aid
stringlengths 9
15
| mid
stringlengths 7
10
| abstract
stringlengths 78
2.56k
| related_work
stringlengths 92
1.77k
| ref_abstract
dict |
|---|---|---|---|---|
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Traditional visual localization methods @cite_68 @cite_7 @cite_57 @cite_29 @cite_70 @cite_32 directly regress the full 6 DoF camera pose of query images, with respect to a known reference. A typical pipeline of such feature-based methods includes acquiring a point-cloud model of the scene through , and computing local feature descriptors like SIFT @cite_43 or LIFT @cite_79 . These descriptors are in turn used to establish 2D-to-3D correspondences with every point of the model, and the query's camera pose can be directly inferred from those matches, using RANSAC @cite_45 @cite_18 combined with an Perspective-n-Point (PP) solver @cite_63 @cite_76 @cite_19 . | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_70",
"@cite_29",
"@cite_32",
"@cite_57",
"@cite_43",
"@cite_79",
"@cite_45",
"@cite_19",
"@cite_63",
"@cite_76",
"@cite_68"
],
"mid": [
"",
"1565312575",
"2728498859",
"",
"2495591376",
"2780675120",
"2151103935",
"2949896259",
"2085261163",
"",
"2091520273",
"",
"1616969904"
],
"abstract": [
"",
"We present a fast, simple location recognition and image localization method that leverages feature correspondence and geometry estimated from large Internet photo collections. Such recovered structure contains a significant amount of useful information about images and image features that is not available when considering images in isolation. For instance, we can predict which views will be the most common, which feature points in a scene are most reliable, and which features in the scene tend to co-occur in the same image. Based on this information, we devise an adaptive, prioritized algorithm for matching a representative set of SIFT features covering a large scene to a query image for efficient localization. Our approach is based on considering features in the scene database, and matching them to query image features, as opposed to more conventional methods that match image features to visual words or database features. We find this approach results in improved performance, due to the richer knowledge of characteristics of the database features compared to query image features. We present experiments on two large city-scale photo collections, showing that our algorithm compares favorably to image retrieval-style approaches to location recognition.",
"Accurate visual localization is a key technology for autonomous navigation. 3D structure-based methods employ 3D models of the scene to estimate the full 6DOF pose of a camera very accurately. However, constructing (and extending) large-scale 3D models is still a significant challenge. In contrast, 2D image retrieval-based methods only require a database of geo-tagged images, which is trivial to construct and to maintain. They are often considered inaccurate since they only approximate the positions of the cameras. Yet, the exact camera pose can theoretically be recovered when enough relevant database images are retrieved. In this paper, we demonstrate experimentally that large-scale 3D models are not strictly necessary for accurate visual localization. We create reference poses for a large and challenging urban dataset. Using these poses, we show that combining image-based methods with local reconstructions results in a pose accuracy similar to the state-of-the-art structure-based methods. Our results suggest that we might want to reconsider the current approach for accurate large-scale localization.",
"",
"We consider the problem of localizing a novel image in a large 3D model, given that the gravitational vector is known. In principle, this is just an instance of camera pose estimation, but the scale of the problem introduces some interesting challenges. Most importantly, it makes the correspondence problem very difficult so there will often be a significant number of outliers to handle. To tackle this problem, we use recent theoretical as well as technical advances. Many modern cameras and phones have gravitational sensors that allow us to reduce the search space. Further, there are new techniques to efficiently and reliably deal with extreme rates of outliers. We extend these methods to camera pose estimation by using accurate approximations and fast polynomial solvers. Experimental results are given demonstrating that it is possible to reliably estimate the camera pose despite cases with more than 99 percent outlier correspondences in city-scale models with several millions of 3D points.",
"Given an image of a street scene in a city, this paper develops a new method that can quickly and precisely pinpoint at which location (as well as viewing direction) the image was taken, against a pre-stored large-scale 3D point-cloud map of the city. We adopt the recently developed 2D-3D direct feature matching framework for this task [23,31,32,42–44]. This is a challenging task especially for large-scale problems. As the map size grows bigger, many 3D points in the wider geographical area can be visually very similar–or even identical–causing severe ambiguities in 2D-3D feature matching. The key is to quickly and unambiguously find the correct matches between a query image and the large 3D map. Existing methods solve this problem mainly via comparing individual features’ visual similarities in a local and per feature manner, thus only local solutions can be found, inadequate for large-scale applications. In this paper, we introduce a global method which harnesses global contextual information exhibited both within the query image and among all the 3D points in the map. This is achieved by a novel global ranking algorithm, applied to a Markov network built upon the 3D map, which takes account of not only visual similarities between individual 2D-3D matches, but also their global compatibilities (as measured by co-visibility) among all matching pairs found in the scene. Tests on standard benchmark datasets show that our method achieved both higher precision and comparable recall, compared with the state-of-the-art.",
"This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene. The features are invariant to image scale and rotation, and are shown to provide robust matching across a substantial range of affine distortion, change in 3D viewpoint, addition of noise, and change in illumination. The features are highly distinctive, in the sense that a single feature can be correctly matched with high probability against a large database of features from many images. This paper also describes an approach to using these features for object recognition. The recognition proceeds by matching individual features to a database of features from known objects using a fast nearest-neighbor algorithm, followed by a Hough transform to identify clusters belonging to a single object, and finally performing verification through least-squares solution for consistent pose parameters. This approach to recognition can robustly identify objects among clutter and occlusion while achieving near real-time performance.",
"We introduce a novel Deep Network architecture that implements the full feature point handling pipeline, that is, detection, orientation estimation, and feature description. While previous works have successfully tackled each one of these problems individually, we show how to learn to do all three in a unified manner while preserving end-to-end differentiability. We then demonstrate that our Deep pipeline outperforms state-of-the-art methods on a number of benchmark datasets, without the need of retraining.",
"A new paradigm, Random Sample Consensus (RANSAC), for fitting a model to experimental data is introduced. RANSAC is capable of interpreting smoothing data containing a significant percentage of gross errors, and is thus ideally suited for applications in automated image analysis where interpretation is based on the data provided by error-prone feature detectors. A major portion of this paper describes the application of RANSAC to the Location Determination Problem (LDP): Given an image depicting a set of landmarks with known locations, determine that point in space from which the image was obtained. In response to a RANSAC requirement, new results are derived on the minimum number of landmarks needed to obtain a solution, and algorithms are presented for computing these minimum-landmark solutions in closed form. These results provide the basis for an automatic system that can solve the LDP under difficult viewing",
"",
"The problem of determining the absolute position and orientation of a camera from a set of 2D-to-3D point correspondences is one of the most important problems in computer vision with a broad range of applications. In this paper we present a new solution to the absolute pose problem for camera with unknown radial distortion and unknown focal length from five 2D-to-3D point correspondences. Our new solver is numerically more stable, more accurate, and significantly faster than the existing state-of-the-art minimal four point absolute pose solvers for this problem. Moreover, our solver results in less solutions and can handle larger radial distortions. The new solver is straightforward and uses only simple concepts from linear algebra. Therefore it is simpler than the state-of-the-art Groebner basis solvers. We compare our new solver with the existing state-of-the-art solvers and show its usefulness on synthetic and real datasets.",
"",
"We address the problem of determining where a photo was taken by estimating a full 6-DOF-plus-intrincs camera pose with respect to a large geo-registered 3D point cloud, bringing together research on image localization, landmark recognition, and 3D pose estimation. Our method scales to datasets with hundreds of thousands of images and tens of millions of 3D points through the use of two new techniques: a co-occurrence prior for RANSAC and bidirectional matching of image features with 3D points. We evaluate our method on several large data sets, and show state-of-the-art results on landmark recognition as well as the ability to locate cameras to within meters, requiring only seconds per query."
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Feature-based methods achieve very competitive results in consistent daytime conditions @cite_5 @cite_2 @cite_56 @cite_38 @cite_11 . However, the pose estimation accuracy for such methods is heavily reliant on the local 2D-3D correspondences. Research in feature-based approaches mostly focuses on improving descriptor matching efficiency @cite_52 @cite_7 @cite_5 @cite_15 @cite_41 @cite_4 , speed @cite_50 @cite_59 and robustness @cite_33 @cite_36 @cite_8 @cite_69 @cite_2 @cite_47 . Despite these efforts, a significant number of erroneous matches appear under strong appearance variations. As a result, dramatic drops in localization accuracy are observed in harder conditions such as night-time @cite_10 . | {
"cite_N": [
"@cite_38",
"@cite_69",
"@cite_47",
"@cite_4",
"@cite_33",
"@cite_7",
"@cite_8",
"@cite_41",
"@cite_36",
"@cite_52",
"@cite_56",
"@cite_50",
"@cite_2",
"@cite_59",
"@cite_5",
"@cite_15",
"@cite_10",
"@cite_11"
],
"mid": [
"",
"",
"",
"",
"2075550860",
"1565312575",
"",
"",
"2474696196",
"2102998745",
"",
"2015529634",
"",
"2017693484",
"2522940611",
"",
"",
""
],
"abstract": [
"",
"",
"",
"",
"We consider the problem of localizing a novel image in a large 3D model. In principle, this is just an instance of camera pose estimation, but the scale introduces some challenging problems. For one, it makes the correspondence problem very difficult and it is likely that there will be a significant rate of outliers to handle. In this paper we use recent theoretical as well as technical advances to tackle these problems. Many modern cameras and phones have gravitational sensors that allow us to reduce the search space. Further, there are new techniques to efficiently and reliably deal with extreme rates of outliers. We extend these methods to camera pose estimation by using accurate approximations and fast polynomial solvers. Experimental results are given demonstrating that it is possible to reliably estimate the camera pose despite more than 99 of outlier correspondences.",
"We present a fast, simple location recognition and image localization method that leverages feature correspondence and geometry estimated from large Internet photo collections. Such recovered structure contains a significant amount of useful information about images and image features that is not available when considering images in isolation. For instance, we can predict which views will be the most common, which feature points in a scene are most reliable, and which features in the scene tend to co-occur in the same image. Based on this information, we devise an adaptive, prioritized algorithm for matching a representative set of SIFT features covering a large scene to a query image for efficient localization. Our approach is based on considering features in the scene database, and matching them to query image features, as opposed to more conventional methods that match image features to visual words or database features. We find this approach results in improved performance, due to the richer knowledge of characteristics of the database features compared to query image features. We present experiments on two large city-scale photo collections, showing that our algorithm compares favorably to image retrieval-style approaches to location recognition.",
"",
"",
"Visual location recognition is the task of determining the place depicted in a query image from a given database of geo-tagged images. Location recognition is often cast as an image retrieval problem and recent research has almost exclusively focused on improving the chance that a relevant database image is ranked high enough after retrieval. The implicit assumption is that the number of inliers found by spatial verification can be used to distinguish between a related and an unrelated database photo with high precision. In this paper, we show that this assumption does not hold for large datasets due to the appearance of geometric bursts, i.e., sets of visual elements appearing in similar geometric configurations in unrelated database photos. We propose algorithms for detecting and handling geometric bursts. Although conceptually simple, using the proposed weighting schemes dramatically improves the recall that can be achieved when high precision is required compared to the standard re-ranking based on the inlier count. Our approach is easy to implement and can easily be integrated into existing location recognition systems.",
"Large scale reconstructions of camera matrices and point clouds have been created using structure from motion from community photo collections. Such a dataset is rich in information; it represents a sampling of the geometry and appearance of the underlying space. In this paper, we encode the visibility information between and among points and cameras as visibility probabilities. The conditional visibility probability of a set of points on a point (or a set of cameras on a camera) can rank points (or cameras) based on their mutual dependence. We combine the conditional probability with a distance measure to prioritize points for fast guided search for the image localization problem. We define dual problem of feature triangulation as finding the 3D coordinates of a given image feature point. We use conditional visibility probability to quickly identify a subset of cameras in which a feature is visible.",
"",
"Image-based localization finds the pose an image was taken in. Our method uses the MPEG Compact Descriptors for Visual Search framework [1] to build a compact 3D model from the 3D reconstruction of a city. Then using an image retrieval the closest match for a query image is found and the pose is estimated. Our method is able to determine an estimate of the pose within a few meters of the actual one.",
"",
"The prevalent approach to image-based localization is matching interest points detected in the query image to a sparse 3D point cloud representing the known world. The obtained correspondences are then used to recover a precise camera pose. The state-of-the-art in this field often ignores the availability of a set of 2D descriptors per 3D point, for example by representing each 3D point by only its centroid. In this paper we demonstrate that these sets contain useful information that can be exploited by formulating matching as a discriminative classification problem. Since memory demands and computational complexity are crucial in such a setup, we base our algorithm on the efficient and effective random fern principle. We propose an extension which projects features to fern-specific embedding spaces, which yields improved matching rates in short runtime. Experiments first show that our novel formulation provides improved matching performance in comparison to the standard nearest neighbor approach and that we outperform related randomization methods in our localization scenario.",
"Accurately determining the position and orientation from which an image was taken, i.e., computing the camera pose, is a fundamental step in many Computer Vision applications. The pose can be recovered from 2D-3D matches between 2D image positions and points in a 3D model of the scene. Recent advances in Structure-from-Motion allow us to reconstruct large scenes and thus create the need for image-based localization methods that efficiently handle large-scale 3D models while still being effective, i.e., while localizing as many images as possible. This paper presents an approach for large scale image-based localization that is both efficient and effective. At the core of our approach is a novel prioritized matching step that enables us to first consider features more likely to yield 2D-to-3D matches and to terminate the correspondence search as soon as enough matches have been found. Matches initially lost due to quantization are efficiently recovered by integrating 3D-to-2D search. We show how visibility information from the reconstruction process can be used to improve the efficiency of our approach. We evaluate the performance of our method through extensive experiments and demonstrate that it offers the best combination of efficiency and effectiveness among current state-of-the-art approaches for localization.",
"",
"",
""
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | One workaround is to avoid using explicit feature matching altogether, and train a CNN to regress dense scene coordinates @cite_6 @cite_34 @cite_72 . A confidence-ranking scheme such as DSAC @cite_13 is then used to discard erroneous pose candidates sampled from the obtained depth map. However, such learning-based variants are hard to initialize @cite_10 @cite_26 and struggle with large environments @cite_10 . Other direct learning-based variants include end-to-end pose regression from images @cite_53 @cite_11 , although these approaches are overall less accurate @cite_10 @cite_12 @cite_54 . | {
"cite_N": [
"@cite_13",
"@cite_26",
"@cite_53",
"@cite_54",
"@cite_6",
"@cite_72",
"@cite_34",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2950681848",
"",
"2951336016",
"",
"2768185495",
"2605111497",
"2803564000",
"",
"",
""
],
"abstract": [
"RANSAC is an important algorithm in robust optimization and a central building block for many computer vision applications. In recent years, traditionally hand-crafted pipelines have been replaced by deep learning pipelines, which can be trained in an end-to-end fashion. However, RANSAC has so far not been used as part of such deep learning pipelines, because its hypothesis selection procedure is non-differentiable. In this work, we present two different ways to overcome this limitation. The most promising approach is inspired by reinforcement learning, namely to replace the deterministic hypothesis selection by a probabilistic selection for which we can derive the expected loss w.r.t. to all learnable parameters. We call this approach DSAC, the differentiable counterpart of RANSAC. We apply DSAC to the problem of camera localization, where deep learning has so far failed to improve on traditional approaches. We demonstrate that by directly minimizing the expected loss of the output camera poses, robustly estimated by RANSAC, we achieve an increase in accuracy. In the future, any deep learning pipeline can use DSAC as a robust optimization component.",
"",
"We present a robust and real-time monocular six degree of freedom relocalization system. Our system trains a convolutional neural network to regress the 6-DOF camera pose from a single RGB image in an end-to-end manner with no need of additional engineering or graph optimisation. The algorithm can operate indoors and outdoors in real time, taking 5ms per frame to compute. It obtains approximately 2m and 6 degree accuracy for large scale outdoor scenes and 0.5m and 10 degree accuracy indoors. This is achieved using an efficient 23 layer deep convnet, demonstrating that convnets can be used to solve complicated out of image plane regression problems. This was made possible by leveraging transfer learning from large scale classification data. We show the convnet localizes from high level features and is robust to difficult lighting, motion blur and different camera intrinsics where point based SIFT registration fails. Furthermore we show how the pose feature that is produced generalizes to other scenes allowing us to regress pose with only a few dozen training examples. PoseNet code, dataset and an online demonstration is available on our project webpage, at this http URL",
"",
"Popular research areas like autonomous driving and augmented reality have renewed the interest in image-based camera localization. In this work, we address the task of predicting the 6D camera pose from a single RGB image in a given 3D environment. With the advent of neural networks, previous works have either learned the entire camera localization process, or multiple components of a camera localization pipeline. Our key contribution is to demonstrate and explain that learning a single component of this pipeline is sufficient. This component is a fully convolutional neural network for densely regressing so-called scene coordinates, defining the correspondence between the input image and the 3D scene space. The neural network is prepended to a new end-to-end trainable pipeline. Our system is efficient, highly accurate, robust in training, and exhibits outstanding generalization capabilities. It exceeds state-of-the-art consistently on indoor and outdoor datasets. Interestingly, our approach surpasses existing techniques even without utilizing a 3D model of the scene during training, since the network is able to discover 3D scene geometry automatically, solely from single-view constraints.",
"Deep learning has shown to be effective for robust and real-time monocular image relocalisation. In particular, PoseNet [22] is a deep convolutional neural network which learns to regress the 6-DOF camera pose from a single image. It learns to localize using high level features and is robust to difficult lighting, motion blur and unknown camera intrinsics, where point based SIFT registration fails. However, it was trained using a naive loss function, with hyper-parameters which require expensive tuning. In this paper, we give the problem a more fundamental theoretical treatment. We explore a number of novel loss functions for learning camera pose which are based on geometry and scene reprojection error. Additionally we show how to automatically learn an optimal weighting to simultaneously regress position and orientation. By leveraging geometry, we demonstrate that our technique significantly improves PoseNets performance across datasets ranging from indoor rooms to a small city.",
"Scene coordinate regression has become an essential part of current camera re-localization methods. Different versions, such as regression forests and deep learning methods, have been successfully applied to estimate the corresponding camera pose given a single input image. In this work, we propose to regress the scene coordinates pixel-wise for a given RGB image by using deep learning. Compared to the recent methods, which usually employ RANSAC to obtain a robust pose estimate from the established point correspondences, we propose to regress confidences of these correspondences, which allows us to immediately discard erroneous predictions and improve the initial pose estimates. Finally, the resulting confidences can be used to score initial pose hypothesis and aid in pose refinement, offering a generalized solution to solve this task.",
"",
"",
""
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Another way of making feature-based methods robust to condition perturbations is to use semantic reasoning. Semantic information can indeed be exploited to enhance either the feature matching stage @cite_42 @cite_3 @cite_46 @cite_26 or the pose estimation stage @cite_0 . While being accurate at small scale, feature-based methods bottleneck is scalability. Both the construction of precise 3D models (and their maintenance), and local feature-matching is challenging and expensive in large-scale scenarios @cite_70 . | {
"cite_N": [
"@cite_26",
"@cite_70",
"@cite_42",
"@cite_3",
"@cite_0",
"@cite_46"
],
"mid": [
"",
"2728498859",
"1245714593",
"2035420386",
"2897085414",
"2495500254"
],
"abstract": [
"",
"Accurate visual localization is a key technology for autonomous navigation. 3D structure-based methods employ 3D models of the scene to estimate the full 6DOF pose of a camera very accurately. However, constructing (and extending) large-scale 3D models is still a significant challenge. In contrast, 2D image retrieval-based methods only require a database of geo-tagged images, which is trivial to construct and to maintain. They are often considered inaccurate since they only approximate the positions of the cameras. Yet, the exact camera pose can theoretically be recovered when enough relevant database images are retrieved. In this paper, we demonstrate experimentally that large-scale 3D models are not strictly necessary for accurate visual localization. We create reference poses for a large and challenging urban dataset. Using these poses, we show that combining image-based methods with local reconstructions results in a pose accuracy similar to the state-of-the-art structure-based methods. Our results suggest that we might want to reconsider the current approach for accurate large-scale localization.",
"Successful large scale object instance retrieval systems are typically based on accurate matching of local descriptors, such as SIFT. However, these local descriptors are often not sufficiently distinctive to prevent false correspondences, as they only consider the gradient appearance of the local patch, without being able to “see the big picture”.",
"Image-to-image feature matching is the single most restrictive time bottleneck in any matching pipeline. We propose two methods for improving the speed and quality by employing semantic scene segmentation. First, we introduce a way of capturing semantic scene context of a key point into a compact description. Second, we propose to learn correct match ability of descriptors from these semantic contexts. Finally, we further reduce the complexity of matching to only a pre-computed set of semantically close key points. All methods can be used independently and in the evaluation we show combinations for maximum speed benefits. Overall, our proposed methods outperform all baselines and provide significant improvements in accuracy and an order of magnitude faster key point matching.",
"Robust and accurate visual localization across large appearance variations due to changes in time of day, seasons, or changes of the environment is a challenging problem which is of importance to application areas such as navigation of autonomous robots. Traditional feature-based methods often struggle in these conditions due to the significant number of erroneous matches between the image and the 3D model. In this paper, we present a method for scoring the individual correspondences by exploiting semantic information about the query image and the scene. In this way, erroneous correspondences tend to get a low semantic consistency score, whereas correct correspondences tend to get a high score. By incorporating this information in a standard localization pipeline, we show that the localization performance can be significantly improved compared to the state-of-the-art, as evaluated on two challenging long-term localization benchmarks.",
"The problem of localization and geo-location estimation of an image has a long-standing history both in robotics and computer vision. With the advent of availability of large amounts of geo-referenced image data, several image retrieval approaches have been deployed to tackle this problem. In this work, we will show how the capability of semantic labeling of both query views and the reference dataset by means of semantic segmentation can aid (1) the problem of retrieval of views similar and possibly overlapping with the query and (2) guide the recognition and discovery of commonly occurring scene layouts in the reference dataset. We will demonstrate the effectiveness of these semantic representations on examples of localization, semantic concept discovery, and intersection recognition in the images of urban scenes."
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Retrieval-based or image-based localization methods trade-off accuracy for scalability, by modeling the scene as an image database, and visual localization as an image retrieval problem. Visual queries are matched with a pose-annotated images, using compact image-level representations. The query's camera pose can then be simply inferred from the top-ranked images @cite_21 @cite_55 @cite_39 @cite_70 , and the need for ground-truth 3D geometry is alleviated. | {
"cite_N": [
"@cite_70",
"@cite_55",
"@cite_21",
"@cite_39"
],
"mid": [
"2728498859",
"1883248133",
"",
"2043732461"
],
"abstract": [
"Accurate visual localization is a key technology for autonomous navigation. 3D structure-based methods employ 3D models of the scene to estimate the full 6DOF pose of a camera very accurately. However, constructing (and extending) large-scale 3D models is still a significant challenge. In contrast, 2D image retrieval-based methods only require a database of geo-tagged images, which is trivial to construct and to maintain. They are often considered inaccurate since they only approximate the positions of the cameras. Yet, the exact camera pose can theoretically be recovered when enough relevant database images are retrieved. In this paper, we demonstrate experimentally that large-scale 3D models are not strictly necessary for accurate visual localization. We create reference poses for a large and challenging urban dataset. Using these poses, we show that combining image-based methods with local reconstructions results in a pose accuracy similar to the state-of-the-art structure-based methods. Our results suggest that we might want to reconsider the current approach for accurate large-scale localization.",
"Finding an image's exact GPS location is a challenging computer vision problem that has many real-world applications. In this paper, we address the problem of finding the GPS location of images with an accuracy which is comparable to hand-held GPS devices. We leverage a structured data set of about 100,000 images build from Google Maps Street View as the reference images. We propose a localization method in which the SIFT descriptors of the detected SIFT interest points in the reference images are indexed using a tree. In order to localize a query image, the tree is queried using the detected SIFT descriptors in the query image. A novel GPS-tag-based pruning method removes the less reliable descriptors. Then, a smoothing step with an associated voting scheme is utilized; this allows each query descriptor to vote for the location its nearest neighbor belongs to, in order to accurately localize the query image. A parameter called Confidence of Localization which is based on the Kurtosis of the distribution of votes is defined to determine how reliable the localization of a particular image is. In addition, we propose a novel approach to localize groups of images accurately in a hierarchical manner. First, each image is localized individually; then, the rest of the images in the group are matched against images in the neighboring area of the found first match. The final location is determined based on the Confidence of Localization parameter. The proposed image group localization method can deal with very unclear queries which are not capable of being geolocated individually.",
"",
"In this paper we present a prototype system for image based localization in urban environments. Given a database of views of city street scenes tagged by GPS locations, the system computes the GPS location of a novel query view. We first use a wide-baseline matching technique based on SIFT features to select the closest views in the database. Often due to a large change of viewpoint and presence of repetitive structures, a large percentage of matches (> 50 ) are not correct correspondences. The subsequent motion estimation between the query view and the reference view, is then handled by a novel and efficient robust estimation technique capable of dealing with large percentage of outliers. This stage is also accompanied by a model selection step among the fundamental matrix and the homography. Once the motion between the closest reference views is estimated, the location of the query view is then obtained by triangulation of translation directions. Approximate solutions for cases when triangulation cannot be obtained reliably are also described. The presented system is tested on the dataset used in ICCV 2005 Computer Vision Contest and is shown to have higher accuracy than previous reported results."
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Robust global descriptors can be obtained by aggregation of local features in the image. VLAD @cite_71 is a popular descriptor, computed by summing and concatenating many descriptors for affine-invariant regions. DenseVLAD @cite_49 modified the VLAD architecture by densely sampling RootSIFT @cite_77 descriptors in the image. Recent learning-based variants cast the task of image retrieval as a metric learning problem. NetVLAD @cite_62 defines a differentiable VLAD layer as the final activation of a siamese network. While never explicitly applied to visual localization, activations such as max-pooling (MAC) @cite_31 @cite_61 , sum-pooling (SPoC) @cite_25 , weighted sum-pooling (CroW) @cite_20 or regional max-pooling (R-MAC) @cite_74 coupled with siamese or triplet architectures, have shown to outperform VLAD-based methods in image-retrieval tasks @cite_23 . The Generalized-Mean-Pooling (GeM) @cite_37 layer proposes a hybrid activation, which combines the benefits of average and max-pooling, and achieves state-the-art results in outdoor visual search @cite_23 . | {
"cite_N": [
"@cite_61",
"@cite_37",
"@cite_62",
"@cite_77",
"@cite_49",
"@cite_71",
"@cite_74",
"@cite_23",
"@cite_31",
"@cite_25",
"@cite_20"
],
"mid": [
"2294444065",
"",
"2951019013",
"1979931042",
"2589016937",
"2103924867",
"2544587078",
"2963154697",
"2174726731",
"2204975001",
""
],
"abstract": [
"This paper provides an extensive study on the availability of image representations based on convolutional networks (ConvNets) for the task of visual instance retrieval. Besides the choice of convolutional layers, we present an efficient pipeline exploiting multi-scale schemes to extract local features, in particular, by taking geometric invariance into explicit account, i.e. positions, scales and spatial consistency. In our experiments using five standard image retrieval datasets, we demonstrate that generic ConvNet image representations can outperform other state-of-the-art methods if they are extracted appropriately.",
"",
"We tackle the problem of large scale visual place recognition, where the task is to quickly and accurately recognize the location of a given query photograph. We present the following three principal contributions. First, we develop a convolutional neural network (CNN) architecture that is trainable in an end-to-end manner directly for the place recognition task. The main component of this architecture, NetVLAD, is a new generalized VLAD layer, inspired by the \"Vector of Locally Aggregated Descriptors\" image representation commonly used in image retrieval. The layer is readily pluggable into any CNN architecture and amenable to training via backpropagation. Second, we develop a training procedure, based on a new weakly supervised ranking loss, to learn parameters of the architecture in an end-to-end manner from images depicting the same places over time downloaded from Google Street View Time Machine. Finally, we show that the proposed architecture significantly outperforms non-learnt image representations and off-the-shelf CNN descriptors on two challenging place recognition benchmarks, and improves over current state-of-the-art compact image representations on standard image retrieval benchmarks.",
"The objective of this work is object retrieval in large scale image datasets, where the object is specified by an image query and retrieval should be immediate at run time in the manner of Video Google [28]. We make the following three contributions: (i) a new method to compare SIFT descriptors (RootSIFT) which yields superior performance without increasing processing or storage requirements; (ii) a novel method for query expansion where a richer model for the query is learnt discriminatively in a form suited to immediate retrieval through efficient use of the inverted index; (iii) an improvement of the image augmentation method proposed by Turcot and Lowe [29], where only the augmenting features which are spatially consistent with the augmented image are kept. We evaluate these three methods over a number of standard benchmark datasets (Oxford Buildings 5k and 105k, and Paris 6k) and demonstrate substantial improvements in retrieval performance whilst maintaining immediate retrieval speeds. Combining these complementary methods achieves a new state-of-the-art performance on these datasets.",
"We address the problem of large-scale visual place recognition for situations where the scene undergoes a major change in appearance, for example, due to illumination (day night), change of seasons, aging, or structural modifications over time such as buildings being built or destroyed. Such situations represent a major challenge for current large-scale place recognition methods. This work has the following three principal contributions. First, we demonstrate that matching across large changes in the scene appearance becomes much easier when both the query image and the database image depict the scene from approximately the same viewpoint. Second, based on this observation, we develop a new place recognition approach that combines (i) an efficient synthesis of novel views with (ii) a compact indexable image representation. Third, we introduce a new challenging dataset of 1,125 camera-phone query images of Tokyo that contain major changes in illumination (day, sunset, night) as well as structural changes in the scene. We demonstrate that the proposed approach significantly outperforms other large-scale place recognition techniques on this challenging data.",
"The objective of this paper is large scale object instance retrieval, given a query image. A starting point of such systems is feature detection and description, for example using SIFT. The focus of this paper, however, is towards very large scale retrieval where, due to storage requirements, very compact image descriptors are required and no information about the original SIFT descriptors can be accessed directly at run time. We start from VLAD, the state-of-the art compact descriptor introduced by for this purpose, and make three novel contributions: first, we show that a simple change to the normalization method significantly improves retrieval performance, second, we show that vocabulary adaptation can substantially alleviate problems caused when images are added to the dataset after initial vocabulary learning. These two methods set a new state-of-the-art over all benchmarks investigated here for both mid-dimensional (20k-D to 30k-D) and small (128-D) descriptors. Our third contribution is a multiple spatial VLAD representation, MultiVLAD, that allows the retrieval and localization of objects that only extend over a small part of an image (again without requiring use of the original image SIFT descriptors).",
"While deep learning has become a key ingredient in the top performing methods for many computer vision tasks, it has failed so far to bring similar improvements to instance-level image retrieval. In this article, we argue that reasons for the underwhelming results of deep methods on image retrieval are threefold: (1) noisy training data, (2) inappropriate deep architecture, and (3) suboptimal training procedure. We address all three issues. First, we leverage a large-scale but noisy landmark dataset and develop an automatic cleaning method that produces a suitable training set for deep retrieval. Second, we build on the recent R-MAC descriptor, show that it can be interpreted as a deep and differentiable architecture, and present improvements to enhance it. Last, we train this network with a siamese architecture that combines three streams with a triplet loss. At the end of the training process, the proposed architecture produces a global image representation in a single forward pass that is well suited for image retrieval. Extensive experiments show that our approach significantly outperforms previous retrieval approaches, including state-of-the-art methods based on costly local descriptor indexing and spatial verification. On Oxford 5k, Paris 6k and Holidays, we respectively report 94.7, 96.6, and 94.8 mean average precision. Our representations can also be heavily compressed using product quantization with little loss in accuracy.",
"In this paper we address issues with image retrieval benchmarking on standard and popular Oxford 5k and Paris 6k datasets. In particular, annotation errors, the size of the dataset, and the level of challenge are addressed: new annotation for both datasets is created with an extra attention to the reliability of the ground truth. Three new protocols of varying difficulty are introduced. The protocols allow fair comparison between different methods, including those using a dataset pre-processing stage. For each dataset, 15 new challenging queries are introduced. Finally, a new set of 1M hard, semi-automatically cleaned distractors is selected. An extensive1 comparison of the state-of-the-art methods is performed on the new benchmark. Different types of methods are evaluated, ranging from local-feature-based to modern CNN based methods. The best results are achieved by taking the best of the two worlds. Most importantly, image retrieval appears far from being solved.",
"Recently, image representation built upon Convolutional Neural Network (CNN) has been shown to provide effective descriptors for image search, outperforming pre-CNN features as short-vector representations. Yet such models are not compatible with geometry-aware re-ranking methods and still outperformed, on some particular object retrieval benchmarks, by traditional image search systems relying on precise descriptor matching, geometric re-ranking, or query expansion. This work revisits both retrieval stages, namely initial search and re-ranking, by employing the same primitive information derived from the CNN. We build compact feature vectors that encode several image regions without the need to feed multiple inputs to the network. Furthermore, we extend integral images to handle max-pooling on convolutional layer activations, allowing us to efficiently localize matching objects. The resulting bounding box is finally used for image re-ranking. As a result, this paper significantly improves existing CNN-based recognition pipeline: We report for the first time results competing with traditional methods on the challenging Oxford5k and Paris6k datasets.",
"Several recent works have shown that image descriptors produced by deep convolutional neural networks provide state-of-the-art performance for image classification and retrieval problems. It also has been shown that the activations from the convolutional layers can be interpreted as local features describing particular image regions. These local features can be aggregated using aggregating methods developed for local features (e.g. Fisher vectors), thus providing new powerful global descriptor. In this paper we investigate possible ways to aggregate local deep features to produce compact descriptors for image retrieval. First, we show that deep features and traditional hand-engineered features have quite different distributions of pairwise similarities, hence existing aggregation methods have to be carefully re-evaluated. Such re-evaluation reveals that in contrast to shallow features, the simple aggregation method based on sum pooling provides the best performance for deep convolutional features. This method is efficient, has few parameters, and bears little risk of overfitting when e.g. learning the PCA matrix. In addition, we suggest a simple yet efficient query expansion scheme suitable for the proposed aggregation method. Overall, the new compact global descriptor improves the state-of-the-art on four common benchmarks considerably.",
""
]
} |
1812.03707 | 2942115372 | Outdoor visual localization is a crucial component to many computer vision systems. We propose an approach to localization from images that is designed to explicitly handle the strong variations in appearance happening between daytime and nighttime. As revealed by recent long-term localization benchmarks, both traditional feature-based and retrieval-based approaches still struggle to handle such changes. Our novel localization method combines a state-of-the-art image retrieval architecture with condition-specific sub-networks allowing the computation of global image descriptors that are explicitly dependent of the capturing conditions. We show that our approach improves localization by a factor of almost 300 compared to the popular VLAD-based methods on nighttime localization. | Compared to 3D feature-based methods, retrieval-based methods offer two key advantages. First, the extension of such methods to city-scale scenarios is trivial @cite_70 . Besides, in a very large database, unsupervised descriptor compression like PCA @cite_16 or Product Quantization (PQ) @cite_51 enables efficient approximate nearest-neighbour search with little loss in performance @cite_74 . Secondly, evaluations in day-to-night @cite_62 @cite_9 or long-term @cite_10 @cite_0 visual localization reveal that retrieval-based methods outperform other feature-based methods like Active Search @cite_5 at a coarse precision level. Image pre-processing @cite_14 or translation @cite_24 can be employed to bring all images to a visually similar and condition-invariant representation. However, image translation methods are complex to train, require a full retraining for every condition and add to the overall computational cost of retrieval methods. Our method outperforms both retrieval and feature-based techniques without the need for such pre-processing. | {
"cite_N": [
"@cite_62",
"@cite_14",
"@cite_70",
"@cite_10",
"@cite_9",
"@cite_0",
"@cite_24",
"@cite_74",
"@cite_5",
"@cite_16",
"@cite_51"
],
"mid": [
"2951019013",
"1998728242",
"2728498859",
"",
"",
"2897085414",
"2969121634",
"2544587078",
"2522940611",
"1679894842",
"2124509324"
],
"abstract": [
"We tackle the problem of large scale visual place recognition, where the task is to quickly and accurately recognize the location of a given query photograph. We present the following three principal contributions. First, we develop a convolutional neural network (CNN) architecture that is trainable in an end-to-end manner directly for the place recognition task. The main component of this architecture, NetVLAD, is a new generalized VLAD layer, inspired by the \"Vector of Locally Aggregated Descriptors\" image representation commonly used in image retrieval. The layer is readily pluggable into any CNN architecture and amenable to training via backpropagation. Second, we develop a training procedure, based on a new weakly supervised ranking loss, to learn parameters of the architecture in an end-to-end manner from images depicting the same places over time downloaded from Google Street View Time Machine. Finally, we show that the proposed architecture significantly outperforms non-learnt image representations and off-the-shelf CNN descriptors on two challenging place recognition benchmarks, and improves over current state-of-the-art compact image representations on standard image retrieval benchmarks.",
"This paper is about extending the reach and endurance of outdoor localisation using stereo vision. At the heart of the localisation is the fundamental task of discovering feature correspondences between recorded and live images. One aspect of this problem involves deciding where to look for correspondences in an image and the second is deciding what to look for. This latter point, which is the main focus of our paper, requires understanding how and why the appearance of visual features can change over time. In particular, such knowledge allows us to better deal with abrupt and challenging changes in lighting. We show how by instantiating a parallel image processing stream which operates on illumination-invariant images, we can substantially improve the performance of an outdoor visual navigation system. We will demonstrate, explain and analyse the effect of the RGB to illumination-invariant transformation and suggest that for little cost it becomes a viable tool for those concerned with having robots operate for long periods outdoors.",
"Accurate visual localization is a key technology for autonomous navigation. 3D structure-based methods employ 3D models of the scene to estimate the full 6DOF pose of a camera very accurately. However, constructing (and extending) large-scale 3D models is still a significant challenge. In contrast, 2D image retrieval-based methods only require a database of geo-tagged images, which is trivial to construct and to maintain. They are often considered inaccurate since they only approximate the positions of the cameras. Yet, the exact camera pose can theoretically be recovered when enough relevant database images are retrieved. In this paper, we demonstrate experimentally that large-scale 3D models are not strictly necessary for accurate visual localization. We create reference poses for a large and challenging urban dataset. Using these poses, we show that combining image-based methods with local reconstructions results in a pose accuracy similar to the state-of-the-art structure-based methods. Our results suggest that we might want to reconsider the current approach for accurate large-scale localization.",
"",
"",
"Robust and accurate visual localization across large appearance variations due to changes in time of day, seasons, or changes of the environment is a challenging problem which is of importance to application areas such as navigation of autonomous robots. Traditional feature-based methods often struggle in these conditions due to the significant number of erroneous matches between the image and the 3D model. In this paper, we present a method for scoring the individual correspondences by exploiting semantic information about the query image and the scene. In this way, erroneous correspondences tend to get a low semantic consistency score, whereas correct correspondences tend to get a high score. By incorporating this information in a standard localization pipeline, we show that the localization performance can be significantly improved compared to the state-of-the-art, as evaluated on two challenging long-term localization benchmarks.",
"Visual localization is a key step in many robotics pipelines, allowing the robot to (approximately) determine its position and orientation in the world. An efficient and scalable approach to visual localization is to use image retrieval techniques. These approaches identify the image most similar to a query photo in a database of geo-tagged images and approximate the query’s pose via the pose of the retrieved database image. However, image retrieval across drastically different illumination conditions, e.g. day and night, is still a problem with unsatisfactory results, even in this age of powerful neural models. This is due to a lack of a suitably diverse dataset with true correspondences to perform end-to-end learning. A recent class of neural models allows for realistic translation of images among visual domains with relatively little training data and, most importantly, without ground-truth pairings.In this paper, we explore the task of accurately localizing images captured from two traversals of the same area in both day and night. We propose ToDayGAN – a modified image-translation model to alter nighttime driving images to a more useful daytime representation. We then compare the daytime and translated night images to obtain a pose estimate for the night image using the known 6-DOF position of the closest day image. Our approach improves localization performance by over 250 compared the current state-of-the-art, in the context of standard metrics in multiple categories.",
"While deep learning has become a key ingredient in the top performing methods for many computer vision tasks, it has failed so far to bring similar improvements to instance-level image retrieval. In this article, we argue that reasons for the underwhelming results of deep methods on image retrieval are threefold: (1) noisy training data, (2) inappropriate deep architecture, and (3) suboptimal training procedure. We address all three issues. First, we leverage a large-scale but noisy landmark dataset and develop an automatic cleaning method that produces a suitable training set for deep retrieval. Second, we build on the recent R-MAC descriptor, show that it can be interpreted as a deep and differentiable architecture, and present improvements to enhance it. Last, we train this network with a siamese architecture that combines three streams with a triplet loss. At the end of the training process, the proposed architecture produces a global image representation in a single forward pass that is well suited for image retrieval. Extensive experiments show that our approach significantly outperforms previous retrieval approaches, including state-of-the-art methods based on costly local descriptor indexing and spatial verification. On Oxford 5k, Paris 6k and Holidays, we respectively report 94.7, 96.6, and 94.8 mean average precision. Our representations can also be heavily compressed using product quantization with little loss in accuracy.",
"Accurately determining the position and orientation from which an image was taken, i.e., computing the camera pose, is a fundamental step in many Computer Vision applications. The pose can be recovered from 2D-3D matches between 2D image positions and points in a 3D model of the scene. Recent advances in Structure-from-Motion allow us to reconstruct large scenes and thus create the need for image-based localization methods that efficiently handle large-scale 3D models while still being effective, i.e., while localizing as many images as possible. This paper presents an approach for large scale image-based localization that is both efficient and effective. At the core of our approach is a novel prioritized matching step that enables us to first consider features more likely to yield 2D-to-3D matches and to terminate the correspondence search as soon as enough matches have been found. Matches initially lost due to quantization are efficiently recovered by integrating 3D-to-2D search. We show how visibility information from the reconstruction process can be used to improve the efficiency of our approach. We evaluate the performance of our method through extensive experiments and demonstrate that it offers the best combination of efficiency and effectiveness among current state-of-the-art approaches for localization.",
"The paper addresses large scale image retrieval with short vector representations. We study dimensionality reduction by Principal Component Analysis (PCA) and propose improvements to its different phases. We show and explicitly exploit relations between i) mean subtraction and the negative evidence, i.e., a visual word that is mutually missing in two descriptions being compared, and ii) the axis de-correlation and the co-occurrences phenomenon. Finally, we propose an effective way to alleviate the quantization artifacts through a joint dimensionality reduction of multiple vocabularies. The proposed techniques are simple, yet significantly and consistently improve over the state of the art on compact image representations. Complementary experiments in image classification show that the methods are generally applicable.",
"This paper introduces a product quantization-based approach for approximate nearest neighbor search. The idea is to decompose the space into a Cartesian product of low-dimensional subspaces and to quantize each subspace separately. A vector is represented by a short code composed of its subspace quantization indices. The euclidean distance between two vectors can be efficiently estimated from their codes. An asymmetric version increases precision, as it computes the approximate distance between a vector and a code. Experimental results show that our approach searches for nearest neighbors efficiently, in particular in combination with an inverted file system. Results for SIFT and GIST image descriptors show excellent search accuracy, outperforming three state-of-the-art approaches. The scalability of our approach is validated on a data set of two billion vectors."
]
} |
1812.03405 | 2905418437 | Classifiers fail to classify correctly input images that have been purposefully and imperceptibly perturbed to cause misclassification. This susceptability has been shown to be consistent across classifiers, regardless of their type, architecture or parameters. Common defenses against adversarial attacks modify the classifer boundary by training on additional adversarial examples created in various ways. In this paper, we introduce AutoGAN, which counters adversarial attacks by enhancing the lower-dimensional manifold defined by the training data and by projecting perturbed data points onto it. AutoGAN mitigates the need for knowing the attack type and magnitude as well as the need for having adversarial samples of the attack. Our approach uses a Generative Adversarial Network (GAN) with an autoencoder generator and a discriminator that also serves as a classifier. We test AutoGAN against adversarial samples generated with state-of-the-art Fast Gradient Sign Method (FGSM) as well as samples generated with random Gaussian noise, both using the MNIST dataset. For different magnitudes of perturbation in training and testing, AutoGAN can surpass the accuracy of FGSM method by up to 25 points on samples perturbed using FGSM. Without an augmented training dataset, AutoGAN achieves an accuracy of 89 compared to 1 achieved by FGSM method on FGSM testing adversarial samples. | Vulnerability to adversarial attacks has been attributed to several factors. Among these factors are the extreme nonlinearity of deep networks @cite_5 ; and linear behavior in high-dimensional spaces @cite_28 , leading to the FGSM method of adversarial direction. Szegedy @cite_5 , Song @cite_25 and Tram e r @cite_35 have attributed the susceptibility to adversarial attacks to low probability pockets in the manifold. | {
"cite_N": [
"@cite_28",
"@cite_5",
"@cite_25",
"@cite_35"
],
"mid": [
"1945616565",
"1673923490",
"",
"2612637113"
],
"abstract": [
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.",
"Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.",
"",
"Adversarial examples are maliciously perturbed inputs designed to mislead machine learning (ML) models at test-time. They often transfer: the same adversarial example fools more than one model. In this work, we propose novel methods for estimating the previously unknown dimensionality of the space of adversarial inputs. We find that adversarial examples span a contiguous subspace of large ( 25) dimensionality. Adversarial subspaces with higher dimensionality are more likely to intersect. We find that for two different models, a significant fraction of their subspaces is shared, thus enabling transferability. In the first quantitative analysis of the similarity of different models' decision boundaries, we show that these boundaries are actually close in arbitrary directions, whether adversarial or benign. We conclude by formally studying the limits of transferability. We derive (1) sufficient conditions on the data distribution that imply transferability for simple model classes and (2) examples of scenarios in which transfer does not occur. These findings indicate that it may be possible to design defenses against transfer-based attacks, even for models that are vulnerable to direct attacks."
]
} |
1812.03405 | 2905418437 | Classifiers fail to classify correctly input images that have been purposefully and imperceptibly perturbed to cause misclassification. This susceptability has been shown to be consistent across classifiers, regardless of their type, architecture or parameters. Common defenses against adversarial attacks modify the classifer boundary by training on additional adversarial examples created in various ways. In this paper, we introduce AutoGAN, which counters adversarial attacks by enhancing the lower-dimensional manifold defined by the training data and by projecting perturbed data points onto it. AutoGAN mitigates the need for knowing the attack type and magnitude as well as the need for having adversarial samples of the attack. Our approach uses a Generative Adversarial Network (GAN) with an autoencoder generator and a discriminator that also serves as a classifier. We test AutoGAN against adversarial samples generated with state-of-the-art Fast Gradient Sign Method (FGSM) as well as samples generated with random Gaussian noise, both using the MNIST dataset. For different magnitudes of perturbation in training and testing, AutoGAN can surpass the accuracy of FGSM method by up to 25 points on samples perturbed using FGSM. Without an augmented training dataset, AutoGAN achieves an accuracy of 89 compared to 1 achieved by FGSM method on FGSM testing adversarial samples. | Many efforts have explored generation of adversarial samples. The FGSM method @cite_28 identifies directions in the input space that reduce the most the probability of correct classification as adversarial directions. Given a target model @math , an input image @math , its label @math , the loss function @math , FGSM creates adversarial samples by adding perturbation to input images: | {
"cite_N": [
"@cite_28"
],
"mid": [
"1945616565"
],
"abstract": [
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset."
]
} |
1812.03405 | 2905418437 | Classifiers fail to classify correctly input images that have been purposefully and imperceptibly perturbed to cause misclassification. This susceptability has been shown to be consistent across classifiers, regardless of their type, architecture or parameters. Common defenses against adversarial attacks modify the classifer boundary by training on additional adversarial examples created in various ways. In this paper, we introduce AutoGAN, which counters adversarial attacks by enhancing the lower-dimensional manifold defined by the training data and by projecting perturbed data points onto it. AutoGAN mitigates the need for knowing the attack type and magnitude as well as the need for having adversarial samples of the attack. Our approach uses a Generative Adversarial Network (GAN) with an autoencoder generator and a discriminator that also serves as a classifier. We test AutoGAN against adversarial samples generated with state-of-the-art Fast Gradient Sign Method (FGSM) as well as samples generated with random Gaussian noise, both using the MNIST dataset. For different magnitudes of perturbation in training and testing, AutoGAN can surpass the accuracy of FGSM method by up to 25 points on samples perturbed using FGSM. Without an augmented training dataset, AutoGAN achieves an accuracy of 89 compared to 1 achieved by FGSM method on FGSM testing adversarial samples. | In addition to FGSM adversarial directions, random adversarial directions can also be identified, for example with random Gaussian noise added to pixels @cite_5 . Sabour @cite_17 generate adversarial samples by manipulating the deep representation of images to make them look like other specific images to neural networks. Ensemble Adversarial Training @cite_21 augment the training data with perturbations from other models. Ensemble Adversarial Training decouples the generation of adversarial samples and the model being trained to counter adversarial attacks because adversarial perturbations can be transferred from one NN to another due to the transferability of adversarial attacks. Other efforts use GANs @cite_15 @cite_37 to generate adversarial training examples @cite_2 @cite_10 @cite_19 @cite_24 . Xiao @cite_2 generate adversarial examples with adversarial networks, where GANs generate the perturbations. Kos @cite_31 utilize variational autoencoder (VAE) and VAE-GAN for generation of adversarial training examples. Baluja @cite_18 uses two approaches for generation of adversarial samples: perturbation in residual blocks in residual networks; adversarial autoencoding using regularizers. | {
"cite_N": [
"@cite_37",
"@cite_31",
"@cite_18",
"@cite_21",
"@cite_24",
"@cite_19",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_10",
"@cite_17"
],
"mid": [
"",
"2951578466",
"2604147826",
"2620038827",
"",
"",
"2783555701",
"1673923490",
"",
"",
"2179402106"
],
"abstract": [
"",
"We explore methods of producing adversarial examples on deep generative models such as the variational autoencoder (VAE) and the VAE-GAN. Deep learning architectures are known to be vulnerable to adversarial examples, but previous work has focused on the application of adversarial examples to classification tasks. Deep generative models have recently become popular due to their ability to model input data distributions and generate realistic examples from those distributions. We present three classes of attacks on the VAE and VAE-GAN architectures and demonstrate them against networks trained on MNIST, SVHN and CelebA. Our first attack leverages classification-based adversaries by attaching a classifier to the trained encoder of the target generative model, which can then be used to indirectly manipulate the latent representation. Our second attack directly uses the VAE loss function to generate a target reconstruction image from the adversarial example. Our third attack moves beyond relying on classification or the standard loss for the gradient and directly optimizes against differences in source and target latent representations. We also motivate why an attacker might be interested in deploying such techniques against a target generative network.",
"Multiple different approaches of generating adversarial examples have been proposed to attack deep neural networks. These approaches involve either directly computing gradients with respect to the image pixels, or directly solving an optimization on the image pixels. In this work, we present a fundamentally new method for generating adversarial examples that is fast to execute and provides exceptional diversity of output. We efficiently train feed-forward neural networks in a self-supervised manner to generate adversarial examples against a target network or set of networks. We call such a network an Adversarial Transformation Network (ATN). ATNs are trained to generate adversarial examples that minimally modify the classifier's outputs given the original input, while constraining the new classification to match an adversarial target class. We present methods to train ATNs and analyze their effectiveness targeting a variety of MNIST classifiers as well as the latest state-of-the-art ImageNet classifier Inception ResNet v2.",
"Adversarial examples are perturbed inputs designed to fool machine learning models. Adversarial training injects such examples into training data to increase robustness. To scale this technique to large datasets, perturbations are crafted using fast single-step methods that maximize a linear approximation of the model's loss. We show that this form of adversarial training converges to a degenerate global minimum, wherein small curvature artifacts near the data points obfuscate a linear approximation of the loss. The model thus learns to generate weak perturbations, rather than defend against strong ones. As a result, we find that adversarial training remains vulnerable to black-box attacks, where we transfer perturbations computed on undefended models, as well as to a powerful novel single-step attack that escapes the non-smooth vicinity of the input data via a small random step. We further introduce Ensemble Adversarial Training, a technique that augments training data with perturbations transferred from other models. On ImageNet, Ensemble Adversarial Training yields models with strong robustness to black-box attacks. In particular, our most robust model won the first round of the NIPS 2017 competition on Defenses against Adversarial Attacks.",
"",
"",
"Deep neural networks (DNNs) have been found to be vulnerable to adversarial examples resulting from adding small-magnitude perturbations to inputs. Such adversarial examples can mislead DNNs to produce adversary-selected results. Different attack strategies have been proposed to generate adversarial examples, but how to produce them with high perceptual quality and more efficiently requires more research efforts. In this paper, we propose AdvGAN to generate adversarial examples with generative adversarial networks (GANs), which can learn and approximate the distribution of original instances. For AdvGAN, once the generator is trained, it can generate adversarial perturbations efficiently for any instance, so as to potentially accelerate adversarial training as defenses. We apply AdvGAN in both semi-whitebox and black-box attack settings. In semi-whitebox attacks, there is no need to access the original target model after the generator is trained, in contrast to traditional white-box attacks. In black-box attacks, we dynamically train a distilled model for the black-box model and optimize the generator accordingly. Adversarial examples generated by AdvGAN on different target models have high attack success rate under state-of-the-art defenses compared to other attacks. Our attack has placed the first with 92.76 accuracy on a public MNIST black-box attack challenge.",
"Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.",
"",
"",
"We show that the representation of an image in a deep neural network (DNN) can be manipulated to mimic those of other natural images, with only minor, imperceptible perturbations to the original image. Previous methods for generating adversarial images focused on image perturbations designed to produce erroneous class labels, while we concentrate on the internal layers of DNN representations. In this way our new class of adversarial images differs qualitatively from others. While the adversary is perceptually similar to one image, its internal representation appears remarkably similar to a different image, one from a different class, bearing little if any apparent similarity to the input; they appear generic and consistent with the space of natural images. This phenomenon raises questions about DNN representations, as well as the properties of natural images themselves."
]
} |
1812.03405 | 2905418437 | Classifiers fail to classify correctly input images that have been purposefully and imperceptibly perturbed to cause misclassification. This susceptability has been shown to be consistent across classifiers, regardless of their type, architecture or parameters. Common defenses against adversarial attacks modify the classifer boundary by training on additional adversarial examples created in various ways. In this paper, we introduce AutoGAN, which counters adversarial attacks by enhancing the lower-dimensional manifold defined by the training data and by projecting perturbed data points onto it. AutoGAN mitigates the need for knowing the attack type and magnitude as well as the need for having adversarial samples of the attack. Our approach uses a Generative Adversarial Network (GAN) with an autoencoder generator and a discriminator that also serves as a classifier. We test AutoGAN against adversarial samples generated with state-of-the-art Fast Gradient Sign Method (FGSM) as well as samples generated with random Gaussian noise, both using the MNIST dataset. For different magnitudes of perturbation in training and testing, AutoGAN can surpass the accuracy of FGSM method by up to 25 points on samples perturbed using FGSM. Without an augmented training dataset, AutoGAN achieves an accuracy of 89 compared to 1 achieved by FGSM method on FGSM testing adversarial samples. | We evaluate AutoGAN against adversarial samples in adversarial directions generated with the FGSM method @cite_28 . To further assess that the scope of AutoGAN is not only limited to adversarial directions, we evaluate AutoGAN against adversarial attacks in random directions created by adding random Gaussian noise to each pixel. | {
"cite_N": [
"@cite_28"
],
"mid": [
"1945616565"
],
"abstract": [
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset."
]
} |
1812.03659 | 2905351737 | High-dimensional auto-regressive models provide a natural way to model influence between @math actors given multi-variate time series data for @math time intervals. While there has been considerable work on network estimation, there is limited work in the context of inference and hypothesis testing. In particular, prior work on hypothesis testing in time series has been restricted to linear Gaussian auto-regressive models. From a practical perspective, it is important to determine suitable statistical tests for connections between actors that go beyond the Gaussian assumption. In the context of time series models, confidence intervals present additional estimators since most estimators such as the Lasso and Dantzig selectors are biased which has led to estimators. In this paper we address these challenges and provide convergence in distribution results and confidence intervals for the multi-variate AR(p) model with sub-Gaussian noise, a generalization of Gaussian noise that broadens applicability and presents numerous technical challenges. The main technical challenge lies in the fact that unlike Gaussian random vectors, for sub-Gaussian vectors zero correlation does not imply independence. The proof relies on using an intricate truncation argument to develop novel concentration bounds for quadratic forms of dependent sub-Gaussian random variables. Our convergence in distribution results hold provided @math , where @math and @math refer to sparsity parameters which matches existed results for hypothesis testing with i.i.d. samples. We validate our theoretical results with simulation results for both block-structured and chain-structured networks. | In the literature on inference for high-dimensional VAR models, most work focuses on the estimation problem. Song and Bickel ( @cite_6 ) investigate penalized least squares algorithms for different penalties, with some externally imposed assumptions on the temporal dependence. Theoretical guarantees on Dantzig type and Lasso type estimators are studied in @cite_4 and @cite_3 , but with Gaussian noise. Barigozzi and Brownlees ( @cite_8 ) consider the inference for stationary dependence structure built among variables, other than the parameters in the VAR model. In our work, we control the error bounds of Lasso and Dantzig type estimators for parameter matrices, with sub-Gaussian noise. Then we establish asymptotic distribution of test statistic based on this. | {
"cite_N": [
"@cite_3",
"@cite_4",
"@cite_6",
"@cite_8"
],
"mid": [
"1925915230",
"1836500409",
"2155350940",
"1751922918"
],
"abstract": [
"Many scientific and economic problems involve the analysis of highdimensional time series datasets. However, theoretical studies in highdimensional statistics to date rely primarily on the assumption of independent and identically distributed (i.i.d.) samples. In this work, we focus on stable Gaussian processes and investigate the theoretical properties of � 1-regularized estimates in two important statistical problems in the context of high-dimensional time series: (a) stochastic regression with serially correlated errors and (b) transition matrix estimation in vector autoregressive (VAR) models. We derive nonasymptotic upper bounds on the estimation errors of the regularized estimates and establish that consistent estimation under high-dimensional scaling is possible via � 1-regularization for a large class of stable processes under sparsity constraints. A key technical contribution of the work is to introduce a measure of stability for stationary processes using their spectral properties that provides insight into the effect of dependence on the accuracy of the regularized estimates. With this proposed stability measure, we establish some useful deviation bounds for dependent data, which can be used to study several important regularized estimates in a time series setting.",
"The vector autoregressive (VAR) model is a powerful tool in learning complex time series and has been exploited in many fields. The VAR model poses some unique challenges to researchers: On one hand, the dimensionality, introduced by incorporating multiple numbers of time series and adding the order of the vector autoregression, is usually much higher than the time series length; On the other hand, the temporal dependence structure naturally present in the VAR model gives rise to extra difficulties in data analysis. The regular way in cracking the VAR model is via \"least squares\" and usually involves adding different penalty terms (e.g., ridge or lasso penalty) in handling high dimensionality. In this manuscript, we propose an alternative way in estimating the VAR model. The main idea is, via exploiting the temporal dependence structure, formulating the estimating problem to a linear program. There is instant advantage of the proposed approach over the lasso-type estimators: The estimation equation can be decomposed to multiple sub-equations and accordingly can be solved efficiently using parallel computing. Besides that, we also bring new theoretical insights into the VAR model analysis. So far the theoretical results developed in high dimensions (e.g., Song and Bickel, 2011 and Kock and Callot, 2015) are based on stringent assumptions that are not transparent. Our results, on the other hand, show that the spectral norms of the transition matrices play an important role in estimation accuracy and build estimation and prediction consistency accordingly. Moreover, we provide some experiments on both synthetic and real-world equity data. We show that there are empirical advantages of our method over the lasso-type estimators in parameter estimation and forecasting.",
"One popular approach for nonstructural economic and financial forecasting is to include a large number of economic and financial variables, which has been shown to lead to significant improvements for forecasting, for example, by the dynamic factor models. A challenging issue is to determine which variables and (their) lags are relevant, especially when there is a mixture of serial correlation (temporal dynamics), high dimensional (spatial) dependence structure and moderate sample size (relative to dimensionality and lags). To this end, an integrated solution that addresses these three challenges simultaneously is appealing. We study the large vector auto regressions here with three types of estimates. We treat each variable's own lags different from other variables' lags, distinguish various lags over time, and is able to select the variables and lags simultaneously. We first show the consequences of using Lasso type estimate directly for time series without considering the temporal dependence. In contrast, our proposed method can still produce an estimate as efficient as an oracle under such scenarios. The tuning parameters are chosen via a data driven \"rolling scheme\" method to optimize the forecasting performance. A macroeconomic and financial forecasting problem is considered to illustrate its superiority over existing estimators.",
"This work proposes novel network analysis techniques for multivariate time series. We define the network of a multivariate time series as a graph where vertices denote the components of the process and edges denote non zero long run partial correlations. We then introduce a two step LASSO procedure, called NETS, to estimate high dimensional sparse Long Run Partial Correlation networks. This approach is based on a VAR approximation of the process and allows to decompose the long run linkages into the contribution of the dynamic and contemporaneous dependence relations of the system. The large sample properties of the estimator are analysed and we establish conditions for consistent selection and estimation of the non zero long run partial correlations. The methodology is illustrated with an application to a panel of U.S. bluechips."
]
} |
1812.03659 | 2905351737 | High-dimensional auto-regressive models provide a natural way to model influence between @math actors given multi-variate time series data for @math time intervals. While there has been considerable work on network estimation, there is limited work in the context of inference and hypothesis testing. In particular, prior work on hypothesis testing in time series has been restricted to linear Gaussian auto-regressive models. From a practical perspective, it is important to determine suitable statistical tests for connections between actors that go beyond the Gaussian assumption. In the context of time series models, confidence intervals present additional estimators since most estimators such as the Lasso and Dantzig selectors are biased which has led to estimators. In this paper we address these challenges and provide convergence in distribution results and confidence intervals for the multi-variate AR(p) model with sub-Gaussian noise, a generalization of Gaussian noise that broadens applicability and presents numerous technical challenges. The main technical challenge lies in the fact that unlike Gaussian random vectors, for sub-Gaussian vectors zero correlation does not imply independence. The proof relies on using an intricate truncation argument to develop novel concentration bounds for quadratic forms of dependent sub-Gaussian random variables. Our convergence in distribution results hold provided @math , where @math and @math refer to sparsity parameters which matches existed results for hypothesis testing with i.i.d. samples. We validate our theoretical results with simulation results for both block-structured and chain-structured networks. | In the high-dimensional hypothesis testing literature, there is some work regarding to testing for high-dimensional mean vector ( @cite_34 ), covariance matrices ( @cite_10 , @cite_33 ) and independence among variables ( @cite_24 ). While for testing on regression parameters, most work assumes i.i.d samples. @cite_19 , @cite_12 and @cite_22 proposes methods to test whether a covariate should be selected conditioning on the selection of some other covariates. A penalized score test depending on the tuning parameter @math is considered in @cite_25 . Our work follows the a line of work by @cite_2 , @cite_14 , @cite_5 and @cite_11 , the de-sparsifying or decorrelated literature. We construct a VAR version of decorrelated score test proposed by @cite_11 . Chen and Wu ( @cite_20 ) tackles the hypothesis testing problem for time series data as well, but they are testing the trend in a time series, instead of the autoregressive parameter which encodes the influence structure among variables. | {
"cite_N": [
"@cite_14",
"@cite_33",
"@cite_22",
"@cite_20",
"@cite_24",
"@cite_19",
"@cite_2",
"@cite_5",
"@cite_34",
"@cite_10",
"@cite_25",
"@cite_12",
"@cite_11"
],
"mid": [
"",
"",
"1535995764",
"2796273868",
"2049513258",
"",
"",
"",
"1976889927",
"",
"1773660948",
"",
"2964331163"
],
"abstract": [
"",
"",
"We develop a general approach to valid inference after model selection. In a nutshell, our approach produces post-selection inferences with the same frequency guarantees as those given by data splitting but are more powerful. At the core of our framework is a result that characterizes the distribution of a post-selection estimator conditioned on the selection event. We specialize the approach to model selection by the lasso to form valid condence intervals for the selected coecients and test whether all relevant variables have been included in the model.",
"ABSTRACTThe article considers statistical inference for trends of high-dimensional time series. Based on a modified L2 distance between parametric and nonparametric trend estimators, we propose a d...",
"A simple statistic is proposed for testing the complete independence of random variables having a multivariate normal distribution. The asymptotic null distribution of this statistic, as both the sample size and the number of variables go to infinity, is shown to be normal. Consequently, this test can be used when the number of variables is not small relative to the sample size and, in particular, even when the number of variables exceeds the sample size. The finite sample size performance of the normal approximation is evaluated in a simulation study and the results are compared to those of the likelihood ratio test. Copyright 2005, Oxford University Press.",
"",
"",
"",
"In this article, we consider the problem of testing that the mean vector @m=0 in the model x\"[email protected]+Cz\"j,j=1,...,N, where z\"j are random p-vectors, z\"j=(z\"i\"j,...,z\"p\"j)^' and z\"i\"j are independently and identically distributed with finite four moments, i=1,...,p,j=1,...,N; that is x\"i need not be normally distributed. We shall assume that C is a pxp non-singular matrix, and there are fewer observations than the dimension, [email protected]?p. We consider the test statistic T=[[email protected]?^'D\"s^-^[email protected]?-np (n-2)] [2trR^2-p^2 n]^1^2, where [email protected]? is the sample mean vector, S=(s\"i\"j) is the sample covariance matrix, D\"S= diag (s\"1\"1,...,s\"p\"p),R=D\"s^-^1^2SD\"s^-^1^2 and n=N-1. The asymptotic null and non-null distributions of the test statistic T are derived.",
"",
"In recent years, there has been considerable theoretical development regarding variable selection consistency of penalized regression techniques, such as the lasso. However, there has been relatively little work on quantifying the uncertainty in these selection procedures. In this paper, we propose a new method for inference in high dimensions using a score test based on penalized regression. In this test, we perform penalized regression of an outcome on all but a single feature, and test for correlation of the residuals with the held-out feature. This procedure is applied to each feature in turn. Interestingly, when an @math penalty is used, the sparsity pattern of the lasso corresponds exactly to a decision based on the proposed test. Further, when an @math penalty is used, the test corresponds precisely to a score test in a mixed effects model, in which the effects of all but one feature are assumed to be random. We formulate the hypothesis being tested as a compromise between the null hypotheses tested in simple linear regression on each feature and in multiple linear regression on all features, and develop reference distributions for some well-known penalties. We also examine the behavior of the test on real and simulated data.",
"",
""
]
} |
1812.03659 | 2905351737 | High-dimensional auto-regressive models provide a natural way to model influence between @math actors given multi-variate time series data for @math time intervals. While there has been considerable work on network estimation, there is limited work in the context of inference and hypothesis testing. In particular, prior work on hypothesis testing in time series has been restricted to linear Gaussian auto-regressive models. From a practical perspective, it is important to determine suitable statistical tests for connections between actors that go beyond the Gaussian assumption. In the context of time series models, confidence intervals present additional estimators since most estimators such as the Lasso and Dantzig selectors are biased which has led to estimators. In this paper we address these challenges and provide convergence in distribution results and confidence intervals for the multi-variate AR(p) model with sub-Gaussian noise, a generalization of Gaussian noise that broadens applicability and presents numerous technical challenges. The main technical challenge lies in the fact that unlike Gaussian random vectors, for sub-Gaussian vectors zero correlation does not imply independence. The proof relies on using an intricate truncation argument to develop novel concentration bounds for quadratic forms of dependent sub-Gaussian random variables. Our convergence in distribution results hold provided @math , where @math and @math refer to sparsity parameters which matches existed results for hypothesis testing with i.i.d. samples. We validate our theoretical results with simulation results for both block-structured and chain-structured networks. | As mentioned earlier, our work is most closely related to the prior work of @cite_7 , which provides a hypothesis testing framework with high-dimensional Gaussian time series as a special case. In our work, we consider the more general and technically challenging case of sub-Gaussian vector auto-regressive models. Throughout this paper, we provide a comparison to results derived in this work for the Gaussian case. | {
"cite_N": [
"@cite_7"
],
"mid": [
"2228628292"
],
"abstract": [
"We propose a new inferential framework for constructing confidence regions and testing hypotheses in statistical models specified by a system of high dimensional estimating equations. We construct an influence function by projecting the fitted estimating equations to a sparse direction obtained by solving a large-scale linear program. Our main theoretical contribution is to establish a unified Z-estimation theory of confidence regions for high dimensional problems. Different from existing methods, all of which require the specification of the likelihood or pseudo-likelihood, our framework is likelihood-free. As a result, our approach provides valid inference for a broad class of high dimensional constrained estimating equation problems, which are not covered by existing methods. Such examples include, noisy compressed sensing, instrumental variable regression, undirected graphical models, discriminant analysis and vector autoregressive models. We present detailed theoretical results for all these examples. Finally, we conduct thorough numerical simulations, and a real dataset analysis to back up the developed theoretical results."
]
} |
1812.03527 | 2902848927 | Skin lesion identification is a key step toward dermatological diagnosis. When describing a skin lesion, it is very important to note its body site distribution as many skin diseases commonly affect particular parts of the body. To exploit the correlation between skin lesions and their body site distributions, in this study, we investigate the possibility of improving skin lesion classification using the additional context information provided by body location. Specifically, we build a deep multi-task learning (MTL) framework to jointly optimize skin lesion classification and body location classification (the latter is used as an inductive bias). Our MTL framework uses the state-of-the-art ImageNet pretrained model with specialized loss functions for the two related tasks. Our experiments show that the proposed MTL based method performs more robustly than its standalone (single-task) counterpart. | Most of the existing works @cite_23 @cite_20 @cite_7 only focus on one or a few skin disease and solve the problem using conventional machine learning approach, i.e., extracting manually engineered features from segmented lesion patches and classifying with a linear classifier such as SVM. While in our study, we target a more challenging problem where all skin diseases are considered. | {
"cite_N": [
"@cite_20",
"@cite_7",
"@cite_23"
],
"mid": [
"25676111",
"146528819",
"197962081"
],
"abstract": [
"Dermoscopy image analysis technology is discussed based on Chinese in this chapter. It includes four aspects: preprocessing, segmentation, feature extraction and classification. Firstly, in preprocessing stage, hair is extracted out according to the elongate of connected region, and then removed from the image by using PDE-based image inpainting technology. Secondly, a novel dermoscopy image segmentation algorithm is introduced using self-generating neural network (SGNN) combined with genetic algorithm (GA). And in the feature description stage, the features including color, texture, shape and border are extracted for the lesion object. Lastly, the model of combined neural network classifier is employed to classify the lesion object successfully with a sensitivity and specificity of 93.3 and 96.7 respectively. Based on the image analysis method disscussed in this chapter, an automatic analysis system of dermoscopy images of Chinese is successfully developed and has been applied for the clinical diagnosis of skin tumors.",
"An image based system implementing a well-known diagnostic method is disclosed for the automatic detection of melanomas as support to clinicians. The software procedure is able to recognize automatically the skin lesion within the digital image, measure morphological and chromatic parameters, carry out a suitable classification for detecting the dermoscopic structures provided by the 7-Point Checklist. Advanced techniques are introduced at different stages of the image processing pipeline, including the border detection, the extraction of low-level features and scoring of high order features.",
"In this chapter a software system for the automated detection of melanoma over dermoscopic images is presented. The analysis is carried out by supporting in the “ABCD Rule” medical algorithm, undertaking the automated detection and characterization of the corresponding indicators. For this purpose the system uses different image processing techniques and three supervised machine learning tasks. To test the robustness of the system the different indicators of the algorithm are tested, obtaining good results of accuracy in all of them, and there also was determined in a direct way the diagnostic capacity of the system, obtaining results of 81.25 of sensitivity and 77.14 of specificity. Moreover, the system is also capable of analyzing macroscopic images, having been designd with multiplatform architecture and being firmly oriented to teledermatology, which is increasingly used."
]
} |
1812.03274 | 2904496675 | This paper presents a manipulation planning algorithm for robots to reorient objects. It automatically finds a sequence of robot motion that manipulates and prepares an object for specific tasks. Examples of the preparatory manipulation planning problems include reorienting an electric drill to cut holes, reorienting workpieces for assembly, and reorienting cargo for packing, etc. The proposed algorithm could plan single and dual arm manipulation sequences to solve the problems. The mechanism under the planner is a regrasp graph which encodes grasp configurations and object poses. The algorithms search the graph to find a sequence of robot motion to reorient objects. The planner is able to plan both single and dual arm manipulation. It could also automatically determine whether to use a single arm, dual arms, or their combinations to finish given tasks. The planner is examined by various humanoid robots like Nextage, HRP2Kai, HRP5P, etc., using both simulation and real-world experiments. | Recent work included more complicated grasp planners and was integrally considered with backtracking, geometrical reasoning, and motion planning. For example, @cite_25 presented a framework which integrated motion planning and preparatory manipulation planning to move objects among obstacles. @cite_28 used shape primitives to plan the grasps of a multi-finger hand and implemented the preparatory manipulation planning of a cup using the multi-finger hand. @cite_20 used Bayesian Network to model task constraints and plan the preparatory manipulation sequence of containers and toys. @cite_12 used integral primitive-based prehensile and non-prehensile grasp planning to exercise preparatory manipulation of everyday objects. Lozano- @cite_1 used symbolic search to find manipulation sequences under various constraints. The method was used to plan sequences for a mobile manipulator to rearrange objects. Similarly, @cite_23 , @cite_16 , Toussaint @cite_18 , and @cite_27 respectively used symbolic reasoning to remove obstacles and pick up or rearrange objects. Especially, Toussaint @cite_18 and @cite_27 presented algorithms to plan motion sequences for robots to stack distributed objects, considering both task and motion constraints. @cite_6 presented a planner considering the constraints from coordinated manipulators. More extensively, @cite_2 and @cite_3 respectively presented a framework work that used integral prehensile and non-prehensile planning to generate sequential manipulation. | {
"cite_N": [
"@cite_18",
"@cite_28",
"@cite_1",
"@cite_16",
"@cite_6",
"@cite_3",
"@cite_27",
"@cite_23",
"@cite_2",
"@cite_20",
"@cite_25",
"@cite_12"
],
"mid": [
"2403069916",
"",
"2031738727",
"",
"2559038552",
"2736374714",
"2468062401",
"2141841102",
"1674160379",
"1548434535",
"2044995998",
"2296486637"
],
"abstract": [
"We consider problems of sequential robot manipulation (aka. combined task and motion planning) where the objective is primarily given in terms of a cost function over the final geometric state, rather than a symbolic goal description. In this case we should leverage optimization methods to inform search over potential action sequences. We propose to formulate the problem holistically as a 1st- order logic extension of a mathematical program: a non-linear constrained program over the full world trajectory where the symbolic state-action sequence defines the (in-)equality constraints. We tackle the challenge of solving such programs by proposing three levels of approximation: The coarsest level introduces the concept of the effective end state kinematics, parametrically describing all possible end state configurations conditional to a given symbolic action sequence. Optimization on this level is fast and can inform symbolic search. The other two levels optimize over interaction keyframes and eventually over the full world trajectory across interactions. We demonstrate the approach on a problem of maximizing the height of a physically stable construction from an assortment of boards, cylinders and blocks.",
"",
"In this paper, we describe a strategy for integrated task and motion planning based on performing a symbolic search for a sequence of high-level operations, such as pick, move and place, while postponing geometric decisions. Partial plans (skeletons) in this search thus pose a geometric constraintsatisfaction problem (CSP), involving sequences of placements and paths for the robot, and grasps and locations of objects. We propose a formulation for these problems in a discretized configuration space for the robot. The resulting problems can be solved using existing methods for discrete CSP.",
"",
"Constrained motion planning and Manipulation planning, for generic non-linear constraints, highly rely on the ability of solving non-linear equations. The Newton-Raphson method, often used in this context, is discontinuous with respect to its input and point wise path projection can lead to discontinuous constrained paths. The discontinuities come from the pseudo-inverse involved at each iteration of the Newton-Raphson algorithm. An interval of continuity for one iteration is derived from an upper bound of the norm of the Hessian of the constraints. Such a bound is easy to compute for constraints involving joint positions and orientations. We introduce two heuristical path projection algorithms and empirically show that they generate continuous constrained paths. The algorithms are integrated into a motion planning algorithm to empirically assert the continuity of the solution.",
"In this paper we propose a method for motion planning and feedback control of hybrid, dynamic, and non-prehensile manipulation tasks. We outline five subproblems to address this: determining a set of manipulation primitives, choosing a sequence of tasks, picking transition states, motion planning for each individual primitive, and stabilizing each mode using feedback control. We apply the framework to plan a sequence of motions for manipulating a block with a planar 3R manipulator. We demonstrate preliminary experimental results for a block resting on the manipulator with a desired goal state on a ledge outside of the robot's workspace. The planned primitives reorient the block using a series of fixed, rolling, and sliding contact modes, and throw it to the goal state.",
"",
"The need for combined task and motion planning in robotics is well understood. Solutions to this problem have typically relied on special purpose, integrated implementations of task planning and motion planning algorithms. We propose a new approach that uses off-the-shelf task planners and motion planners and makes no assumptions about their implementation. Doing so enables our approach to directly build on, and benefit from, the vast literature and latest advances in task planning and motion planning. It uses a novel representational abstraction and requires only that failures in computing a motion plan for a high-level action be identifiable and expressible in the form of logical predicates at the task level. We evaluate the approach and illustrate its robustness through a number of experiments using a state-of-the-art robotics simulator and a PR2 robot. These experiments show the system accomplishing a diverse set of challenging tasks such as taking advantage of a tray when laying out a table for dinner and picking objects from cluttered environments where other objects need to be re-arranged before the target object can be reached.",
"Manipulation planning involves planning the combined motion of objects in the environment as well as the robot motions to achieve them. In this paper, we explore a hierarchical approach to planning sequences of non-prehensile and prehensile actions. We subdivide the planning problem into three stages (object contacts, object poses and robot contacts) and thereby reduce the size of search space that is explored. We show that this approach is more efficient than earlier strategies that search in the combined robot-object configuration space directly.",
"In this paper, we present an approach towards autonomous grasping of objects according to their category and a given task. Recent advances in the field of object segmentation and categorization as well as task-based grasp inference have been leveraged by integrating them into one pipeline. This allows us to transfer task-specific grasp experience between objects of the same category. The effectiveness of the approach is demonstrated on the humanoid robot ARMAR-IIIa.",
"This paper deals with motion planning for robots manipulating movable objects among obstacles. We propose a general manipulation planning approach capable of addressing continuous sets for modeling both the possible grasps and the stable placements of the movable object, rather than discrete sets generally assumed by the previous approaches. The proposed algorithm relies on a topological property that characterizes the existence of solutions in the subspace of configurations where the robot grasps the object placed at a stable position. It allows us to devise a manipulation planner that captures in a probabilistic roadmap the connectivity of sub-dimensional manifolds of the composite configuration space. Experiments conducted with the planner in simulated environments demonstrate its efficacy to solve complex manipulation problems.",
""
]
} |
1812.03274 | 2904496675 | This paper presents a manipulation planning algorithm for robots to reorient objects. It automatically finds a sequence of robot motion that manipulates and prepares an object for specific tasks. Examples of the preparatory manipulation planning problems include reorienting an electric drill to cut holes, reorienting workpieces for assembly, and reorienting cargo for packing, etc. The proposed algorithm could plan single and dual arm manipulation sequences to solve the problems. The mechanism under the planner is a regrasp graph which encodes grasp configurations and object poses. The algorithms search the graph to find a sequence of robot motion to reorient objects. The planner is able to plan both single and dual arm manipulation. It could also automatically determine whether to use a single arm, dual arms, or their combinations to finish given tasks. The planner is examined by various humanoid robots like Nextage, HRP2Kai, HRP5P, etc., using both simulation and real-world experiments. | There are also lots of studies on dual-arm or multi-arm preparatory manipulation planning. For example, @cite_30 used predefined grasps to plan sequential manipulation for multiple manipulators in early time. @cite_17 applied the grasp-placement tables developed by @cite_14 to dual-arm robots. @cite_22 used decomposition to plan grasps and implemented dual-arm regrasp of complicated mesh models. Graphs, instead of grasp-placement tables, were used in the planner. @cite_31 presented regrasp and hand-over planning of dual-arm robots across multi-modal configuration spaces, and implemented a practical system in @cite_8 . Details of multi-modal motion planning could be found in @cite_0 , which made a concrete description of the concept and presented several implementations. More recent work like @cite_9 used dual-arm sequential manipulation planning to wrap up the fabric. It is also quite relevant except that reorienting directions are constrained by the wrapping heuristics. @cite_10 , @cite_11 , and Suarez- @cite_29 respectively applied integral prehensile and non-prehensile sequential dual-arm motion planning to the preparatory manipulation of a chair. @cite_7 presented a framework for simultaneous handover and visual recognition. | {
"cite_N": [
"@cite_30",
"@cite_14",
"@cite_11",
"@cite_22",
"@cite_7",
"@cite_8",
"@cite_9",
"@cite_29",
"@cite_0",
"@cite_31",
"@cite_10",
"@cite_17"
],
"mid": [
"1965244358",
"",
"2585588704",
"2088078109",
"",
"2005777097",
"2736497041",
"",
"",
"2036411373",
"",
"1934518279"
],
"abstract": [
"We apply manipulation planning to computer animation. A new path planner is presented that automatically computes the collision-free trajectories for several cooperating arms to manipulate a movable object between two configurations. This implemented planner is capable of dealing with complicated tasks where regrasping is involved. In addition, we present a new inverse kinematics algorithm for the human arms. This algorithm is utilized by the planner for the generation of realistic human arm motions as they manipulate objects. We view our system as a tool for facilitating the production of animation.",
"",
"Closed kinematic chains are created whenever multiple robot arms concurrently manipulate a single object. The closed-chain constraint, when coupled with robot joint limits, dramatically changes the connectivity of the configuration space. We propose a regrasping move, termed “IK-switch,” which allows efficiently bridging components of the configuration space that are otherwise mutually disconnected. This move, combined with several other developments, such as a method to stabilize the manipulated object using the environment, a new tree structure, and a compliant control scheme, enables us to address complex closed-chain manipulation tasks, such as flipping a chair frame, which is otherwise impossible to realize using existing multi-arm planning methods.",
"This paper proposes a planning framework to deal with the problem of computing the motion of a robot with dual arm hand, during an object pick-and-place task. We consider the situation where the start and goal configurations of the object constrain the robot to grasp the object with one hand, to give it to the other hand, before placing it in its final configuration. To realize such a task, the proposed framework treats the grasp computation, for one or two multi-fingered hands, of an arbitrarily-shaped object, the exchange configuration and finally the motion of the robot arms and body. In order to improve the planner performance, a context-independent grasp list is computed offline for each hand and for the given object as well as computed offline roadmap that will be adapted according to the environment composition. Simulation results show the planner performance on a complex scenario.",
"",
"This research develops a robotic vision and manipulation technologies for random bin-picking. Especially, we focus on the manipulation technology where a dual-arm manipulator first picks up an object from the pile, then regrasps it from the right hand to the left hand, and finally places it to the fixture. We first explain an overview of our research project. Then, we explain about the grasp manipulation planner for a dual-arm manipulator. Here, our grasp planner is well applied for objects which can be approximated by multiple cylinders. Also, our pick-and-place planner can effectively find a regrasping posture of an object. In addition to the manipulation technology, we briefly explain about our vision technology to measure the position orientation of an object. Finally, to show the effectiveness of our proposed approach, we show an experimental result of a dual-arm manipulator.",
"This paper proposes a motion-planning method for a wrapping-with-fabric task. To wrap, robots require not only offhand paths (which represent the movement of a grasping point on fabric) but also actual robot motion. When a robot handles fabric, it must consider the range of motion and collisions. To overcome such limitations, the robot must pass fabric inter- and intra-hand. The proposition is based on the reliability of its movements. The method may be used to plan a combination of inter- and intra-hand passing motions to generate the basic movements for wrapping along an offhand path. Experiments show how, by using the proposed planning method, a dual-arm robot can wrap an L-shaped pipe.",
"",
"",
"",
"",
"A pick-and-place operation in a 3-dimensional environment is a basic operation for humans and multi-purpose manipulators. However, there may be a difficult problem for such manipulators. Especially, if the object cannot be moved with a single grasp, regrasping, which can be a time-consuming process, should be carried out. Regrasping, given initial and final poses of the target object, is a construction of sequential transition of object poses that are compatible with two poses in the point of grasp configuration. This paper presents a novel approach for solving the regrasp problem. The approach consists of a preprocessing and a planning stage. Preprocessing, which is done only once for a given robot, generates a look-up table which has information on kinematically feasible task space of the end-effector throughout the entire workspace. Then, using this table, the planning automatically determines a possible intermediate location, pose and regrasp sequence leading from the pick-up to put-down grasp. With a redundant robot, it is shown experimentally that the presented method is complete in the entire workspace and can be implemented in real-time applications due to rapid regrasp planning time. The regrasp planner was combined with an existing path."
]
} |
1812.03245 | 2905505753 | We propose a self-supervised learning framework that uses unlabeled monocular video sequences to generate large-scale supervision for training a Visual Odometry (VO) frontend, a network which computes pointwise data associations across images. Our self-improving method enables a VO frontend to learn over time, unlike other VO and SLAM systems which require time-consuming hand-tuning or expensive data collection to adapt to new environments. Our proposed frontend operates on monocular images and consists of a single multi-task convolutional neural network which outputs 2D keypoints locations, keypoint descriptors, and a novel point stability score. We use the output of VO to create a self-supervised dataset of point correspondences to retrain the frontend. When trained using VO at scale on 2.5 million monocular images from ScanNet, the stability classifier automatically discovers a ranking for keypoints that are not likely to help in VO, such as t-junctions across depth discontinuities, features on shadows and highlights, and dynamic objects like people. The resulting frontend outperforms both traditional methods (SIFT, ORB, AKAZE) and deep learning methods (SuperPoint and LF-Net) in a 3D-to-2D pose estimation task on ScanNet. | SIPS @cite_13 uses a ranking loss to estimate a concise set of interest points. IMIPS @cite_25 uses a similar approach to SIPS for learning implicit keypoint correspondence between pairs of images, without the need for descriptors. MegaDepth @cite_9 uses SfM combined with semantic segmentation to label a large set of outdoor images. All these methods rely on external supervision which is problematic because it introduces a second SLAM system with an additional set of limitations and dependencies, which all are inherited by the learned system (see Figure b). | {
"cite_N": [
"@cite_9",
"@cite_13",
"@cite_25"
],
"mid": [
"2796059865",
"2799196020",
"2903234512"
],
"abstract": [
"Single-view depth prediction is a fundamental problem in computer vision. Recently, deep learning methods have led to significant progress, but such methods are limited by the available training data. Current datasets based on 3D sensors have key limitations, including indoor-only images (NYU), small numbers of training examples (Make3D), and sparse sampling (KITTI). We propose to use multi-view Internet photo collections, a virtually unlimited data source, to generate training data via modern structure-from-motion and multi-view stereo (MVS) methods, and present a large depth dataset called MegaDepth based on this idea. Data derived from MVS comes with its own challenges, including noise and unreconstructable objects. We address these challenges with new data cleaning methods, as well as automatically augmenting our data with ordinal depth relations generated using semantic segmentation. We validate the use of large amounts of Internet data by showing that models trained on MegaDepth exhibit strong generalization-not only to novel scenes, but also to other diverse datasets including Make3D, KITTI, and DIW, even when no images from those datasets are seen during training.",
"Detecting interest points is a key component of vision-based estimation algorithms, such as visual odometry or visual SLAM. Classically, interest point detection has been done with methods such as Harris, FAST, or DoG. Recently, better detectors have been proposed based on Neural Networks. Traditionally, interest point detectors have been designed to maximize repeatability or matching score. Instead, we pursue another metric, which we call succinctness. This metric captures the minimum amount of interest points that need to be extracted in order to achieve accurate relative pose estimation. Extracting a minimum amount of interest points is attractive for many applications, because it reduces computational load, memory, and, potentially, data transmission. We propose a novel reinforcement- and ranking-based training framework, which uses a full relative pose estimation pipeline during training. It can be trained in an unsupervised manner, without pose or 3D point ground truth. Using this training framework, we present a detector which outperforms previous interest point detectors in terms of succinctness on a variety of publicly available datasets.",
"The extraction and matching of interest points is a prerequisite for many geometric computer vision problems. Traditionally, matching has been achieved by assigning descriptors to interest points and matching points that have similar descriptors. In this paper, we propose a method by which interest points are instead already implicitly matched at detection time. With this, descriptors do not need to be calculated, stored, communicated, or matched any more. This is achieved by a convolutional neural network with multiple output channels and can be thought of as a collection of a variety of detectors, each specialized to specific visual features. This paper describes how to design and train such a network in a way that results in successful relative pose estimation performance despite the limitation on interest point count. While the overall matching score is slightly lower than with traditional methods, the approach is descriptor free and thus enables localization systems with a significantly smaller memory footprint and multi-agent localization systems with lower bandwidth requirements. The network also outputs the confidence for a specific interest point resulting in a valid match. We evaluate performance relative to state-of-the-art alternatives."
]
} |
1812.03245 | 2905505753 | We propose a self-supervised learning framework that uses unlabeled monocular video sequences to generate large-scale supervision for training a Visual Odometry (VO) frontend, a network which computes pointwise data associations across images. Our self-improving method enables a VO frontend to learn over time, unlike other VO and SLAM systems which require time-consuming hand-tuning or expensive data collection to adapt to new environments. Our proposed frontend operates on monocular images and consists of a single multi-task convolutional neural network which outputs 2D keypoints locations, keypoint descriptors, and a novel point stability score. We use the output of VO to create a self-supervised dataset of point correspondences to retrain the frontend. When trained using VO at scale on 2.5 million monocular images from ScanNet, the stability classifier automatically discovers a ranking for keypoints that are not likely to help in VO, such as t-junctions across depth discontinuities, features on shadows and highlights, and dynamic objects like people. The resulting frontend outperforms both traditional methods (SIFT, ORB, AKAZE) and deep learning methods (SuperPoint and LF-Net) in a 3D-to-2D pose estimation task on ScanNet. | Self-supervised VO. SuperPoint @cite_5 takes a different approach by self-labeling a large set of images and using homographies of images to learn correspondence. While this is a surprisingly powerful technique given its simplicity, it is limited by its reliance on static images and cannot learn from real illumination changes and correspondence across difficult non-planar scenes. The backend in this method is a simple homography model, where the homographies are generated synthetically. This method falls into the self-supervised VO paradigm (see Figure c). | {
"cite_N": [
"@cite_5"
],
"mid": [
"2963462995"
],
"abstract": [
"This paper presents a self-supervised framework for training interest point detectors and descriptors suitable for a large number of multiple-view geometry problems in computer vision. As opposed to patch-based neural networks, our fully-convolutional model operates on full-sized images and jointly computes pixel-level interest point locations and associated descriptors in one forward pass. We introduce Homographic Adaptation, a multi-scale, multi-homography approach for boosting interest point detection repeatability and performing cross-domain adaptation (e.g., synthetic-to-real). Our model, when trained on the MS-COCO generic image dataset using Homographic Adaptation, is able to repeatedly detect a much richer set of interest points than the initial pre-adapted deep model and any other traditional corner detector. The final system gives rise to state-of-the-art homography estimation results on HPatches when compared to LIFT, SIFT and ORB."
]
} |
1812.03245 | 2905505753 | We propose a self-supervised learning framework that uses unlabeled monocular video sequences to generate large-scale supervision for training a Visual Odometry (VO) frontend, a network which computes pointwise data associations across images. Our self-improving method enables a VO frontend to learn over time, unlike other VO and SLAM systems which require time-consuming hand-tuning or expensive data collection to adapt to new environments. Our proposed frontend operates on monocular images and consists of a single multi-task convolutional neural network which outputs 2D keypoints locations, keypoint descriptors, and a novel point stability score. We use the output of VO to create a self-supervised dataset of point correspondences to retrain the frontend. When trained using VO at scale on 2.5 million monocular images from ScanNet, the stability classifier automatically discovers a ranking for keypoints that are not likely to help in VO, such as t-junctions across depth discontinuities, features on shadows and highlights, and dynamic objects like people. The resulting frontend outperforms both traditional methods (SIFT, ORB, AKAZE) and deep learning methods (SuperPoint and LF-Net) in a 3D-to-2D pose estimation task on ScanNet. | Geometric Matching Networks @cite_0 and Deep Image Homography Estimation @cite_20 use a similar self-supervision strategy to create training data for estimating image transformations. However, these methods lack interest points and point correspondences, which are typically required for doing higher-level computer vision tasks such as SLAM and SfM. | {
"cite_N": [
"@cite_0",
"@cite_20"
],
"mid": [
"2604233003",
"2439114332"
],
"abstract": [
"We address the problem of determining correspondences between two images in agreement with a geometric model such as an affine or thin-plate spline transformation, and estimating its parameters. The contributions of this work are three-fold. First, we propose a convolutional neural network architecture for geometric matching. The architecture is based on three main components that mimic the standard steps of feature extraction, matching and simultaneous inlier detection and model parameter estimation, while being trainable end-to-end. Second, we demonstrate that the network parameters can be trained from synthetically generated imagery without the need for manual annotation and that our matching layer significantly increases generalization capabilities to never seen before images. Finally, we show that the same model can perform both instance-level and category-level matching giving state-of-the-art results on the challenging Proposal Flow dataset.",
"We present a deep convolutional neural network for estimating the relative homography between a pair of images. Our feed-forward network has 10 layers, takes two stacked grayscale images as input, and produces an 8 degree of freedom homography which can be used to map the pixels from the first image to the second. We present two convolutional neural network architectures for HomographyNet: a regression network which directly estimates the real-valued homography parameters, and a classification network which produces a distribution over quantized homographies. We use a 4-point homography parameterization which maps the four corners from one image into the second image. Our networks are trained in an end-to-end fashion using warped MS-COCO images. Our approach works without the need for separate local feature detection and transformation estimation stages. Our deep models are compared to a traditional homography estimator based on ORB features and we highlight the scenarios where HomographyNet outperforms the traditional technique. We also describe a variety of applications powered by deep homography estimation, thus showcasing the flexibility of a deep learning approach."
]
} |
1812.03245 | 2905505753 | We propose a self-supervised learning framework that uses unlabeled monocular video sequences to generate large-scale supervision for training a Visual Odometry (VO) frontend, a network which computes pointwise data associations across images. Our self-improving method enables a VO frontend to learn over time, unlike other VO and SLAM systems which require time-consuming hand-tuning or expensive data collection to adapt to new environments. Our proposed frontend operates on monocular images and consists of a single multi-task convolutional neural network which outputs 2D keypoints locations, keypoint descriptors, and a novel point stability score. We use the output of VO to create a self-supervised dataset of point correspondences to retrain the frontend. When trained using VO at scale on 2.5 million monocular images from ScanNet, the stability classifier automatically discovers a ranking for keypoints that are not likely to help in VO, such as t-junctions across depth discontinuities, features on shadows and highlights, and dynamic objects like people. The resulting frontend outperforms both traditional methods (SIFT, ORB, AKAZE) and deep learning methods (SuperPoint and LF-Net) in a 3D-to-2D pose estimation task on ScanNet. | Learning Stability. In @cite_3 , Zhou al present an unsupervised approach for learning monocular depth and relative pose that does not rely on external ground truth. The model also predicts an explainability mask, which is similar to the stability classifier presented in this work because it discovers dynamic objects like people without explicitly being trained to do so. It operates on a pair of images, rather than SuperPointVO which operates on a single image. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2609883120"
],
"abstract": [
"We present an unsupervised learning framework for the task of monocular depth and camera motion estimation from unstructured video sequences. In common with recent work [10, 14, 16], we use an end-to-end learning approach with view synthesis as the supervisory signal. In contrast to the previous work, our method is completely unsupervised, requiring only monocular video sequences for training. Our method uses single-view depth and multiview pose networks, with a loss based on warping nearby views to the target using the computed depth and pose. The networks are thus coupled by the loss during training, but can be applied independently at test time. Empirical evaluation on the KITTI dataset demonstrates the effectiveness of our approach: 1) monocular depth performs comparably with supervised methods that use either ground-truth pose or depth for training, and 2) pose estimation performs favorably compared to established SLAM systems under comparable input settings."
]
} |
1812.03235 | 2904940351 | Knowledge graphs are used to represent relational information in terms of triples. To enable learning about domains, embedding models, such as tensor factorization models, can be used to make predictions of new triples. Often there is background taxonomic information (in terms of subclasses and subproperties) that should also be taken into account. We show that existing fully expressive (a.k.a. universal) models cannot provably respect subclass and subproperty information. We show that minimal modifications to an existing knowledge graph completion method enables injection of taxonomic information. Moreover, we prove that our model is fully expressive, assuming a lower-bound on the size of the embeddings. Experimental results on public knowledge graphs show that despite its simplicity our approach is surprisingly effective. | There is a large family of link prediction models based on soft first-order logic rules @cite_2 @cite_11 @cite_9 . While these models can be easily integrated with background taxonomic information, they typically cannot generalize to unseen cases beyond their rules. Exceptions include @cite_3 @cite_4 which combine (stacked layers of) soft rules with entity embeddings, but these models have only applied to property prediction. Approaches based on path-constrained random walks (e.g., @cite_6 ) suffer from similar limitations as they have been shown to be a subset of probabilistic logic-based models @cite_7 . | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_9",
"@cite_3",
"@cite_6",
"@cite_2",
"@cite_11"
],
"mid": [
"2775300540",
"2789351899",
"2185448989",
"2464810161",
"1529533208",
"1977970897",
"1824971879"
],
"abstract": [
"Statistical relational AI (StarAI) aims at reasoning and learning in noisy domains described in terms of objects and relationships by combining probability with first-order logic. With huge advances in deep learning in the current years, combining deep networks with first-order logic has been the focus of several recent studies. Many of the existing attempts, however, only focus on relations and ignore object properties. The attempts that do consider object properties are limited in terms of modelling power or scalability. In this paper, we develop relational neural networks (RelNNs) by adding hidden layers to relational logistic regression (the relational counterpart of logistic regression). We learn latent properties for objects both directly and through general rules. Back-propagation is used for training these models. A modular, layer-wise architecture facilitates utilizing the techniques developed within deep learning community to our architecture. Initial experiments on eight tasks over three real-world datasets show that RelNNs are promising models for relational learning.",
"The aim of statistical relational learning is to learn statistical models from relational or graph-structured data. Three main statistical relational learning paradigms include weighted rule learning, random walks on graphs, and tensor factorization. These paradigms have been mostly developed and studied in isolation for many years, with few works attempting at understanding the relationship among them or combining them. In this paper, we study the relationship between the path ranking algorithm (PRA), one of the most well-known relational learning methods in the graph random walk paradigm, and relational logistic regression (RLR), one of the recent developments in weighted rule learning. We provide a simple way to normalize relations and prove that relational logistic regression using normalized relations generalizes the path ranking algorithm. This result provides a better understanding of relational learning, especially for the weighted rule learning and graph random walk paradigms. It opens up the possibility of using the more flexible RLR rules within PRA models and even generalizing both by including normalized and unnormalized relations in the same model.",
"Logistic regression is a commonly used representation for aggregators in Bayesian belief networks when a child has multiple parents. In this paper we consider extending logistic regression to relational models, where we want to model varying populations and interactions among parents. In this paper, we first examine the representational problems caused by population variation. We show how these problems arise even in simple cases with a single parametrized parent, and propose a linear relational logistic regression which we show can represent arbitrary linear (in population size) decision thresholds, whereas the traditional logistic regression cannot. Then we examine representing interactions among the parents of a child node, and representing non-linear dependency on population size. We propose a multi-parent relational logistic regression which can represent interactions among parents and arbitrary polynomial decision thresholds. Finally, we show how other well-known aggregators can be represented using this relational logistic regression.",
"Relational logistic regression (RLR) is a representation of conditional probability in terms of weighted formulae for modelling multi-relational data. In this paper, we develop a learning algorithm for RLR models. Learning an RLR model from data consists of two steps: 1- learning the set of formulae to be used in the model (a.k.a. structure learning) and learning the weight of each formula (a.k.a. parameter learning). For structure learning, we deploy Schmidt and Murphy's hierarchical assumption: first we learn a model with simple formulae, then more complex formulae are added iteratively only if all their sub-formulae have proven effective in previous learned models. For parameter learning, we convert the problem into a non-relational learning problem and use an off-the-shelf logistic regression learning algorithm from Weka, an open-source machine learning tool, to learn the weights. We also indicate how hidden features about the individuals can be incorporated into RLR to boost the learning performance. We compare our learning algorithm to other structure and parameter learning algorithms in the literature, and compare the performance of RLR models to standard logistic regression and RDN-Boost on a modified version of the MovieLens data-set.",
"Relational machine learning studies methods for the statistical analysis of relational, or graph-structured, data. In this paper, we provide a review of how such statistical models can be “trained” on large knowledge graphs, and then used to predict new facts about the world (which is equivalent to predicting new edges in the graph). In particular, we discuss two fundamentally different kinds of statistical relational models, both of which can scale to massive data sets. The first is based on latent feature models such as tensor factorization and multiway neural networks. The second is based on mining observable patterns in the graph. We also show how to combine these latent and observable models to get improved modeling power at decreased computational cost. Finally, we discuss how such statistical models of graphs can be combined with text-based information extraction methods for automatically constructing knowledge graphs from the Web. To this end, we also discuss Google's knowledge vault project as an example of such combination.",
"We propose a simple approach to combining first-order logic and probabilistic graphical models in a single representation. A Markov logic network (MLN) is a first-order knowledge base with a weight attached to each formula (or clause). Together with a set of constants representing objects in the domain, it specifies a ground Markov network containing one feature for each possible grounding of a first-order formula in the KB, with the corresponding weight. Inference in MLNs is performed by MCMC over the minimal subset of the ground network required for answering the query. Weights are efficiently learned from relational databases by iteratively optimizing a pseudo-likelihood measure. Optionally, additional clauses are learned using inductive logic programming techniques. Experiments with a real-world database and knowledge base in a university domain illustrate the promise of this approach.",
"We introduce ProbLog, a probabilistic extension of Prolog. A ProbLog program defines a distribution over logic programs by specifying for each clause the probability that it belongs to a randomly sampled program, and these probabilities are mutually independent. The semantics of ProbLog is then defined by the success probability of a query, which corresponds to the probability that the query succeeds in a randomly sampled program. The key contribution of this paper is the introduction of an effective solver for computing success probabilities. It essentially combines SLD-resolution with methods for computing the probability of Boolean formulae. Our implementation further employs an approximation algorithm that combines iterative deepening with binary decision diagrams. We report on experiments in the context of discovering links in real biological networks, a demonstration of the practical usefulness of the approach."
]
} |
1812.03213 | 2904548465 | Phrase Grounding aims to detect and localize objects in images that are referred to and are queried by natural language phrases. Phrase grounding finds applications in tasks such as Visual Dialog, Visual Search and Image-text co-reference resolution. In this paper, we present a framework that leverages information such as phrase category, relationships among neighboring phrases in a sentence and context to improve the performance of phrase grounding systems. We propose three modules: Proposal Indexing Network(PIN); Inter-phrase Regression Network(IRN) and Proposal Ranking Network(PRN) each of which analyze the region proposals of an image at increasing levels of detail by incorporating the above information. Also, in the absence of ground-truth spatial locations of the phrases(weakly-supervised), we propose knowledge transfer mechanisms that leverages the framework of PIN module. We demonstrate the effectiveness of our approach on the Flickr 30k Entities and ReferItGame datasets, for which we achieve improvements over state-of-the-art approaches in both supervised and weakly-supervised variants. | Improving hugely on early phrase grouding attempts that used limited vocabulary @cite_13 @cite_27 , @cite_14 employ bidirectional RNN to align the sentence fragments and image regions in common embedding space. @cite_21 proposed to rank proposals using knowledge transfer from image captioning. @cite_0 employ attention to rank proposals in a latent subspace. @cite_22 extended this approach to account for regression based on query semantics. @cite_30 suggest using Canonical Correlation Analysis (CCA) and @cite_34 suggest Deep CCA to learn similarity among visual and language modalities. @cite_15 employ structured matching and boost performance using partial matching of phrase pairs. Plummer et. al @cite_35 further augment CCA model to take advantage of extensive linguistic cues in the phrases. All these approaches are upperbound by the performance of the proposal generation systems. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_14",
"@cite_22",
"@cite_21",
"@cite_0",
"@cite_27",
"@cite_15",
"@cite_34",
"@cite_13"
],
"mid": [
"",
"",
"2951805548",
"2620365397",
"2963735856",
"",
"2048343491",
"2520141964",
"2287889828",
"2953276893"
],
"abstract": [
"",
"",
"We present a model that generates natural language descriptions of images and their regions. Our approach leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between language and visual data. Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Multimodal Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate that our alignment model produces state of the art results in retrieval experiments on Flickr8K, Flickr30K and MSCOCO datasets. We then show that the generated descriptions significantly outperform retrieval baselines on both full images and on a new dataset of region-level annotations.",
"Given an image and a natural language query phrase, a grounding system localizes the mentioned objects in the image according to the query's specifications. State-of-the-art methods address the problem by ranking a set of proposal bounding boxes according to the query's semantics, which makes them dependent on the performance of proposal generation systems. Besides, query phrases in one sentence may be semantically related in one sentence and can provide useful cues to ground objects. We propose a novel Multimodal Spatial Regression with semantic Context (MSRC) system which not only predicts the location of ground truth based on proposal bounding boxes, but also refines prediction results by penalizing similarities of different queries coming from same sentences. The advantages of MSRC are twofold: first, it removes the limitation of performance from proposal generation algorithms by using a spatial regression network. Second, MSRC not only encodes the semantics of a query phrase, but also deals with its relation with other queries in the same sentence (i.e., context) by a context refinement network. Experiments show MSRC system provides a significant improvement in accuracy on two popular datasets: Flickr30K Entities and Refer-it Game, with 6.64 and 5.28 increase over the state-of-the-arts respectively.",
"In this paper, we address the task of natural language object retrieval, to localize a target object within a given image based on a natural language query of the object. Natural language object retrieval differs from text-based image retrieval task as it involves spatial information about objects within the scene and global scene context. To address this issue, we propose a novel Spatial Context Recurrent ConvNet (SCRC) model as scoring function on candidate boxes for object retrieval, integrating spatial configurations and global scene-level contextual information into the network. Our model processes query text, local image descriptors, spatial configurations and global context features through a recurrent network, outputs the probability of the query text conditioned on each candidate box as a score for the box, and can transfer visual-linguistic knowledge from image captioning domain to our task. Experimental results demonstrate that our method effectively utilizes both local and global information, outperforming previous baseline methods significantly on different datasets and scenarios, and can exploit large scale vision and language datasets for knowledge transfer.",
"",
"In this paper we exploit natural sentential descriptions of RGB-D scenes in order to improve 3D semantic parsing. Importantly, in doing so, we reason about which particular object each noun pronoun is referring to in the image. This allows us to utilize visual information in order to disambiguate the so-called coreference resolution problem that arises in text. Towards this goal, we propose a structure prediction model that exploits potentials computed from text and RGB-D imagery to reason about the class of the 3D objects, the scene type, as well as to align the nouns pronouns with the referred visual objects. We demonstrate the effectiveness of our approach on the challenging NYU-RGBD v2 dataset, which we enrich with natural lingual descriptions. We show that our approach significantly improves 3D detection and scene classification accuracy, and is able to reliably estimate the text-to-image alignment. Furthermore, by using textual and visual information, we are also able to successfully deal with coreference in text, improving upon the state-of-the-art Stanford coreference system [15].",
"In this paper we introduce a new approach to phrase localization: grounding phrases in sentences to image regions. We propose a structured matching of phrases and regions that encourages the semantic relations between phrases to agree with the visual relations between regions. We formulate structured matching as a discrete optimization problem and relax it to a linear program. We use neural networks to embed regions and phrases into vectors, which then define the similarities (matching weights) between regions and phrases. We integrate structured matching with neural networks to enable end-to-end training. Experiments on Flickr30K Entities demonstrate the empirical effectiveness of our approach.",
"This paper proposes a method for learning joint embeddings of images and text using a two-branch neural network with multiple layers of linear projections followed by nonlinearities. The network is trained using a large margin objective that combines cross-view ranking constraints with within-view neighborhood structure preservation constraints inspired by metric learning literature. Extensive experiments show that our approach gains significant improvements in accuracy for image-to-text and text-to-image retrieval. Our method achieves new state-of-the-art results on the Flickr30K and MSCOCO image-sentence datasets and shows promise on the new task of phrase localization on the Flickr30K Entities dataset.",
"We introduce a model for bidirectional retrieval of images and sentences through a multi-modal embedding of visual and natural language data. Unlike previous models that directly map images or sentences into a common embedding space, our model works on a finer level and embeds fragments of images (objects) and fragments of sentences (typed dependency tree relations) into a common space. In addition to a ranking objective seen in previous work, this allows us to add a new fragment alignment objective that learns to directly associate these fragments across modalities. Extensive experimental evaluation shows that reasoning on both the global level of images and sentences and the finer level of their respective fragments significantly improves performance on image-sentence retrieval tasks. Additionally, our model provides interpretable predictions since the inferred inter-modal fragment alignment is explicit."
]
} |
1812.03213 | 2904548465 | Phrase Grounding aims to detect and localize objects in images that are referred to and are queried by natural language phrases. Phrase grounding finds applications in tasks such as Visual Dialog, Visual Search and Image-text co-reference resolution. In this paper, we present a framework that leverages information such as phrase category, relationships among neighboring phrases in a sentence and context to improve the performance of phrase grounding systems. We propose three modules: Proposal Indexing Network(PIN); Inter-phrase Regression Network(IRN) and Proposal Ranking Network(PRN) each of which analyze the region proposals of an image at increasing levels of detail by incorporating the above information. Also, in the absence of ground-truth spatial locations of the phrases(weakly-supervised), we propose knowledge transfer mechanisms that leverages the framework of PIN module. We demonstrate the effectiveness of our approach on the Flickr 30k Entities and ReferItGame datasets, for which we achieve improvements over state-of-the-art approaches in both supervised and weakly-supervised variants. | Recently, @cite_5 proposed QRC that overcomes some limitations of region proposal generators by regressing proposals based on query and employs reinforcement learning techniques to punish conflicting predictions. | {
"cite_N": [
"@cite_5"
],
"mid": [
"2745166287"
],
"abstract": [
"Given a textual description of an image, phrase grounding localizes objects in the image referred by query phrases in the description. State-of-the-art methods address the problem by ranking a set of proposals based on the relevance to each query, which are limited by the performance of independent proposal generation systems and ignore useful cues from context in the description. In this paper, we adopt a spatial regression method to break the performance limit, and introduce reinforcement learning techniques to further leverage semantic context information. We propose a novel Query-guided Regression network with Context policy (QRC Net) which jointly learns a Proposal Generation Network (PGN), a Query-guided Regression Network (QRN) and a Context Policy Network (CPN). Experiments show QRC Net provides a significant improvement in accuracy on two popular datasets: Flickr30K Entities and Referit Game, with 14.25 and 17.14 increase over the state-of-the-arts respectively."
]
} |
1812.03213 | 2904548465 | Phrase Grounding aims to detect and localize objects in images that are referred to and are queried by natural language phrases. Phrase grounding finds applications in tasks such as Visual Dialog, Visual Search and Image-text co-reference resolution. In this paper, we present a framework that leverages information such as phrase category, relationships among neighboring phrases in a sentence and context to improve the performance of phrase grounding systems. We propose three modules: Proposal Indexing Network(PIN); Inter-phrase Regression Network(IRN) and Proposal Ranking Network(PRN) each of which analyze the region proposals of an image at increasing levels of detail by incorporating the above information. Also, in the absence of ground-truth spatial locations of the phrases(weakly-supervised), we propose knowledge transfer mechanisms that leverages the framework of PIN module. We demonstrate the effectiveness of our approach on the Flickr 30k Entities and ReferItGame datasets, for which we achieve improvements over state-of-the-art approaches in both supervised and weakly-supervised variants. | Context provides broader information that can be leveraged to resolve semantic ambiguities and rank proposals. @cite_21 used global image context and bounding box encoding as context to augment visual features. @cite_9 further encode the size information and jointly predict all query regions to boost performance in a referring task. @cite_35 jointly optimize neighboring phrases by encoding their relations for better grounding performance. @cite_22 @cite_5 employ semantic context to jointly ground multiple phrases to filter conflicting predictions among neighboring phrases. The existing approaches for grounding do not take full advantage of the rich context provided by visual context, semantic context and inter-phrase relationships. | {
"cite_N": [
"@cite_35",
"@cite_22",
"@cite_9",
"@cite_21",
"@cite_5"
],
"mid": [
"",
"2620365397",
"2489434015",
"2963735856",
"2745166287"
],
"abstract": [
"",
"Given an image and a natural language query phrase, a grounding system localizes the mentioned objects in the image according to the query's specifications. State-of-the-art methods address the problem by ranking a set of proposal bounding boxes according to the query's semantics, which makes them dependent on the performance of proposal generation systems. Besides, query phrases in one sentence may be semantically related in one sentence and can provide useful cues to ground objects. We propose a novel Multimodal Spatial Regression with semantic Context (MSRC) system which not only predicts the location of ground truth based on proposal bounding boxes, but also refines prediction results by penalizing similarities of different queries coming from same sentences. The advantages of MSRC are twofold: first, it removes the limitation of performance from proposal generation algorithms by using a spatial regression network. Second, MSRC not only encodes the semantics of a query phrase, but also deals with its relation with other queries in the same sentence (i.e., context) by a context refinement network. Experiments show MSRC system provides a significant improvement in accuracy on two popular datasets: Flickr30K Entities and Refer-it Game, with 6.64 and 5.28 increase over the state-of-the-arts respectively.",
"Humans refer to objects in their environments all the time, especially in dialogue with other people. We explore generating and comprehending natural language referring expressions for objects in images. In particular, we focus on incorporating better measures of visual context into referring expression models and find that visual comparison to other objects within an image helps improve performance significantly. We also develop methods to tie the language generation process together, so that we generate expressions for all objects of a particular category jointly. Evaluation on three recent datasets - RefCOCO, RefCOCO+, and RefCOCOg (Datasets and toolbox can be downloaded from https: github.com lichengunc refer), shows the advantages of our methods for both referring expression generation and comprehension.",
"In this paper, we address the task of natural language object retrieval, to localize a target object within a given image based on a natural language query of the object. Natural language object retrieval differs from text-based image retrieval task as it involves spatial information about objects within the scene and global scene context. To address this issue, we propose a novel Spatial Context Recurrent ConvNet (SCRC) model as scoring function on candidate boxes for object retrieval, integrating spatial configurations and global scene-level contextual information into the network. Our model processes query text, local image descriptors, spatial configurations and global context features through a recurrent network, outputs the probability of the query text conditioned on each candidate box as a score for the box, and can transfer visual-linguistic knowledge from image captioning domain to our task. Experimental results demonstrate that our method effectively utilizes both local and global information, outperforming previous baseline methods significantly on different datasets and scenarios, and can exploit large scale vision and language datasets for knowledge transfer.",
"Given a textual description of an image, phrase grounding localizes objects in the image referred by query phrases in the description. State-of-the-art methods address the problem by ranking a set of proposals based on the relevance to each query, which are limited by the performance of independent proposal generation systems and ignore useful cues from context in the description. In this paper, we adopt a spatial regression method to break the performance limit, and introduce reinforcement learning techniques to further leverage semantic context information. We propose a novel Query-guided Regression network with Context policy (QRC Net) which jointly learns a Proposal Generation Network (PGN), a Query-guided Regression Network (QRN) and a Context Policy Network (CPN). Experiments show QRC Net provides a significant improvement in accuracy on two popular datasets: Flickr30K Entities and Referit Game, with 14.25 and 17.14 increase over the state-of-the-arts respectively."
]
} |
1812.03213 | 2904548465 | Phrase Grounding aims to detect and localize objects in images that are referred to and are queried by natural language phrases. Phrase grounding finds applications in tasks such as Visual Dialog, Visual Search and Image-text co-reference resolution. In this paper, we present a framework that leverages information such as phrase category, relationships among neighboring phrases in a sentence and context to improve the performance of phrase grounding systems. We propose three modules: Proposal Indexing Network(PIN); Inter-phrase Regression Network(IRN) and Proposal Ranking Network(PRN) each of which analyze the region proposals of an image at increasing levels of detail by incorporating the above information. Also, in the absence of ground-truth spatial locations of the phrases(weakly-supervised), we propose knowledge transfer mechanisms that leverages the framework of PIN module. We demonstrate the effectiveness of our approach on the Flickr 30k Entities and ReferItGame datasets, for which we achieve improvements over state-of-the-art approaches in both supervised and weakly-supervised variants. | Knowledge transfer involves solving a target task by learning from a different but related source task. @cite_11 employ knowledge transfer, @cite_25 use linguistic knowledge bases to automatically transfer information from source to target classes. @cite_29 and @cite_19 employ knowledge transfer in weakly-supervised object detection and localization respectively using skip vectors @cite_28 as a knowledge base. In this work, we propose to use knowledge transfer from object detection task to index region proposals for weakly supervised phrase grounding. | {
"cite_N": [
"@cite_28",
"@cite_29",
"@cite_19",
"@cite_25",
"@cite_11"
],
"mid": [
"2950133940",
"2061629163",
"1945443124",
"2077071968",
""
],
"abstract": [
"The recently introduced continuous Skip-gram model is an efficient method for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. In this paper we present several extensions that improve both the quality of the vectors and the training speed. By subsampling of the frequent words we obtain significant speedup and also learn more regular word representations. We also describe a simple alternative to the hierarchical softmax called negative sampling. An inherent limitation of word representations is their indifference to word order and their inability to represent idiomatic phrases. For example, the meanings of \"Canada\" and \"Air\" cannot be easily combined to obtain \"Air Canada\". Motivated by this example, we present a simple method for finding phrases in text, and show that learning good vector representations for millions of phrases is possible.",
"Learning a new object class from cluttered training images is very challenging when the location of object instances is unknown, i.e. in a weakly supervised setting. Many previous works require objects covering a large portion of the images. We present a novel approach that can cope with extensive clutter as well as large scale and appearance variations between object instances. To make this possible we exploit generic knowledge learned beforehand from images of other classes for which location annotation is available. Generic knowledge facilitates learning any new class from weakly supervised images, because it reduces the uncertainty in the location of its object instances. We propose a conditional random field that starts from generic knowledge and then progressively adapts to the new class. Our approach simultaneously localizes object instances while learning an appearance model specific for the class. We demonstrate this on several datasets, including the very challenging Pascal VOC 2007. Furthermore, our method allows training any state-of-the-art object detector in a weakly supervised fashion, although it would normally require object location annotations.",
"We consider the problem of localizing unseen objects in weakly labeled image collections. Given a set of images annotated at the image level, our goal is to localize the object in each image. The novelty of our proposed work is that, in addition to building object appearance model from the weakly labeled data, we also make use of existing detectors of some other object classes (which we call “familiar objects”). We propose a method for transferring the appearance models of the familiar objects to the unseen object. Our experimental results on both image and video datasets demonstrate the effectiveness of our approach.",
"While knowledge transfer (KT) between object classes has been accepted as a promising route towards scalable recognition, most experimental KT studies are surprisingly limited in the number of object classes considered. To support claims of KT w.r.t. scalability we thus advocate to evaluate KT in a large-scale setting. To this end, we provide an extensive evaluation of three popular approaches to KT on a recently proposed large-scale data set, the ImageNet Large Scale Visual Recognition Competition 2010 data set. In a first setting they are directly compared to one-vs-all classification often neglected in KT papers and in a second setting we evaluate their ability to enable zero-shot learning. While none of the KT methods can improve over one-vs-all classification they prove valuable for zero-shot learning, especially hierarchical and direct similarity based KT. We also propose and describe several extensions of the evaluated approaches that are necessary for this large-scale study.",
""
]
} |
1812.03085 | 2904818016 | In this paper, we formulate the color constancy task as an image-to-image translation problem using GANs. By conducting a large set of experiments on different datasets, an experimental survey is provided on the use of different types of GANs to solve for color constancy i.e. CC-GANs (Color Constancy GANs). Based on the experimental review, recommendations are given for the design of CC-GAN architectures based on different criteria, circumstances and datasets. | To mimic the human ability of color constancy for cameras, a wide range of algorithms are proposed. These algorithms can be classified into two main categories: (1) static-based methods which are either relying on reflection models or low-level image statistics, and (2) learning-based methods which rely on training sets @cite_48 . Static methods assume that an image that is taken under a canonical light source follow certain properties. For example, the grey-world algorithm @cite_34 assumes that the average reflectance of the surfaces in a scene is achromatic. Further, the white-patch algorithm @cite_5 assumes that the maximum pixel response in an image corresponds to white reflectance. These assumptions are extended to first- and higher-order image statistics to provide more accurate estimators such as the grey-edge assumption @cite_16 . In their work, @cite_16 proves that all the statistics-based methods can be unified into a single framework. | {
"cite_N": [
"@cite_48",
"@cite_5",
"@cite_34",
"@cite_16"
],
"mid": [
"2100001370",
"2076205488",
"2052090926",
""
],
"abstract": [
"Computational color constancy is a fundamental prerequisite for many computer vision applications. This paper presents a survey of many recent developments and state-of-the-art methods. Several criteria are proposed that are used to assess the approaches. A taxonomy of existing algorithms is proposed and methods are separated in three groups: static methods, gamut-based methods, and learning-based methods. Further, the experimental setup is discussed including an overview of publicly available datasets. Finally, various freely available methods, of which some are considered to be state of the art, are evaluated on two datasets.",
"Cone shell erosion between inserts is substantially reduced by positioning small, flat-topped compacts in the vulnerable cutter shell areas. At least one row of substantially outwardly projecting formation contacting inserts are located on the rock cutter. A row of substantially flush compacts are embedded in the cutter shell alternately positioned between the outwardly projecting formation contacting inserts.",
"A comprehensive mathematical model to account for colour constancy is formulated. Since the visual system is able to measure true object colour in complex scenes under a broad range of spectral compositions, for the illumination; it is assumed that the visual system must implicitly estimate and illuminant. The basic hypothesis is that the estimate of the illuminant is made on the basis of spatial information from the entire visual field. This estimate is then used by the visual system to arrive at an estimate of the (object) reflectance of the various subfields in the complex visual scene. The estimates are made by matching the inputs to the system to linear combinations of fixed bases and standards in the colour space. The model provides a general unified mathematical framework for related psychophysical phenomenology.",
""
]
} |
1812.03085 | 2904818016 | In this paper, we formulate the color constancy task as an image-to-image translation problem using GANs. By conducting a large set of experiments on different datasets, an experimental survey is provided on the use of different types of GANs to solve for color constancy i.e. CC-GANs (Color Constancy GANs). Based on the experimental review, recommendations are given for the design of CC-GAN architectures based on different criteria, circumstances and datasets. | More recent work uses powerful deep CNN models to approach the color constancy problem. @cite_26 uses an end-to-end deep model to estimate the illuminant. @cite_14 extracts patch features from a pre-trained CNN and feed them to a regression module to estimate the scene illumination. Both works provide state-of-the-art results supported by powerful CNN models. In another work, @cite_7 formulates the problem as an illumination classification problem using deep learning. @cite_24 proposes a two-branch deep CNN. One generates multiple hypotheses of the illuminant of a given patch. The other branch adaptively selects the best hypothesis. Moreover, @cite_35 estimates the local illuminant color of image patches. @cite_42 proposes a method to weight the input patches to differentiate between semantically valuable and semantically ambiguous local regions. Finally, @cite_30 exploits multiple frames over time within a deep learning framework. On the other hand, @cite_33 bypasses the illuminant color estimation step and directly learns a mapping from the input image to the white balanced image in an end-to-end fashion. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_14",
"@cite_26",
"@cite_33",
"@cite_7",
"@cite_42",
"@cite_24"
],
"mid": [
"2777263104",
"1950407018",
"2951446647",
"2306965214",
"2896707699",
"",
"",
"2519060067"
],
"abstract": [
"We introduce a novel formulation of temporal color constancy which considers multiple frames preceding the frame for which illumination is estimated. We propose an end-to-end trainable recurrent color constancy network – the RCC-Net – which exploits convolutional LSTMs and a simulated sequence to learn compositional representations in space and time. We use a standard single frame color constancy benchmark, the SFU Gray Ball Dataset, which can be adapted to a temporal setting. Extensive experiments show that the proposed method consistently outperforms single-frame state-of-the-art methods and their temporal variants.",
"In this paper, we present a three-stage method for the estimation of the color of the illuminant in RAW images. The first stage uses a convolutional neural network that has been specially designed to produce multiple local estimates of the illuminant. The second stage, given the local estimates, determines the number of illuminants in the scene. Finally, local illuminant estimates are refined by non-linear local aggregation, resulting in a global estimate in case of single illuminant. An extensive comparison with both local and global illuminant estimation methods in the state of the art, on standard data sets with single and multiple illuminants, proves the effectiveness of our method.",
"In this work we describe a Convolutional Neural Network (CNN) to accurately predict the scene illumination. Taking image patches as input, the CNN works in the spatial domain without using hand-crafted features that are employed by most previous methods. The network consists of one convolutional layer with max pooling, one fully connected layer and three output nodes. Within the network structure, feature learning and regression are integrated into one optimization process, which leads to a more effective model for estimating scene illumination. This approach achieves state-of-the-art performance on a standard dataset of RAW images. Preliminary experiments on images with spatially varying illumination demonstrate the stability of the local illuminant estimation ability of our CNN.",
"Computational color constancy aims to estimate the color of the light source. The performance of many vision tasks, such as object detection and scene understanding, may benefit from color constancy by estimating the correct object colors. Since traditional color constancy methods are based on specific assumptions, none of those methods can be used as a universal predictor. Further, shallow learning schemes are used for training-based color constancy approaches, suffering from limited learning capacity. In this paper, we propose a framework using Deep Neural Networks (DNNs) to obtain an accurate light source estimator to achieve color constancy. We formulate color constancy as a DNN-based regression approach to estimate the color of the light source. The model is trained using datasets of more than a million images. Experiments show that the proposed algorithm outperforms the state-of-the-art by 9 . Especially in cross dataset validation, reducing the median angular error by 35 . Further, in our implementation, the algorithm operates at more than @math fps during",
"Most popular color constancy algorithms assume that the light source obeys a uniform distribution across the scene. However, in the real world, the illuminations can vary a lot according to their spatial distribution. To overcome this problem, in this paper, we adopt a method based on a full end-to-end deep neural model to directly learn a mapping from the original image to the corresponding well-colored image. With this formulation, the network is able to determine pixel-wise illumination and produce a final visually compelling image. The training and evaluation of the network were performed on a standard dataset of two-dominant-illuminants. In this dataset, this approach achieves state-of-the-art performance. Besides, the main architecture of the network simply consists of a stack of fully convolutional blocks which can take the input of arbitrary size and produce correspondingly-sized output with effective learning. The experimental result shows that our customized loss function can help to reach a better performance than simply using MSE.",
"",
"",
"Illuminant estimation to achieve color constancy is an ill-posed problem. Searching the large hypothesis space for an accurate illuminant estimation is hard due to the ambiguities of unknown reflections and local patch appearances. In this work, we propose a novel Deep Specialized Network (DS-Net) that is adaptive to diverse local regions for estimating robust local illuminants. This is achieved through a new convolutional network architecture with two interacting sub-networks, i.e. an hypotheses network (HypNet) and a selection network (SelNet). In particular, HypNet generates multiple illuminant hypotheses that inherently capture different modes of illuminants with its unique two-branch structure. SelNet then adaptively picks for confident estimations from these plausible hypotheses. Extensive experiments on the two largest color constancy benchmark datasets show that the proposed ‘hypothesis selection’ approach is effective to overcome erroneous estimation. Through the synergy of HypNet and SelNet, our approach outperforms state-of-the-art methods such as [1, 2, 3]."
]
} |
1812.03216 | 2950819666 | Although deep reinforcement learning (deep RL) methods have lots of strengths that are favorable if applied to autonomous driving, real deep RL applications in autonomous driving have been slowed down by the modeling gap between the source (training) domain and the target (deployment) domain. Unlike current policy transfer approaches, which generally limit to the usage of uninterpretable neural network representations as the transferred features, we propose to transfer concrete kinematic quantities in autonomous driving. The proposed robust-control-based (RC) generic transfer architecture, which we call RL-RC, incorporates a transferable hierarchical RL trajectory planner and a robust tracking controller based on disturbance observer (DOB). The deep RL policies trained with known nominal dynamics model are transfered directly to the target domain, DOB-based robust tracking control is applied to tackle the modeling gap including the vehicle dynamics errors and the external disturbances such as side forces. We provide simulations validating the capability of the proposed method to achieve zero-shot transfer across multiple driving scenarios such as lane keeping, lane changing and obstacle avoidance. | Deep RL Policy Transfer. One category of transfer learning uses source domain randomization to embed robustness. The EPOpt algorithm @cite_18 randomly samples dynamic parameters from a prior distribution and optimizes the policy for cases with worst-performing dynamic parameters. @cite_14 uses RNN to take in historical paths, expecting the policy to be adaptive by implicitly identifying the parameters. Similar idea has been adopted in @cite_4 , but an on-line system identification module is explicitly introduced to identify the parameters. The estimated parameters are fed into a parameter-universal policy. All these works are limited in that they only handle the discrepancy of dynamic parameters, while robustness against external disturbances is excluded from consideration. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4"
],
"mid": [
"2529477964",
"2767050701",
"2586431331"
],
"abstract": [
"Sample complexity and safety are major challenges when learning policies with reinforcement learning for real-world tasks, especially when the policies are represented using rich function approximators like deep neural networks. Model-based methods where the real-world target domain is approximated using a simulated source domain provide an avenue to tackle the above challenges by augmenting real data with simulated data. However, discrepancies between the simulated source domain and the target domain pose a challenge for simulated training. We introduce the EPOpt algorithm, which uses an ensemble of simulated source domains and a form of adversarial training to learn policies that are robust and generalize to a broad range of possible target domains, including unmodeled effects. Further, the probability distribution over source domains in the ensemble can be adapted using data from target domain and approximate Bayesian methods, to progressively make it a better approximation. Thus, learning on a model ensemble, along with source domain adaptation, provides the benefit of both robustness and learning adaptation.",
"Simulations are attractive environments for training agents as they provide an abundant source of data and alleviate certain safety concerns during the training process. But the behaviours developed by agents in simulation are often specific to the characteristics of the simulator. Due to modeling error, strategies that are successful in simulation may not transfer to their real world counterparts. In this paper, we demonstrate a simple method to bridge this \"reality gap.\" By randomizing the dynamics of the simulator during training, we are able to develop policies that are capable of adapting to very different dynamics, including ones that differ significantly from the dynamics on which the policies were trained. This adaptivity enables the policies to generalize to the dynamics of the real world without any training on the physical system. Our approach is demonstrated on an object pushing task using a robotic arm. Despite being trained exclusively in simulation, our policies are able to maintain a similar level of performance when deployed on a real robot, reliably moving an object to a desired location from random initial configurations. We explore the impact of various design decisions and show that the resulting policies are robust to significant calibration error.",
"We present a new method of learning control policies that successfully operate under unknown dynamic models. We create such policies by leveraging a large number of training examples that are generated using a physical simulator. Our system is made of two components: a Universal Policy (UP) and a function for Online System Identification (OSI). We describe our control policy as universal because it is trained over a wide array of dynamic models. These variations in the dynamic model may include differences in mass and inertia of the robots' components, variable friction coefficients, or unknown mass of an object to be manipulated. By training the Universal Policy with this variation, the control policy is prepared for a wider array of possible conditions when executed in an unknown environment. The second part of our system uses the recent state and action history of the system to predict the dynamics model parameters mu. The value of mu from the Online System Identification is then provided as input to the control policy (along with the system state). Together, UP-OSI is a robust control policy that can be used across a wide range of dynamic models, and that is also responsive to sudden changes in the environment. We have evaluated the performance of this system on a variety of tasks, including the problem of cart-pole swing-up, the double inverted pendulum, locomotion of a hopper, and block-throwing of a manipulator. UP-OSI is effective at these tasks across a wide range of dynamic models. Moreover, when tested with dynamic models outside of the training range, UP-OSI outperforms the Universal Policy alone, even when UP is given the actual value of the model dynamics. In addition to the benefits of creating more robust controllers, UP-OSI also holds out promise of narrowing the Reality Gap between simulated and real physical systems."
]
} |
1812.03216 | 2950819666 | Although deep reinforcement learning (deep RL) methods have lots of strengths that are favorable if applied to autonomous driving, real deep RL applications in autonomous driving have been slowed down by the modeling gap between the source (training) domain and the target (deployment) domain. Unlike current policy transfer approaches, which generally limit to the usage of uninterpretable neural network representations as the transferred features, we propose to transfer concrete kinematic quantities in autonomous driving. The proposed robust-control-based (RC) generic transfer architecture, which we call RL-RC, incorporates a transferable hierarchical RL trajectory planner and a robust tracking controller based on disturbance observer (DOB). The deep RL policies trained with known nominal dynamics model are transfered directly to the target domain, DOB-based robust tracking control is applied to tackle the modeling gap including the vehicle dynamics errors and the external disturbances such as side forces. We provide simulations validating the capability of the proposed method to achieve zero-shot transfer across multiple driving scenarios such as lane keeping, lane changing and obstacle avoidance. | Another line of work is through model adaptation. @cite_16 trains a neural network for dynamics model and adapts its local linear model on-line for model-based control methods. In @cite_2 , a deep inverse dynamics model of the target system is trained, and transfer is achieved by executing inverse model outputs to reach the nominal state generated by performing source policy in source environment. One problem of such approaches is that, unlike tracking by feedback control, a feasible and accurate inverse model for feedforward tracking is hardly guaranteed in the target domain. Also, a neural network approximating system dynamics is non-trivial to find in practice. Furthermore, fine-tuning in the target domain is not desirable due to safety. | {
"cite_N": [
"@cite_16",
"@cite_2"
],
"mid": [
"2198225532",
"2530944449"
],
"abstract": [
"One of the key challenges in applying reinforcement learning to complex robotic control tasks is the need to gather large amounts of experience in order to find an effective policy for the task at hand. Model-based reinforcement learning can achieve good sample efficiency, but requires the ability to learn a model of the dynamics that is good enough to learn an effective policy. In this work, we develop a model-based reinforcement learning algorithm that combines prior knowledge from previous tasks with online adaptation of the dynamics model. These two ingredients enable highly sample-efficient learning even in regimes where estimating the true dynamics is very difficult, since the online model adaptation allows the method to locally compensate for unmodeled variation in the dynamics. We encode the prior experience into a neural network dynamics model, adapt it online by progressively refitting a local linear model of the dynamics, and use model predictive control to plan under these dynamics. Our experimental results show that this approach can be used to solve a variety of complex robotic manipulation tasks in just a single attempt, using prior data from other manipulation behaviors.",
"Developing control policies in simulation is often more practical and safer than directly running experiments in the real world. This applies to policies obtained from planning and optimization, and even more so to policies obtained from reinforcement learning, which is often very data demanding. However, a policy that succeeds in simulation often doesn't work when deployed on a real robot. Nevertheless, often the overall gist of what the policy does in simulation remains valid in the real world. In this paper we investigate such settings, where the sequence of states traversed in simulation remains reasonable for the real world, even if the details of the controls are not, as could be the case when the key differences lie in detailed friction, contact, mass and geometry properties. During execution, at each time step our approach computes what the simulation-based control policy would do, but then, rather than executing these controls on the real robot, our approach computes what the simulation expects the resulting next state(s) will be, and then relies on a learned deep inverse dynamics model to decide which real-world action is most suitable to achieve those next states. Deep models are only as good as their training data, and we also propose an approach for data collection to (incrementally) learn the deep inverse dynamics model. Our experiments shows our approach compares favorably with various baselines that have been developed for dealing with simulation to real world model discrepancy, including output error control and Gaussian dynamics adaptation."
]
} |
1812.03216 | 2950819666 | Although deep reinforcement learning (deep RL) methods have lots of strengths that are favorable if applied to autonomous driving, real deep RL applications in autonomous driving have been slowed down by the modeling gap between the source (training) domain and the target (deployment) domain. Unlike current policy transfer approaches, which generally limit to the usage of uninterpretable neural network representations as the transferred features, we propose to transfer concrete kinematic quantities in autonomous driving. The proposed robust-control-based (RC) generic transfer architecture, which we call RL-RC, incorporates a transferable hierarchical RL trajectory planner and a robust tracking controller based on disturbance observer (DOB). The deep RL policies trained with known nominal dynamics model are transfered directly to the target domain, DOB-based robust tracking control is applied to tackle the modeling gap including the vehicle dynamics errors and the external disturbances such as side forces. We provide simulations validating the capability of the proposed method to achieve zero-shot transfer across multiple driving scenarios such as lane keeping, lane changing and obstacle avoidance. | A method developed on a similar idea to the one in this paper is in @cite_1 , in which a MPC controller is designed to stabilize the target system around the nominal trajectory generated by consecutively applying the policy in the source system. Theorems on tube-based MPC ensure that the states are bounded under certain modeling error. However, the bound cannot be explicitly found and no asymptotic stability can be guaranteed. Moreover, solving on-line optimization problem is computationally expensive, while robust controllers are usually easier and faster for the trajectory tracking problem of automated vehicles. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2739211902"
],
"abstract": [
"Model-free policy learning has enabled robust performance of complex tasks with relatively simple algorithms. However, this simplicity comes at the cost of requiring an Oracle and arguably very poor sample complexity. This renders such methods unsuitable for physical systems. Variants of model-based methods address this problem through the use of simulators, however, this gives rise to the problem of policy transfer from simulated to the physical system. Model mismatch due to systematic parameter shift and unmodelled dynamics error may cause sub-optimal or unsafe behavior upon direct transfer. We introduce the Adaptive Policy Transfer for Stochastic Dynamics (ADAPT) algorithm that achieves provably safe and robust, dynamically-feasible zero-shot transfer of RL-policies to new domains with dynamics error. ADAPT combines the strengths of offline policy learning in a black-box source simulator with online tube-based MPC to attenuate bounded model mismatch between the source and target dynamics. ADAPT allows online transfer of policy, trained solely in a simulation offline, to a family of unknown targets without fine-tuning. We also formally show that (i) ADAPT guarantees state and control safety through state-action tubes under the assumption of Lipschitz continuity of the divergence in dynamics and, (ii) ADAPT results in a bounded loss of reward accumulation relative to a policy trained and evaluated in the source environment. We evaluate ADAPT on 2 continuous, non-holonomic simulated dynamical systems with 4 different disturbance models, and find that ADAPT performs between 50 -300 better on mean reward accrual than direct policy transfer."
]
} |
1812.03216 | 2950819666 | Although deep reinforcement learning (deep RL) methods have lots of strengths that are favorable if applied to autonomous driving, real deep RL applications in autonomous driving have been slowed down by the modeling gap between the source (training) domain and the target (deployment) domain. Unlike current policy transfer approaches, which generally limit to the usage of uninterpretable neural network representations as the transferred features, we propose to transfer concrete kinematic quantities in autonomous driving. The proposed robust-control-based (RC) generic transfer architecture, which we call RL-RC, incorporates a transferable hierarchical RL trajectory planner and a robust tracking controller based on disturbance observer (DOB). The deep RL policies trained with known nominal dynamics model are transfered directly to the target domain, DOB-based robust tracking control is applied to tackle the modeling gap including the vehicle dynamics errors and the external disturbances such as side forces. We provide simulations validating the capability of the proposed method to achieve zero-shot transfer across multiple driving scenarios such as lane keeping, lane changing and obstacle avoidance. | DOB for Vehicle Lateral Control. Vehicle lateral control for trajectory tracking has been well-established with many existing methods @cite_15 . DOB is a robust control technique to reject disturbances with guaranteed robust stability for linear system @cite_0 . Moreover, given a stabilizing nominal controller, Q-filter in DOB can be an arbitrary stable filter to make the sensitivity function shaped as desired @cite_12 . It has been applied to robust lateral trajectory tracking and its effectiveness has been verified by experiments @cite_5 . | {
"cite_N": [
"@cite_0",
"@cite_5",
"@cite_15",
"@cite_12"
],
"mid": [
"",
"2025420234",
"2343568200",
"1990083550"
],
"abstract": [
"",
"In this contribution a structure for high level lateral vehicle tracking control is presented. It is based on the two degrees of freedom structure that allows to separately define the command response and disturbance attenuation. The application of the disturbance observer guarantees robust compensation of the disturbances. An advantage of the presented structure is its robustness against variable vehicle parameters. Only a considerably reduced model is necessary which significantly simplifies the application process. Moreover the presented approach is characterized by its extensibility and its modularity.",
"Self-driving vehicles are a maturing technology with the potential to reshape mobility by enhancing the safety, accessibility, efficiency, and convenience of automotive transportation. Safety-critical tasks that must be executed by a self-driving vehicle include planning of motions through a dynamic environment shared with other vehicles and pedestrians, and their robust executions via feedback control. The objective of this paper is to survey the current state of the art on planning and control algorithms with particular regard to the urban setting. A selection of proposed techniques is reviewed along with a discussion of their effectiveness. The surveyed approaches differ in the vehicle mobility model used, in assumptions on the structure of the environment, and in computational requirements. The side by side comparison presented in this survey helps to gain insight into the strengths and limitations of the reviewed approaches and assists with system level design choices.",
"We have many servo systems that require nano micro level positioning accuracy. This requirement sets a number of interesting challenges from the viewpoint of sensing, actuation, and control algorithms. This paper considers the control aspect for precision positioning. In motion control systems at nano micro levels, it has been noted that the control algorithm should be customized as much as possible to the spectrum of disturbances. We will examine how prior knowledge about the disturbance spectrum should be utilized in the design of control algorithms and what are advantages of such prior knowledge. The algorithms will be evaluated on a simulated hard disk drive (HDD) benchmark problem and a laboratory setup for a wafer scanner that is equipped with a laser interferometer for position measurement."
]
} |
1812.03282 | 2904432539 | Most of current person re-identification (ReID) methods neglect a spatial-temporal constraint. Given a query image, conventional methods compute the feature distances between the query image and all the gallery images and return a similarity ranked table. When the gallery database is very large in practice, these approaches fail to obtain a good performance due to appearance ambiguity across different camera views. In this paper, we propose a novel two-stream spatial-temporal person ReID (st-ReID) framework that mines both visual semantic information and spatial-temporal information. To this end, a joint similarity metric with Logistic Smoothing (LS) is introduced to integrate two kinds of heterogeneous information into a unified framework. To approximate a complex spatial-temporal probability distribution, we develop a fast Histogram-Parzen (HP) method. With the help of the spatial-temporal constraint, the st-ReID model eliminates lots of irrelevant images and thus narrows the gallery database. Without bells and whistles, our st-ReID method achieves rank-1 accuracy of 98.1 on Market-1501 and 94.4 on DukeMTMC-reID, improving from the baselines 91.2 and 83.8 , respectively, outperforming all previous state-of-the-art methods by a large margin. | Recent person ReID methods concentrate on deep learning for visual feature representation. Basically, these deep models either attempt to design effective convolutional neural networks or adopt different kinds of loss functions, e.g., classification loss @cite_36 @cite_7 @cite_30 , verification loss @cite_4 @cite_34 , and triplet loss @cite_18 @cite_33 @cite_26 @cite_2 . Due to the remarkable ability of CNN representation, state-of-the-art approaches achieve a good performance, e.g., rank-1 accuracy of 80-90 In order to achieve this goal, many studies try to exploit person structure information @cite_1 @cite_11 @cite_24 @cite_31 @cite_23 @cite_10 . For example, a multi-scale context-aware network @cite_1 is used to learn powerful features over full body and body parts to capture the local context information. A pose-driven deep convolutional model @cite_24 is introduced to alleviate the pose variations and learn robust feature representations from both the global images and different local parts. A human parsing method @cite_10 is adopted to improve the performance of person ReID with the help of the pixel-level accuracy. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_26",
"@cite_4",
"@cite_33",
"@cite_7",
"@cite_36",
"@cite_1",
"@cite_24",
"@cite_23",
"@cite_2",
"@cite_31",
"@cite_34",
"@cite_10",
"@cite_11"
],
"mid": [
"2900373870",
"1971955426",
"2598634450",
"",
"",
"",
"",
"2964163358",
"",
"",
"2336626189",
"",
"2253171278",
"",
""
],
"abstract": [
"Cross-domain transfer learning (CDTL) is an extremely challenging task for the person re-identification (ReID). Given a source domain with annotations and a target domain without annotations, CDTL seeks an effective method to transfer the knowledge from the source domain to the target domain. However, such a simple two-domain transfer learning method is unavailable for the person ReID in that the source target domain consists of several sub-domains, e.g., camera-based sub-domains. To address this intractable problem, we propose a novel Many-to-Many Generative Adversarial Transfer Learning method (M2M-GAN) that takes multiple source sub-domains and multiple target sub-domains into consideration and performs each sub-domain transferring mapping from the source domain to the target domain in a unified optimization process. The proposed method first translates the image styles of source sub-domains into that of target sub-domains, and then performs the supervised learning by using the transferred images and the corresponding annotations in source domain. As the gap is reduced, M2M-GAN achieves a promising result for the cross-domain person ReID. Experimental results on three benchmark datasets Market-1501, DukeMTMC-reID and MSMT17 show the effectiveness of our M2M-GAN.",
"Identifying the same individual across different scenes is an important yet difficult task in intelligent video surveillance. Its main difficulty lies in how to preserve similarity of the same person against large appearance and structure variation while discriminating different individuals. In this paper, we present a scalable distance driven feature learning framework based on the deep neural network for person re-identification, and demonstrate its effectiveness to handle the existing challenges. Specifically, given the training images with the class labels (person IDs), we first produce a large number of triplet units, each of which contains three images, i.e. one person with a matched reference and a mismatched reference. Treating the units as the input, we build the convolutional neural network to generate the layered representations, and follow with the L 2 distance metric. By means of parameter optimization, our framework tends to maximize the relative distance between the matched pair and the mismatched pair for each triplet unit. Moreover, a nontrivial issue arising with the framework is that the triplet organization cubically enlarges the number of training triplets, as one image can be involved into several triplet units. To overcome this problem, we develop an effective triplet generation scheme and an optimized gradient descent algorithm, making the computational load mainly depend on the number of original images instead of the number of triplets. On several challenging databases, our approach achieves very promising results and outperforms other state-of-the-art approaches. HighlightsWe present a novel feature learning framework for person re-identification.Our framework is based on the maximum relative distance comparison.The learning algorithm is scalable to process large amount of data.We demonstrate superior performances over other state-of-the-arts.",
"In the past few years, the field of computer vision has gone through a revolution fueled mainly by the advent of large datasets and the adoption of deep convolutional neural networks for end-to-end learning. The person re-identification subfield is no exception to this. Unfortunately, a prevailing belief in the community seems to be that the triplet loss is inferior to using surrogate losses (classification, verification) followed by a separate metric learning step. We show that, for models trained from scratch as well as pretrained ones, using a variant of the triplet loss to perform end-to-end deep metric learning outperforms most other published methods by a large margin.",
"",
"",
"",
"",
"Person Re-identification (ReID) is to identify the same person across different cameras. It is a challenging task due to the large variations in person pose, occlusion, background clutter, etc. How to extract powerful features is a fundamental problem in ReID and is still an open problem today. In this paper, we design a Multi-Scale Context-Aware Network (MSCAN) to learn powerful features over full body and body parts, which can well capture the local context knowledge by stacking multi-scale convolutions in each layer. Moreover, instead of using predefined rigid parts, we propose to learn and localize deformable pedestrian parts using Spatial Transformer Networks (STN) with novel spatial constraints. The learned body parts can release some difficulties, e.g. pose variations and background clutters, in part-based representation. Finally, we integrate the representation learning processes of full body and body parts into a unified framework for person ReID through multi-class person identification tasks. Extensive evaluations on current challenging large-scale person ReID datasets, including the image-based Market1501, CUHK03 and sequence-based MARS datasets, show that the proposed method achieves the state-of-the-art results.",
"",
"",
"The past decade has witnessed the rapid development of feature representation learning and distance metric learning, whereas the two steps are often discussed separately. To explore their interaction, this work proposes an end-to-end learning framework called DARI, i.e. Distance metric And Representation Integration, and validates the effectiveness of DARI in the challenging task of person verification. Given the training images annotated with the labels, we first produce a large number of triplet units, and each one contains three images, i.e. one person and the matched mismatch references. For each triplet unit, the distance disparity between the matched pair and the mismatched pair tends to be maximized. We solve this objective by building a deep architecture of convolutional neural networks. In particular, the Mahalanobis distance matrix is naturally factorized as one top fully-connected layer that is seamlessly integrated with other bottom layers representing the image feature. The image feature and the distance metric can be thus simultaneously optimized via the one-shot backward propagation. On several public datasets, DARI shows very promising performance on re-identifying individuals cross cameras against various challenges, and outperforms other state-of-the-art approaches.",
"",
"This paper proposes a novel approach to person re-identification, a fundamental task in distributed multi-camera surveillance systems. Although a variety of powerful algorithms have been presented in the past few years, most of them usually focus on designing hand-crafted features and learning metrics either individually or sequentially. Different from previous works, we formulate a unified deep ranking framework that jointly tackles both of these key components to maximize their strengths. We start from the principle that the correct match of the probe image should be positioned in the top rank within the whole gallery set. An effective learning-to-rank algorithm is proposed to minimize the cost corresponding to the ranking disorders of the gallery. The ranking model is solved with a deep convolutional neural network (CNN) that builds the relation between input image pairs and their similarity scores through joint representation learning directly from raw image pixels. The proposed framework allows us to get rid of feature engineering and does not rely on any assumption. An extensive comparative evaluation is given, demonstrating that our approach significantly outperforms all the state-of-the-art approaches, including both traditional and CNN-based methods on the challenging VIPeR, CUHK-01, and CAVIAR4REID datasets. In addition, our approach has better ability to generalize across datasets without fine-tuning.",
"",
""
]
} |
1812.03258 | 2952002940 | Popularity is a critical success factor for a politician and her his party to win in elections and implement their plans. Finding the reasons behind the popularity can provide a stable political movement. This research attempts to measure popularity in Twitter using a mixed method. In recent years, Twitter data has provided an excellent opportunity for exploring public opinions by analyzing a large number of tweets. This study has collected and examined 4.5 million tweets related to a US politician, Senator Bernie Sanders. This study investigated eight economic reasons behind the senator's popularity in Twitter. This research has benefits for politicians, informatics experts, and policymakers to explore public opinion. The collected data will also be available for further investigation. | The fast growth of Twitter and its large-scale public available have drawn the attention of researchers for political applications of Twitter data in three directions: (1) social movement analysis, (2) election prediction, and (3)election analysis. Two examples of the first direction are exploring the role of social media in organizing protesters @cite_10 @cite_4 @cite_16 and studying the behavior of protesters in social media and its effect on social movements @cite_36 @cite_59 . The second direction has adopted quantitative methods to determine the popularity of candidates @cite_38 @cite_33 @cite_44 and find the most popular candidate and predict the elections @cite_14 @cite_42 . The third research category attempts to investigate an election at a macro level such as studying the social media strategy @cite_56 or analyzing economic factors @cite_25 . | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_4",
"@cite_33",
"@cite_36",
"@cite_42",
"@cite_56",
"@cite_44",
"@cite_59",
"@cite_16",
"@cite_10",
"@cite_25"
],
"mid": [
"1590495275",
"",
"2328027452",
"2216627755",
"2095739955",
"2127925090",
"1986316755",
"1991015079",
"2322636197",
"2021905382",
"2123835189",
""
],
"abstract": [
"Twitter is a microblogging website where users read and write millions of short messages on a variety of topics every day. This study uses the context of the German federal election to investigate whether Twitter is used as a forum for political deliberation and whether online messages on Twitter validly mirror offline political sentiment. Using LIWC text analysis software, we conducted a content-analysis of over 100,000 messages containing a reference to either a political party or a politician. Our results show that Twitter is indeed used extensively for political deliberation. We find that the mere number of messages mentioning a party reflects the election result. Moreover, joint mentions of two parties are in line with real world political ties and coalitions. An analysis of the tweets’ political sentiment demonstrates close correspondence to the parties' and politicians’ political positions indicating that the content of Twitter messages plausibly reflects the offline political landscape. We discuss the use of microblogging message content as a valid indicator of political sentiment and derive suggestions for further research.",
"",
"The “Arab Spring” has proven astonishing and exhilarating to Middle East analysts and activists alike. Starting in Tunisia and spreading quickly to Egypt, Bahrain, Yemen, Libya, Syria and beyond, a wave of political protest, unprecedented in scope and ambition, swept the region in 2011. In short order, two deeply entrenched authoritarian rulers were jettisoned from office, and by early summer the leaders of at least three other Arab regimes appeared to be in grave jeopardy. In the wake of this wave, nearly every authoritarian regime in the region scrambled to concoct the “right” mix of repression and cooptation in the hope of stemming the protest. And even authoritarian regimes as distant as China took nervous notice of developments in the region. For Middle East specialists, the events of the Arab Spring proved especially jarring, even if welcomed, because of their extensive investment in analyzing the underpinnings of authoritarian persistence, long the region’s political hallmark. The empirical surprise of 2011 raises a pressing question—do we need to rethink the logic of authoritarianism in the Arab world or, even more broadly, authoritarian persistence writ large? What follows is a reconsideration of the “robustness of authoritarianism” in the Arab world and beyond. The surprises of the Arab Spring, and especially the internal variation within the region, suggest new theoretical insights as well as new empirical realities that govern the dynamics of authoritarianism in the twenty-first century. At the same time, recent events confirm some long-held truisms about the dynamics of authoritarian durability. In addition, the events of the Arab Spring suggest insights into a host of other issues, including the dynamics of military defection; the logic of social mobilization; the complementary roles of structure, agency, intention, and contingency in complex political phenomena such as political uprisings (and, consequently, the limits",
"Nowadays, the use of social media such as Twitter is necessary to monitor trends of people on political issues. As a case study, we collected the main stream of Twitter related to the 2010 UK general election during the associated period. We analyse the characteristics of the three main parties in the election. Also, we propose a simple and practical algorithm to identify the political leaning of users using the amount of Twitter messages which seem related to political parties. The experimental results showed that the best-performing classification method -- which uses the number of Twitter messages referring to a particular political party -- achieved about 86 classification accuracy without any training phase.",
"Protest activity has become a central means for political change in Chile. We examine the association between social media use and youth protest, as well as mediating and moderating mechanisms of this relationship, using survey data collected in Chile in 2010. We found that Facebook use was associated significantly with protest activity, even after taking into account political grievances, material and psychological resources, values, and news media use. The link between overall Facebook use and protest activity was explained by using the social network for news and socializing rather than when it was used for self-expression. Postmaterialist values and political ideologies were not found to moderate the association between Facebook use and protest.",
"To what extend can one use Twitter in opinion polls for political elections? Merely counting Twitter messages mentioning political party names is no guarantee for obtaining good election predictions. By improving the quality of the document collection and by performing sentiment analysis, predictions based on entity counts in tweets can be considerably improved, and become nearly as good as traditionally obtained opinion polls.",
"Abstract Considering the integral relationship between public relations and democracy ( Martinelli, 2011 ) coupled with the growing use of social media for democratic aims ( Smith, 2011 ) the current study examines the effectiveness of Twitter as a public relations communications tool for congressional campaigns. Specifically, as a means of testing Twitter's effectiveness in informing and engaging voters, congressional candidate and political party Twitter use for all 435 U.S. House of Representatives races ( N = 1284) are compared with 2010 election outcomes. Results indicate that candidates’ Twitter use significantly increased their odds of winning, controlling for incumbency and Party ID. Additionally, significant differences between incumbents’ and challengers’ Twitter use during the election cycle emerged, which has important implications for public relations practices aimed at achieving democratic outcomes.",
"In recent years, Twitter has become one of the most important modes for social networking and disseminating content on a variety of topics. It has developed into a popular medium for political discourse and social organization during elections. There has been growing body of literature demonstrating the ability to predict the outcome of elections from Twitter data. This works aims to test the predictive power of Twitter in inferring the winning candidate and vote percentages of the candidates in an election. Our prediction is based on the number of times the name of a candidate is mentioned in tweets prior to elections. We develop new methods to augment the counts by counting not only the presence of candidate's official names but also their aliases and commonly appearing names. In addition, we devised a technique to include relevant and filter irrelevant tweets based on predefined set of keywords. Our approach is successful in predicting the winner of all three presidential elections held in Latin America during the months of February through April, 2013.",
"Recent studies have shown a positive link between frequency of social media use and political participation. However, there has been no clear elaboration of how using social media translates into increased political activity. The current study examines three explanations for this relationship in the context of citizens’ protest behavior: information (social media as a source for news), opinion expression (using social media to express political opinions), and activism (joining causes and finding mobilizing information through social media). To test these relationships, the study uses survey data collected in Chile in 2011, amid massive demonstrations demanding wholesale changes in education and energy policy. Findings suggest that using social media for opinion expression and activism mediates the relationship between overall social media use and protest behavior. These findings deepen our knowledge of the uses and effects of social media and provide new evidence on the role of digital platforms as facili...",
"As thousands of demonstrators took to the streets of Ferguson, Missouri, to protest the fatal police shooting of unarmed African American teenager Michael Brown in the summer of 2014, news and commentary on the shooting, the protests, and the militarized response that followed circulated widely through social media networks. Through a theorization of hashtag usage, we discuss how and why social media platforms have become powerful sites for documenting and challenging episodes of police brutality and the misrepresentation of racialized bodies in mainstream media. We show how engaging in “hashtag activism” can forge a shared political temporality, and, additionally, we examine how social media platforms can provide strategic outlets for contesting and reimagining the materiality of racialized bodies. Our analysis combines approaches from linguistic anthropology and social movements research to investigate the semiotics of digital protest and to interrogate both the possibilities and the pitfalls of engaging in “hashtag ethnography.”",
"Based on a survey of participants in Egypt’s Tahrir Square protests, we demonstrate that social media in general, and Facebook in particular, provided new sources of information the regime could not easily control and were crucial in shaping how citizens made individual decisions about participating in protests, the logistics of protest, and the likelihood of success. We demonstrate that people learned about the protests primarily through interpersonal communication using Facebook, phone contact, or face-to-face conversation. Controlling for other factors, social media use greatly increased the odds that a respondent attended protests on the first day. Half of those surveyed produced and disseminated visuals from the demonstrations, mainly through Facebook.",
""
]
} |
1812.03128 | 2905379489 | We present a new type of backdoor attack that exploits a vulnerability of convolutional neural networks (CNNs) that has been previously unstudied. In particular, we examine the application of facial recognition. Deep learning techniques are at the top of the game for facial recognition, which means they have now been implemented in many production-level systems. Alarmingly, unlike other commercial technologies such as operating systems and network devices, deep learning-based facial recognition algorithms are not presently designed with security requirements or audited for security vulnerabilities before deployment. Given how young the technology is and how abstract many of the internal workings of these algorithms are, neural network-based facial recognition systems are prime targets for security breaches. As more and more of our personal information begins to be guarded by facial recognition (e.g., the iPhone X), exploring the security vulnerabilities of these systems from a penetration testing standpoint is crucial. Along these lines, we describe a general methodology for backdooring CNNs via targeted weight perturbations. Using a five-layer CNN and ResNet-50 as case studies, we show that an attacker is able to significantly increase the chance that inputs they supply will be falsely accepted by a CNN while simultaneously preserving the error rates for legitimate enrolled classes. | More modern attacks implement strategies such as dumping the original operating system on a virtual machine, making it a guest OS and making the rootkit the host OS @cite_1 . The only way to detect that this has happened is to look for discrepancies between how physical memory and virtual memory operate. Hardware Performance Counters, which are specific registers in microprocessors that monitor performance, are the newest and most effective way of detecting modern rootkits @cite_8 , but there still exists malware that can evade detection. The ability to evade detection is another property of our proposed backdoor attack. | {
"cite_N": [
"@cite_1",
"@cite_8"
],
"mid": [
"2117882778",
"2602229646"
],
"abstract": [
"Attackers and defenders of computer systems both strive to gain complete control over the system. To maximize their control, both attackers and defenders have migrated to low-level, operating system code. In this paper, we assume the perspective of the attacker, who is trying to run malicious software and avoid detection. By assuming this perspective, we hope to help defenders understand and defend against the threat posed by a new class of rootkits. We evaluate a new type of malicious software that gains qualitatively more control over a system. This new type of malware, which we call a virtual-machine based rootkit (VMBR), installs a virtual-machine monitor underneath an existing operating system and hoists the original operating system into a virtual machine. Virtual-machine based rootkits are hard to detect and remove because their state cannot be accessed by software running in the target system. Further, VMBRs support general-purpose malicious services by allowing such services to run in a separate operating system that is protected from the target system. We evaluate this new threat by implementing two proof-of-concept VMBRs. We use our proof-of-concept VMBRs to subvert Windows XP and Linux target systems, and we implement four example malicious services using the VMBR platform. Last, we use what we learn from our proof-of-concept VMBRs to explore ways to defend against this new threat. We discuss possible ways to detect and prevent VMBRs, and we implement a defense strategy suitable for protecting systems against this threat.",
"Recent work has investigated the use of hardware performance counters (HPCs) for the detection of malware running on a system. These works gather traces of HPCs for a variety of applications (both malicious and non-malicious) and then apply machine learning to train a detector to distinguish between benign applications and malware. In this work, we provide a more comprehensive analysis of the applicability of using machine learning and HPCs for a specific subset of malware: kernel rootkits. We design five synthetic rootkits, each providing a single piece of rootkit functionality, and execute each while collecting HPC traces of its impact on a specific benchmark application. We then apply machine learning feature selection techniques in order to determine the most relevant HPCs for the detection of these rootkits. We identify 16 HPCs that are useful for the detection of hooking based roots, and also find that rootkits employing direct kernel object manipulation (DKOM) do not significantly impact HPCs. We then use these synthetic rootkit traces to train a detection system capable of detecting new rootkits it has not seen previously with an accuracy of over 99 . Our results indicate that HPCs have the potential to be an effective tool for rootkit detection, even against new rootkits not previously seen by the detector."
]
} |
1812.03264 | 2964137521 | Chinese traditional painting is one of the most historical artworks in the world. It is very popular in Eastern and Southeast Asia due to being aesthetically appealing. Compared with western artistic painting, it is usually more visually abstract and textureless. Recently, neural network based style transfer methods have shown promising and appealing results which are mainly focused on western painting. It remains a challenging problem to preserve abstraction in neural style transfer. In this paper, we present a Neural Abstract Style Transfer method for Chinese traditional painting. It learns to preserve abstraction and other style jointly end-to-end via a novel MXDoG-guided filter (Modified version of the eXtended Difference-of-Gaussians) and three fully differentiable loss terms. To the best of our knowledge, there is little work study on neural style transfer of Chinese traditional painting. To promote research on this direction, we collect a new dataset with diverse photo-realistic images and Chinese traditional paintings (The dataset will be released at https: github.com lbsswu Chinese_style_transfer.). In experiments, the proposed method shows more appealing stylized results in transferring the style of Chinese traditional painting than state-of-the-art neural style transfer methods. | Above neural style transfer methods utilize optimization for image style transfer, usually, it takes more than @math seconds to process an image. @cite_18 utilize the perceptual loss to train feed-forward neural networks, which can be running in real time on GPU. Almost at the same time, @cite_26 propose an unsupervised real time method, but a multi-scale neural network is used. Li and Wand @cite_37 also propose a feed-forward method to accelerate their patch-based Markov method @cite_10 . Recently, @cite_14 further propose an instance normalization method which significantly improves the quality of fast neural style transfer. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_26",
"@cite_37",
"@cite_10"
],
"mid": [
"2331128040",
"2502312327",
"2952226636",
"2339754110",
"2275363859"
],
"abstract": [
"We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a per-pixel loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing perceptual loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results.",
"It this paper we revisit the fast stylization method introduced in Ulyanov et. al. (2016). We show how a small change in the stylization architecture results in a significant qualitative improvement in the generated images. The change is limited to swapping batch normalization with instance normalization, and to apply the latter both at training and testing times. The resulting method can be used to train high-performance architectures for real-time image generation. The code will is made available on github at this https URL. Full paper can be found at arXiv:1701.02096.",
"recently demonstrated that deep networks can generate beautiful textures and stylized images from a single texture example. However, their methods requires a slow and memory-consuming optimization process. We propose here an alternative approach that moves the computational burden to a learning stage. Given a single example of a texture, our approach trains compact feed-forward convolutional networks to generate multiple samples of the same texture of arbitrary size and to transfer artistic style from a given image to any other image. The resulting networks are remarkably light-weight and can generate textures of quality comparable to Gatys et al., but hundreds of times faster. More generally, our approach highlights the power and flexibility of generative feed-forward models trained with complex and expressive loss functions.",
"This paper proposes Markovian Generative Adversarial Networks (MGANs), a method for training generative networks for efficient texture synthesis. While deep neural network approaches have recently demonstrated remarkable results in terms of synthesis quality, they still come at considerable computational costs (minutes of run-time for low-res images). Our paper addresses this efficiency issue. Instead of a numerical deconvolution in previous work, we precompute a feed-forward, strided convolutional network that captures the feature statistics of Markovian patches and is able to directly generate outputs of arbitrary dimensions. Such network can directly decode brown noise to realistic texture, or photos to artistic paintings. With adversarial training, we obtain quality comparable to recent neural texture synthesis methods. As no optimization is required at generation time, our run-time performance (0.25 M pixel images at 25 Hz) surpasses previous neural texture synthesizers by a significant margin (at least 500 times faster). We apply this idea to texture synthesis, style transfer, and video stylization.",
"This paper studies a combination of generative Markov random field (MRF) models and discriminatively trained deep convolutional neural networks (dCNNs) for synthesizing 2D images. The generative MRF acts on higher-levels of a dCNN feature pyramid, controling the image layout at an abstract level. We apply the method to both photographic and non-photo-realistic (artwork) synthesis tasks. The MRF regularizer prevents over-excitation artifacts and reduces implausible feature mixtures common to previous dCNN inversion approaches, permitting synthezing photographic content with increased visual plausibility. Unlike standard MRF-based texture synthesis, the combined system can both match and adapt local features with considerable variability, yielding results far out of reach of classic generative MRF methods."
]
} |
1812.03264 | 2964137521 | Chinese traditional painting is one of the most historical artworks in the world. It is very popular in Eastern and Southeast Asia due to being aesthetically appealing. Compared with western artistic painting, it is usually more visually abstract and textureless. Recently, neural network based style transfer methods have shown promising and appealing results which are mainly focused on western painting. It remains a challenging problem to preserve abstraction in neural style transfer. In this paper, we present a Neural Abstract Style Transfer method for Chinese traditional painting. It learns to preserve abstraction and other style jointly end-to-end via a novel MXDoG-guided filter (Modified version of the eXtended Difference-of-Gaussians) and three fully differentiable loss terms. To the best of our knowledge, there is little work study on neural style transfer of Chinese traditional painting. To promote research on this direction, we collect a new dataset with diverse photo-realistic images and Chinese traditional paintings (The dataset will be released at https: github.com lbsswu Chinese_style_transfer.). In experiments, the proposed method shows more appealing stylized results in transferring the style of Chinese traditional painting than state-of-the-art neural style transfer methods. | Recently, several work @cite_8 @cite_37 @cite_11 @cite_28 try to use or incorporate generative adversarial networks (GAN) for image style transfer. Specifically, the Cycle-GAN method @cite_8 produces amazing results in transferring an image with a painter's style, e.g., Vincent van Gogh, Monet. However, this method is not stable, needs much more time and requires large number of unpaired content and style images for training. What's more, the style of a painting maybe quite different from another even they are painted by the same artist, thus it may be not desirable when we just want to transfer the style of a specific artwork. | {
"cite_N": [
"@cite_28",
"@cite_37",
"@cite_11",
"@cite_8"
],
"mid": [
"2623975327",
"2339754110",
"2610846231",
""
],
"abstract": [
"This paper aims at synthesizing filamentary structured images such as retinal fundus images and neuronal images, as follows: Given a ground-truth, to generate multiple realistic looking phantoms. A ground-truth could be a binary segmentation map containing the filamentary structured morphology, while the synthesized output image is of the same size as the ground-truth and has similar visual appearance to what have been presented in the training set. Our approach is inspired by the recent progresses in generative adversarial nets (GANs) as well as image style transfer. In particular, it is dedicated to our problem context with the following properties: Rather than large-scale dataset, it works well in the presence of as few as 10 training examples, which is common in medical image analysis; It is capable of synthesizing diverse images from the same ground-truth; Last and importantly, the synthetic images produced by our approach are demonstrated to be useful in boosting image analysis performance. Empirical examination over various benchmarks of fundus and neuronal images demonstrate the advantages of the proposed approach.",
"This paper proposes Markovian Generative Adversarial Networks (MGANs), a method for training generative networks for efficient texture synthesis. While deep neural network approaches have recently demonstrated remarkable results in terms of synthesis quality, they still come at considerable computational costs (minutes of run-time for low-res images). Our paper addresses this efficiency issue. Instead of a numerical deconvolution in previous work, we precompute a feed-forward, strided convolutional network that captures the feature statistics of Markovian patches and is able to directly generate outputs of arbitrary dimensions. Such network can directly decode brown noise to realistic texture, or photos to artistic paintings. With adversarial training, we obtain quality comparable to recent neural texture synthesis methods. As no optimization is required at generation time, our run-time performance (0.25 M pixel images at 25 Hz) surpasses previous neural texture synthesizers by a significant margin (at least 500 times faster). We apply this idea to texture synthesis, style transfer, and video stylization.",
"Recently, realistic image generation using deep neural networks has become a hot topic in machine learning and computer vision. Images can be generated at the pixel level by learning from a large collection of images. Learning to generate colorful cartoon images from black-and-white sketches is not only an interesting research problem, but also a potential application in digital entertainment. In this paper, we investigate the sketch-to-image synthesis problem by using conditional generative adversarial networks (cGAN). We propose the auto-painter model which can automatically generate compatible colors for a sketch. The new model is not only capable of painting hand-draw sketch with proper colors, but also allowing users to indicate preferred colors. Experimental results on two sketch datasets show that the auto-painter performs better that existing image-to-image methods.",
""
]
} |
1812.02793 | 2949159381 | Machine learning (ML) and Natural Language Processing (NLP) have achieved remarkable success in many fields and have brought new opportunities and high expectation in the analyses of medical data. The most common type of medical data is the massive free-text electronic medical records (EMR). It is widely regarded that mining such massive data can bring up important information for improving medical practices as well as for possible new discoveries on complex diseases. However, the free EMR texts are lacking consistent standards, rich of private information, and limited in availability. Also, as they are accumulated from everyday practices, it is often hard to have a balanced number of samples for the types of diseases under study. These problems hinder the development of ML and NLP methods for EMR data analysis. To tackle these problems, we developed a model to generate synthetic text of EMRs called Medical Text Generative Adversarial Network or mtGAN. It is based on the GAN framework and is trained by the REINFORCE algorithm. It takes disease features as inputs and generates synthetic texts as EMRs for the corresponding diseases. We evaluate the model from micro-level, macro-level and application-level on a Chinese EMR text dataset. The results show that the method has a good capacity to fit real data and can generate realistic and diverse EMR samples. This provides a novel way to avoid potential leakage of patient privacy while still supply sufficient well-controlled cohort data for developing downstream ML and NLP methods. It can also be used as a data augmentation method to assist studies based on real EMR data. | Text generation has been one of the most challenging problems in natural language processing. RNN and its variants Long Short-Term Memory (LSTM) @cite_0 and Gated Recurrent Unit (GRU) @cite_7 have achieved impressive performance in several complex tasks, such as machine translation and dialogue generation @cite_3 . The RNN language models are commonly used for sequence generation, and they are trained by MLE in an approach called @cite_9 . This mainstream method for auto-regressive models predicts the next token given the previous ground-truth tokens, which leads to the exposure bias problem. | {
"cite_N": [
"@cite_0",
"@cite_9",
"@cite_3",
"@cite_7"
],
"mid": [
"",
"2016589492",
"2951520714",
"2950635152"
],
"abstract": [
"",
"The exact form of a gradient-following learning algorithm for completely recurrent networks running in continually sampled time is derived and used as the basis for practical algorithms for temporal supervised learning tasks. These algorithms have (1) the advantage that they do not require a precisely defined training interval, operating while the network runs; and (2) the disadvantage that they require nonlocal communication in the network being trained and are computationally expensive. These algorithms allow networks having recurrent connections to learn complex tasks that require the retention of information over time periods having either fixed or indefinite length.",
"In this paper, drawing intuition from the Turing test, we propose using adversarial training for open-domain dialogue generation: the system is trained to produce sequences that are indistinguishable from human-generated dialogue utterances. We cast the task as a reinforcement learning (RL) problem where we jointly train two systems, a generative model to produce response sequences, and a discriminator---analagous to the human evaluator in the Turing test--- to distinguish between the human-generated dialogues and the machine-generated ones. The outputs from the discriminator are then used as rewards for the generative model, pushing the system to generate dialogues that mostly resemble human dialogues. In addition to adversarial training we describe a model for adversarial evaluation that uses success in fooling an adversary as a dialogue evaluation metric, while avoiding a number of potential pitfalls. Experimental results on several metrics, including adversarial evaluation, demonstrate that the adversarially-trained system generates higher-quality responses than previous baselines.",
"In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases."
]
} |
1812.02793 | 2949159381 | Machine learning (ML) and Natural Language Processing (NLP) have achieved remarkable success in many fields and have brought new opportunities and high expectation in the analyses of medical data. The most common type of medical data is the massive free-text electronic medical records (EMR). It is widely regarded that mining such massive data can bring up important information for improving medical practices as well as for possible new discoveries on complex diseases. However, the free EMR texts are lacking consistent standards, rich of private information, and limited in availability. Also, as they are accumulated from everyday practices, it is often hard to have a balanced number of samples for the types of diseases under study. These problems hinder the development of ML and NLP methods for EMR data analysis. To tackle these problems, we developed a model to generate synthetic text of EMRs called Medical Text Generative Adversarial Network or mtGAN. It is based on the GAN framework and is trained by the REINFORCE algorithm. It takes disease features as inputs and generates synthetic texts as EMRs for the corresponding diseases. We evaluate the model from micro-level, macro-level and application-level on a Chinese EMR text dataset. The results show that the method has a good capacity to fit real data and can generate realistic and diverse EMR samples. This provides a novel way to avoid potential leakage of patient privacy while still supply sufficient well-controlled cohort data for developing downstream ML and NLP methods. It can also be used as a data augmentation method to assist studies based on real EMR data. | GAN proposed by Goodfellow provides an alternative framework to generate synthetic data. The GAN model consists of two neural networks: a generator G trying to generate synthetic data, and a discriminator D trying to distinguish the real data from the synthetic. The training procedure is a two-player zero-sum game between G and D. GANs have enjoyed great success in image generation, but they are not as much widely applied in natural language processing tasks. One reason is that the gradient from the discriminator cannot be back-propagated to the generator due to the discrete outputs. To address this problem, Yu proposed seqGAN @cite_1 , where the generator is updated through the policy gradient using Reinforcement Learning (RL), and the reward is calculated by the discriminator on a complete sequence via Monte Carlo search. However, no one has ever used this kind of approach in the case of generating synthetic EMR text. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2523469089"
],
"abstract": [
"As a new way of training generative models, Generative Adversarial Nets (GAN) that uses a discriminative model to guide the training of the generative model has enjoyed considerable success in generating real-valued data. However, it has limitations when the goal is for generating sequences of discrete tokens. A major reason lies in that the discrete outputs from the generative model make it difficult to pass the gradient update from the discriminative model to the generative model. Also, the discriminative model can only assess a complete sequence, while for a partially generated sequence, it is non-trivial to balance its current score and the future one once the entire sequence has been generated. In this paper, we propose a sequence generation framework, called SeqGAN, to solve the problems. Modeling the data generator as a stochastic policy in reinforcement learning (RL), SeqGAN bypasses the generator differentiation problem by directly performing gradient policy update. The RL reward signal comes from the GAN discriminator judged on a complete sequence, and is passed back to the intermediate state-action steps using Monte Carlo search. Extensive experiments on synthetic data and real-world tasks demonstrate significant improvements over strong baselines."
]
} |
1812.02809 | 2905557635 | This paper proposes a systematic framework to assess the complementarity of renewable resources over arbitrary geographical scopes and temporal scales which is particularly well-suited to exploit very large data sets of climatological data. The concept of critical time windows is introduced, and a spatio-temporal criticality indicator is proposed, consisting in a parametrised family of scalar indicators quantifying the complementarity between renewable resources in both space and time. The criticality indicator is leveraged to devise a family of optimisation problems identifying sets of locations with maximum complementarity under arbitrary geographical deployment constraints. The applicability of the framework is shown in a case study investigating the complementarity between the wind regimes in continental western Europe and southern Greenland, and its usefulness in a power system planning context is demonstrated. Besides showing that the occurrence of low wind power production events can be significantly reduced on a regional scale by exploiting diversity in local wind patterns, results highlight the fact that aggregating wind power production sites located on different continents may result in a lower occurrence of system-wide low wind power production events and indicate potential benefits of intercontinental electrical interconnections. | In @cite_18 , the smoothing effect that may result from integrating wind power plants over large geographical areas is considered using frequency-domain analysis methods introduced in @cite_28 instead of statistical tools. More precisely, the level of smoothing is mapped to a feature of the power spectral density of the aggregate system, highlighting the reduction in short-term variability as the geographical scope increases. A similar study is performed for utility-scale solar power plants in Gujarat in @cite_21 , particularly highlighting diminishing marginal returns in variability reduction as more plants are aggregated within the state. Finally, the use of scalar indices to study the complementarity between renewable resources has been proposed in @cite_4 and @cite_26 . In particular, in @cite_26 , the index is defined as the product of three partial indices describing temporal aspects, energy content and the amplitude of the signals considered, respectively. This index is then used to quantify complementarity between wind and solar resources across Oklahoma and identify physical factors such as moisture and temperature directly influencing complementarity. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_4",
"@cite_28",
"@cite_21"
],
"mid": [
"2121932946",
"2014283427",
"1995795464",
"2013711389",
"2205961124"
],
"abstract": [
"We present the first frequency-dependent analyses of the geographic smoothing of wind power's variability, analyzing the interconnected measured output of 20 wind plants in Texas. Reductions in variability occur at frequencies corresponding to times shorter than 24Â h and are quantified by measuring the departure from a Kolmogorov spectrum. At a frequency of 2.8x10-4Â Hz (corresponding to 1Â h), an 87 reduction of the variability of a single wind plant is obtained by interconnecting 4 wind plants. Interconnecting the remaining 16 wind plants produces only an additional 8 reduction. We use step change analyses and correlation coefficients to compare our results with previous studies, finding that wind power ramps up faster than it ramps down for each of the step change intervals analyzed and that correlation between the power output of wind plants 200Â km away is half that of co-located wind plants. To examine variability at very low frequencies, we estimate yearly wind energy production in the Great Plains region of the United States from automated wind observations at airports covering 36 years. The estimated wind power has significant inter-annual variability and the severity of wind drought years is estimated to be about half that observed nationally for hydroelectric power.",
"Energy in wind and incoming solar radiation has been observed to be complementary over time. That is, as one of these resources slackens, the other tends to increase. Considering the intermittency of wind and solar radiation, complementarity provides the potential to better exploit the two resources in combination. How well a hybrid system using both resources can improve the reliability of energy generation is partially a function of their complementarity. This study developed an approach to calculate the Complementarity Index of Wind and Solar Radiation (CIWS). Geographic analyses including principal component analysis and geographically weighted regression modeling were applied to investigate the impacts of diverse geographic factors on complementarity. Oklahoma was used as the case study area because of the availability of quality-controlled five-minute data from 127 Mesonet sites. The results indicate average CIWS (unitless) is 10.99, about 46 percent of the theoretical maximum value of 24; the stand...",
"The power plants based on renewable sources face various acceptance difficulties mainly due to high initial costs and low overall efficiencies. A way to improve performance of these systems is to utilize more than one type of source of energy chosen to provide some degree of complementarity. An interesting, albeit improbable, combination is obtained using hydroelectric and photovoltaic energy, taking advantage of the complementarity between these two types of sources coupled to reservoir and or battery storage. This paper discusses energy complementarity in time and proposes a numerical dimensionless index, evaluating this energy complementarity between two types of energy sources, in the same or in different locations, or between two sources of the same type in different locations. In the end, the results of the application of this index to the solar and water availabilities over the state of Rio Grande do Sul, in southern Brazil, is presented in the form of maps.",
"The power spectral density of the output of wind turbines provides information on the character of fluctuations in turbine output. Here both 1-second and 1-hour samples are used to estimate the power spectrum of several wind farms. The measured output power is found to follow a Kolmogorov spectrum over more than four orders of magnitude, from 30 s to 2.6 days. This result is in sharp contrast to the only previous study covering long time periods, published 50 years ago. The spectrum defines the character of fill-in power that must be provided to compensate for wind's fluctuations when wind is deployed at large scale. Installing enough linear ramp rate generation (such as a gas generator) to fill in fast fluctuations with amplitudes of 1 of the maximum fluctuation would oversize the fill-in generation capacity by a factor of two for slower fluctuations, greatly increasing capital costs. A wind system that incorporates batteries, fuel cells, supercapacitors, or other fast-ramp-rate energy storage systems would match fluctuations much better, and can provide an economic route for deployment of energy storage systems when renewable portfolio standards require large amounts of intermittent renewable generating sources.",
"We examine the potential for geographic smoothing of solar photovoltaic (PV) electricity generation using 13 months of observed power production from utility-scale plants in Gujarat, India. To our knowledge, this is the first published analysis of geographic smoothing of solar PV using actual generation data at high time resolution from utility-scale solar PV plants. We use geographic correlation and Fourier transform estimates of the power spectral density (PSD) to characterize the observed variability of operating solar PV plants as a function of time scale. Most plants show a spectrum that is linear in the log–log domain at high frequencies f, ranging from to (slopes of −1.23 and −1.56), thus exhibiting more relative variability at high frequencies than exhibited by wind plants. PSDs for large PV plants have a steeper slope than those for small plants, hence more smoothing at short time scales. Interconnecting 20 Gujarat plants yields a spectrum, reducing fluctuations at frequencies corresponding to 6 h and 1 h by 23 and 45 , respectively. Half of this smoothing can be obtained through connecting 4–5 plants; reaching marginal improvement of 1 per added plant occurs at 12–14 plants. The largest plant (322 MW) showed an spectrum. This suggests that in Gujarat the potential for smoothing is limited to that obtained by one large plant."
]
} |
1812.02897 | 2904463595 | It is well known that popular optimization techniques can lead to overfitting or even a lack of convergence altogether; thus, practitioners often utilize ad hoc regularization terms added to the energy functional. When carefully crafted, these regularizations can produce compelling results. However, regularization changes both the energy landscape and the solution to the optimization problem, which can result in underfitting. Surprisingly, many practitioners both add regularization and claim that their model lacks the expressivity to fit the data. Motivated by a geometric interpretation of the linearized search space, we propose an approach that ameliorates overfitting without the need for regularization terms that restrict the expressiveness of the underlying model. We illustrate the efficacy of our approach on minimization problems related to three-dimensional facial expression estimation where overfitting clouds semantic understanding and regularization may lead to underfitting that misses or misinterprets subtle expressions. | At each iteration, coordinate descent chooses a single coordinate direction to minimize. This allows the coordinate descent algorithm to avoid null spaces making it an attractive option for use on ill-posed, poorly conditioned problems. The column can be chosen stochastically @cite_12 or deterministically. Popular deterministic methods for choosing the next search direction include cyclic coordinate descent @cite_10 , the Gauss-Southwell and Gauss-Southwell-Lipschitz rule @cite_16 , and the maximum block improvement (MBI) rule @cite_9 . Generally, either a line search is performed to determine the step size or a fixed step size learning rate is used. However, the cost of a line search is prohibitive when the function takes a long time to evaluate ( in the case of simulations) so we instead use coordinate descent to solve the linear problem found at every iteration of Gauss-Newton. Instead of looking at a single column at a time, block coordinate descent can be used to update multiple columns simultaneously @cite_23 ; however, with block coordinate descent, regularization may still be needed to avoid ill-posed problems when the block of columns does not have full rank. A more complete overview of coordinate descent methods can be found in @cite_26 @cite_2 . | {
"cite_N": [
"@cite_26",
"@cite_9",
"@cite_23",
"@cite_2",
"@cite_16",
"@cite_10",
"@cite_12"
],
"mid": [
"2529163417",
"2126381689",
"1540764732",
"",
"1871644925",
"",
"2095984592"
],
"abstract": [
"This monograph presents a class of algorithms called coordinate descent algorithms for mathematicians, statisticians, and engineers outside the field of optimization. This particular class of algorithms has recently gained popularity due to their effectiveness in solving large-scale optimization problems in machine learning, compressed sensing, image processing, and computational statistics. Coordinate descent algorithms solve optimization problems by successively minimizing along each coordinate or coordinate hyperplane, which is ideal for parallelized and distributed computing. Avoiding detailed technicalities and proofs, this monograph gives relevant theory and examples for practitioners to effectively apply coordinate descent to modern problems in data science and engineering.",
"In this paper we propose an efficient method for solving the spherically constrained homogeneous polynomial optimization problem. The new approach has the following three main ingredients. First, we establish a block coordinate descent type search method for nonlinear optimization, with the novelty being that we accept only a block update that achieves the maximum improvement, hence the name of our new search method: maximum block improvement (MBI). Convergence of the sequence produced by the MBI method to a stationary point is proved. Second, we establish that maximizing a homogeneous polynomial over a sphere is equivalent to its tensor relaxation problem; thus we can maximize a homogeneous polynomial function over a sphere by its tensor relaxation via the MBI approach. Third, we propose a scheme to reach a KKT point of the polynomial optimization, provided that a stationary solution for the relaxed tensor problem is available. Numerical experiments have shown that our new method works very efficiently: for a majority of the test instances that we have experimented with, the method finds the global optimal solution at a low computational cost.",
"We study the convergence properties of a (block) coordinate descent method applied to minimize a nondifferentiable (nonconvex) function f(x1, . . . , x N ) with certain separability and regularity properties. Assuming that f is continuous on a compact level set, the subsequence convergence of the iterates to a stationary point is shown when either f is pseudoconvex in every pair of coordinate blocks from among N-1 coordinate blocks or f has at most one minimum in each of N-2 coordinate blocks. If f is quasiconvex and hemivariate in every coordinate block, then the assumptions of continuity of f and compactness of the level set may be relaxed further. These results are applied to derive new (and old) convergence results for the proximal minimization algorithm, an algorithm of Arimoto and Blahut, and an algorithm of Han. They are applied also to a problem of blind source separation.",
"",
"There has been significant recent work on the theory and application of randomized coordinate descent algorithms, beginning with the work of Nesterov [SIAM J. Optim., 22(2), 2012], who showed that a random-coordinate selection rule achieves the same convergence rate as the Gauss-Southwell selection rule. This result suggests that we should never use the Gauss-Southwell rule, as it is typically much more expensive than random selection. However, the empirical behaviours of these algorithms contradict this theoretical result: in applications where the computational costs of the selection rules are comparable, the Gauss-Southwell selection rule tends to perform substantially better than random coordinate selection. We give a simple analysis of the Gauss-Southwell rule showing that---except in extreme cases---it's convergence rate is faster than choosing random coordinates. Further, in this work we (i) show that exact coordinate optimization improves the convergence rate for certain sparse problems, (ii) propose a Gauss-Southwell-Lipschitz rule that gives an even faster convergence rate given knowledge of the Lipschitz constants of the partial derivatives, (iii) analyze the effect of approximate Gauss-Southwell rules, and (iv) analyze proximal-gradient variants of the Gauss-Southwell rule.",
"",
"In this paper we propose new methods for solving huge-scale optimization problems. For problems of this size, even the simplest full-dimensional vector operations are very expensive. Hence, we propose to apply an optimization technique based on random partial update of decision variables. For these methods, we prove the global estimates for the rate of convergence. Surprisingly enough, for certain classes of objective functions, our results are better than the standard worst-case bounds for deterministic algorithms. We present constrained and unconstrained versions of the method, and its accelerated variant. Our numerical test confirms a high efficiency of this technique on problems of very big size."
]
} |
1812.02897 | 2904463595 | It is well known that popular optimization techniques can lead to overfitting or even a lack of convergence altogether; thus, practitioners often utilize ad hoc regularization terms added to the energy functional. When carefully crafted, these regularizations can produce compelling results. However, regularization changes both the energy landscape and the solution to the optimization problem, which can result in underfitting. Surprisingly, many practitioners both add regularization and claim that their model lacks the expressivity to fit the data. Motivated by a geometric interpretation of the linearized search space, we propose an approach that ameliorates overfitting without the need for regularization terms that restrict the expressiveness of the underlying model. We illustrate the efficacy of our approach on minimization problems related to three-dimensional facial expression estimation where overfitting clouds semantic understanding and regularization may lead to underfitting that misses or misinterprets subtle expressions. | Although coordinate descent algorithms can avoid the null space, they may still choose poorly correlated coordinates potentially resulting in many poorly correlated coordinates in place of fewer more strongly correlated coordinates. Using correlation to choose the next coordinate to add to the model can alleviate this problem and is the central idea behind MBI @cite_9 , forward and backward stepwise regression @cite_20 , and least angle regression (LARS) @cite_35 . The latter statistical regression methods are often used to gain better prediction accuracy and interpretability of the model @cite_11 . However, LARS converges to the least squares solution @cite_35 which often means eventually using uncorrelated coordinates. | {
"cite_N": [
"@cite_35",
"@cite_9",
"@cite_20",
"@cite_11"
],
"mid": [
"2063978378",
"2126381689",
"2052779929",
""
],
"abstract": [
"The purpose of model selection algorithms such as All Subsets, Forward Selection and Backward Elimination is to choose a linear model on the basis of the same set of data to which the model will be applied. Typically we have available a large collection of possible covariates from which we hope to select a parsimonious set for the efficient prediction of a response variable. Least Angle Regression (LARS), a new model selection algorithm, is a useful and less greedy version of traditional forward selection methods. Three main properties are derived: (1) A simple modification of the LARS algorithm implements the Lasso, an attractive version of ordinary least squares that constrains the sum of the absolute regression coefficients; the LARS modification calculates all possible Lasso estimates for a given problem, using an order of magnitude less computer time than previous methods. (2) A different LARS modification efficiently implements Forward Stagewise linear regression, another promising new model selection method; this connection explains the similar numerical results previously observed for the Lasso and Stagewise, and helps us understand the properties of both methods, which are seen as constrained versions of the simpler LARS algorithm. (3) A simple approximation for the degrees of freedom of a LARS estimate is available, from which we derive a Cp estimate of prediction error; this allows a principled choice among the range of possible LARS estimates. LARS and its variants are computationally efficient: the paper describes a publicly available algorithm that requires only the same order of magnitude of computational effort as ordinary least squares applied to the full set of covariates.",
"In this paper we propose an efficient method for solving the spherically constrained homogeneous polynomial optimization problem. The new approach has the following three main ingredients. First, we establish a block coordinate descent type search method for nonlinear optimization, with the novelty being that we accept only a block update that achieves the maximum improvement, hence the name of our new search method: maximum block improvement (MBI). Convergence of the sequence produced by the MBI method to a stationary point is proved. Second, we establish that maximizing a homogeneous polynomial over a sphere is equivalent to its tensor relaxation problem; thus we can maximize a homogeneous polynomial function over a sphere by its tensor relaxation via the MBI approach. Third, we propose a scheme to reach a KKT point of the polynomial optimization, provided that a stationary solution for the relaxed tensor problem is available. Numerical experiments have shown that our new method works very efficiently: for a majority of the test instances that we have experimented with, the method finds the global optimal solution at a low computational cost.",
"The use of automated subset search algorithms is reviewed and issues concerning model selection and selection criteria are discussed. In addition, a Monte Carlo study is reported which presents data regarding the frequency with which authentic and noise variables are selected by automated subset algorithms. In particular, the effects of the correlation between predictor variables, the number of candidate predictor variables, the size of the sample, and the level of significance for entry and deletion of variables were studied for three automated subset algorithms: BACKWARD ELIMINATION, FORWARD SELECTION, and STEPWISE. Results indicated that: (1) the degree of correlation between the predictor variables affected the frequency with which authentic predictor variables found their way into the final model; (2) the number of candidate predictor variables affected the number of noise variables that gained entry to the model; (3) the size of the sample was of little practical importance in determining the number of authentic variables contained in the final model; and (4) the population multiple coefficient of determination could be faithfully estimated by adopting a statistic that is adjusted by the total number of candidate predictor variables rather than the number of variables in the final model.",
""
]
} |
1812.02897 | 2904463595 | It is well known that popular optimization techniques can lead to overfitting or even a lack of convergence altogether; thus, practitioners often utilize ad hoc regularization terms added to the energy functional. When carefully crafted, these regularizations can produce compelling results. However, regularization changes both the energy landscape and the solution to the optimization problem, which can result in underfitting. Surprisingly, many practitioners both add regularization and claim that their model lacks the expressivity to fit the data. Motivated by a geometric interpretation of the linearized search space, we propose an approach that ameliorates overfitting without the need for regularization terms that restrict the expressiveness of the underlying model. We illustrate the efficacy of our approach on minimization problems related to three-dimensional facial expression estimation where overfitting clouds semantic understanding and regularization may lead to underfitting that misses or misinterprets subtle expressions. | When solving for facial blendshape parameters @cite_18 , L2-norm regularization of the parameters is commonly used by practitioners to obtain more reasonable solutions due to its ease of use in commodity least squares solvers @cite_17 @cite_34 @cite_3 @cite_7 . Others instead choose to bound the blendshape weights between some minimum and maximum value (typically @math and @math or @math and @math ) @cite_24 @cite_29 @cite_37 . The usage of the L1-norm regularization is relatively rare, even though it results in sparser solutions @cite_4 @cite_21 @cite_19 . Other methods use Laplacian regularization @cite_24 @cite_15 or anatomically motivated regularization @cite_6 to constrain the deformation of the face. | {
"cite_N": [
"@cite_18",
"@cite_37",
"@cite_4",
"@cite_7",
"@cite_29",
"@cite_21",
"@cite_3",
"@cite_6",
"@cite_24",
"@cite_19",
"@cite_15",
"@cite_34",
"@cite_17"
],
"mid": [
"788203010",
"",
"1973757125",
"2301937176",
"1927238661",
"2013871799",
"",
"2474891805",
"",
"2134389879",
"",
"2017107803",
"2237250383"
],
"abstract": [
"",
"",
"We present a new algorithm for realtime face tracking on commodity RGB-D sensing devices. Our method requires no user-specific training or calibration, or any other form of manual assistance, thus enabling a range of new applications in performance-based facial animation and virtual interaction at the consumer level. The key novelty of our approach is an optimization algorithm that jointly solves for a detailed 3D expression model of the user and the corresponding dynamic tracking parameters. Realtime performance and robust computations are facilitated by a novel subspace parameterization of the dynamic facial expression space. We provide a detailed evaluation that shows that our approach significantly simplifies the performance capture workflow, while achieving accurate facial tracking for realtime applications.",
"We present a novel approach for real-time facial reenactment of a monocular target video sequence (e.g., Youtube video). The source sequence is also a monocular video stream, captured live with a commodity webcam. Our goal is to animate the facial expressions of the target video by a source actor and re-render the manipulated output video in a photo-realistic fashion. To this end, we first address the under-constrained problem of facial identity recovery from monocular video by non-rigid model-based bundling. At run time, we track facial expressions of both source and target video using a dense photometric consistency measure. Reenactment is then achieved by fast and efficient deformation transfer between source and target. The mouth interior that best matches the re-targeted expression is retrieved from the target sequence and warped to produce an accurate fit. Finally, we convincingly re-render the synthesized target face on top of the corresponding video stream such that it seamlessly blends with the real-world illumination. We demonstrate our method in a live setup, where Youtube videos are reenacted in real time.",
"We introduce a realtime facial tracking system specifically designed for performance capture in unconstrained settings using a consumer-level RGB-D sensor. Our framework provides uninterrupted 3D facial tracking, even in the presence of extreme occlusions such as those caused by hair, hand-to-face gestures, and wearable accessories. Anyone's face can be instantly tracked and the users can be switched without an extra calibration step. During tracking, we explicitly segment face regions from any occluding parts by detecting outliers in the shape and appearance input using an exponentially smoothed and user-adaptive tracking model as prior. Our face segmentation combines depth and RGB input data and is also robust against illumination changes. To enable continuous and reliable facial feature tracking in the color channels, we synthesize plausible face textures in the occluded regions. Our tracking model is personalized on-the-fly by progressively refining the user's identity, expressions, and texture with reliable samples and temporal filtering. We demonstrate robust and high-fidelity facial tracking on a wide range of subjects with highly incomplete and largely occluded data. Our system works in everyday environments and is fully unobtrusive to the user, impacting consumer AR applications and surveillance.",
"We present a complete pipeline for creating fully rigged, personalized 3D facial avatars from hand-held video. Our system faithfully recovers facial expression dynamics of the user by adapting a blendshape template to an image sequence of recorded expressions using an optimization that integrates feature tracking, optical flow, and shape from shading. Fine-scale details such as wrinkles are captured separately in normal maps and ambient occlusion maps. From this user- and expression-specific data, we learn a regressor for on-the-fly detail synthesis during animation to enhance the perceptual realism of the avatars. Our system demonstrates that the use of appropriate reconstruction priors yields compelling face rigs even with a minimalistic acquisition system and limited user assistance. This facilitates a range of new applications in computer animation and consumer-level online communication based on personalized avatars. We present realtime application demos to validate our method.",
"",
"We present a new anatomically-constrained local face model and fitting approach for tracking 3D faces from 2D motion data in very high quality. In contrast to traditional global face models, often built from a large set of blendshapes, we propose a local deformation model composed of many small subspaces spatially distributed over the face. Our local model offers far more flexibility and expressiveness than global blendshape models, even with a much smaller model size. This flexibility would typically come at the cost of reduced robustness, in particular during the under-constrained task of monocular reconstruction. However, a key contribution of this work is that we consider the face anatomy and introduce subspace skin thickness constraints into our model, which constrain the face to only valid expressions and helps counteract depth ambiguities in monocular tracking. Given our new model, we present a novel fitting optimization that allows 3D facial performance reconstruction from a single view at extremely high quality, far beyond previous fitting approaches. Our model is flexible, and can be applied also when only sparse motion data is available, for example with marker-based motion capture or even face posing from artistic sketches. Furthermore, by incorporating anatomical constraints we can automatically estimate the rigid motion of the skull, obtaining a rigid stabilization of the performance for free. We demonstrate our model and single-view fitting method on a number of examples, including, for the first time, extreme local skin deformation caused by external forces such as wind, captured from a single high-speed camera.",
"",
"We propose a method that extracts sparse and spatially localized deformation modes from an animated mesh sequence. To this end, we propose a new way to extend the theory of sparse matrix decompositions to 3D mesh sequence processing, and further contribute with an automatic way to ensure spatial locality of the decomposition in a new optimization framework. The extracted dimensions often have an intuitive and clear interpretable meaning. Our method optionally accepts user-constraints to guide the process of discovering the underlying latent deformation space. The capabilities of our efficient, versatile, and easy-to-implement method are extensively demonstrated on a variety of data sets and application contexts. We demonstrate its power for user friendly intuitive editing of captured mesh animations, such as faces, full body motion, cloth animations, and muscle deformations. We further show its benefit for statistical geometry processing and biomechanically meaningful animation editing. It is further shown qualitatively and quantitatively that our method outperforms other unsupervised decomposition methods and other animation parameterization approaches in the above use cases.",
"",
"We present FaceWarehouse, a database of 3D facial expressions for visual computing applications. We use Kinect, an off-the-shelf RGBD camera, to capture 150 individuals aged 7-80 from various ethnic backgrounds. For each person, we captured the RGBD data of her different expressions, including the neutral expression and 19 other expressions such as mouth-opening, smile, kiss, etc. For every RGBD raw data record, a set of facial feature points on the color image such as eye corners, mouth contour, and the nose tip are automatically localized, and manually adjusted if better accuracy is required. We then deform a template facial mesh to fit the depth data as closely as possible while matching the feature points on the color image to their corresponding points on the mesh. Starting from these fitted face meshes, we construct a set of individual-specific expression blendshapes for each person. These meshes with consistent topology are assembled as a rank-3 tensor to build a bilinear face model with two attributes: identity and expression. Compared with previous 3D facial databases, for every person in our database, there is a much richer matching collection of expressions, enabling depiction of most human facial actions. We demonstrate the potential of FaceWarehouse for visual computing with four applications: facial image manipulation, face component transfer, real-time performance-based facial image animation, and facial animation retargeting from video to image.",
"In this paper, a new technique for modeling textured 3D faces is introduced. 3D faces can either be generated automatically from one or more photographs, or modeled directly through an intuitive user interface. Users are assisted in two key problems of computer aided face modeling. First, new face images or new 3D face models can be registered automatically by computing dense one-to-one correspondence to an internal face model. Second, the approach regulates the naturalness of modeled faces avoiding faces with an “unlikely” appearance. Starting from an example set of 3D face models, we derive a morphable face model by transforming the shape and texture of the examples into a vector space representation. New faces and expressions can be modeled by forming linear combinations of the prototypes. Shape and texture constraints derived from the statistics of our example faces are used to guide manual modeling or automated matching algorithms. We show 3D face reconstructions from single images and their applications for photo-realistic image manipulations. We also demonstrate face manipulations according to complex parameters such as gender, fullness of a face or its distinctiveness."
]
} |
1812.02843 | 2904837360 | This work was performed under the following financial assistance award: 60NANB18D279 from U.S. Department of Commerce, National Institute of Standards and Technology, and also funding from SAP SE | Adversarial examples: Adversarial examples were discovered by @cite_9 who showed that state-of-the-art machine learning classifiers can be fooled comprehensively by simple backpropagation algorithms. @cite_4 improved this by Fast Gradient Sign Method (FGSM) that needs only one iteration of optimization. The possibility of extending these examples to the real world was shown in @cite_13 @cite_20 and recently, @cite_2 showed that adversarial examples could be robust to affine transformations. @cite_15 proposed Projected Gradient Descent (PGD) which has been shown to be the best first-order adversary for fooling classifiers. @cite_1 showed how saliency methods are unreliable by adding constant shift to input data and checking against different saliency methods. In our work, we show that it is not only possible to fool the classifier using an adversary, but also hide it from standard network interpretation algorithms. | {
"cite_N": [
"@cite_4",
"@cite_9",
"@cite_1",
"@cite_2",
"@cite_15",
"@cite_13",
"@cite_20"
],
"mid": [
"1945616565",
"1673923490",
"2766047647",
"2736899637",
"2640329709",
"2460937040",
"2535873859"
],
"abstract": [
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.",
"Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.",
"Saliency methods aim to explain the predictions of deep neural networks. These methods lack reliability when the explanation is sensitive to factors that do not contribute to the model prediction. We use a simple and common pre-processing step ---adding a mean shift to the input data--- to show that a transformation with no effect on the model can cause numerous methods to incorrectly attribute. We define input invariance as the requirement that a saliency method mirror the sensitivity of the model with respect to transformations of the input. We show, through several examples, that saliency methods that do not satisfy a input invariance property are unreliable and can lead to misleading and inaccurate attribution.",
"Standard methods for generating adversarial examples for neural networks do not consistently fool neural network classifiers in the physical world due to a combination of viewpoint shifts, camera noise, and other natural transformations, limiting their relevance to real-world systems. We demonstrate the existence of robust 3D adversarial objects, and we present the first algorithm for synthesizing examples that are adversarial over a chosen distribution of transformations. We synthesize two-dimensional adversarial images that are robust to noise, distortion, and affine transformation. We apply our algorithm to complex three-dimensional objects, using 3D-printing to manufacture the first physical adversarial objects. Our results demonstrate the existence of 3D adversarial objects in the physical world.",
"Recent work has demonstrated that neural networks are vulnerable to adversarial examples, i.e., inputs that are almost indistinguishable from natural data and yet classified incorrectly by the network. In fact, some of the latest findings suggest that the existence of adversarial attacks may be an inherent weakness of deep learning models. To address this problem, we study the adversarial robustness of neural networks through the lens of robust optimization. This approach provides us with a broad and unifying view on much of the prior work on this topic. Its principled nature also enables us to identify methods for both training and attacking neural networks that are reliable and, in a certain sense, universal. In particular, they specify a concrete security guarantee that would protect against any adversary. These methods let us train networks with significantly improved resistance to a wide range of adversarial attacks. They also suggest the notion of security against a first-order adversary as a natural and broad security guarantee. We believe that robustness against such well-defined classes of adversaries is an important stepping stone towards fully resistant deep learning models.",
"Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model. Up to now, all previous work have assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as an input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera.",
"Machine learning is enabling a myriad innovations, including new algorithms for cancer diagnosis and self-driving cars. The broad use of machine learning makes it important to understand the extent to which machine-learning algorithms are subject to attack, particularly when used in applications where physical security or safety is at risk. In this paper, we focus on facial biometric systems, which are widely used in surveillance and access control. We define and investigate a novel class of attacks: attacks that are physically realizable and inconspicuous, and allow an attacker to evade recognition or impersonate another individual. We develop a systematic method to automatically generate such attacks, which are realized through printing a pair of eyeglass frames. When worn by the attacker whose image is supplied to a state-of-the-art face-recognition algorithm, the eyeglasses allow her to evade being recognized or to impersonate another individual. Our investigation focuses on white-box face-recognition systems, but we also demonstrate how similar techniques can be used in black-box scenarios, as well as to avoid face detection."
]
} |
1812.02843 | 2904837360 | This work was performed under the following financial assistance award: 60NANB18D279 from U.S. Department of Commerce, National Institute of Standards and Technology, and also funding from SAP SE | Interpretation of deep networks: As neural networks are getting closer towards deployment in real world applications, it is important that their results are interpretable. Researchers have proposed various algorithms in this direction. One of the earliest attempt was done in @cite_8 where they calculate the derivative of the network's outputs w.r.t the input to compute class specific saliency maps. @cite_0 calculates the change in the network output when a small portion of the image (say @math pixels) is covered by a random occluder. We call this . CAM @cite_12 used weighted average map for each image based on their activations. The most popular one that we consider in this paper is called @cite_11 , a gradient based method which provides visual explanations for any neural network architecture. Kunpeng al @cite_6 recently improved upon Grad-CAM using Guided attention mechanism with state-of-the-art results on segmentation tasks. Although the above methods have shown great improvement in explaining the network's decision, our work highlights that it is important to ensure that they are robust enough to adversaries as well. | {
"cite_N": [
"@cite_8",
"@cite_6",
"@cite_0",
"@cite_12",
"@cite_11"
],
"mid": [
"2962851944",
"2788250925",
"1899185266",
"2950328304",
""
],
"abstract": [
"This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [5], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [13].",
"Weakly supervised learning with only coarse labels can obtain visual explanations of deep neural network such as attention maps by back-propagating gradients. These attention maps are then available as priors for tasks such as object localization and semantic segmentation. In one common framework we address three shortcomings of previous approaches in modeling such attention maps: We (1) first time make attention maps an explicit and natural component of the end-to-end training, (2) provide self-guidance directly on these maps by exploring supervision form the network itself to improve them, and (3) seamlessly bridge the gap between using weak and extra supervision if available. Despite its simplicity, experiments on the semantic segmentation task demonstrate the effectiveness of our methods. We clearly surpass the state-of-the-art on Pascal VOC 2012 val. and test set. Besides, the proposed framework provides a way not only explaining the focus of the learner but also feeding back with direct guidance towards specific tasks. Under mild assumptions our method can also be understood as a plug-in to existing weakly supervised learners to improve their generalization performance.",
"With the success of new computational architectures for visual processing, such as convolutional neural networks (CNN) and access to image databases with millions of labeled examples (e.g., ImageNet, Places), the state of the art in computer vision is advancing rapidly. One important factor for continued progress is to understand the representations that are learned by the inner layers of these deep architectures. Here we show that object detectors emerge from training CNNs to perform scene classification. As scenes are composed of objects, the CNN for scene classification automatically discovers meaningful objects detectors, representative of the learned scene categories. With object detectors emerging as a result of learning to recognize scenes, our work demonstrates that the same network can perform both scene recognition and object localization in a single forward-pass, without ever having been explicitly taught the notion of objects.",
"In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network to have remarkable localization ability despite being trained on image-level labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that can be applied to a variety of tasks. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1 top-5 error for object localization on ILSVRC 2014, which is remarkably close to the 34.2 top-5 error achieved by a fully supervised CNN approach. We demonstrate that our network is able to localize the discriminative image regions on a variety of tasks despite not being trained for them",
""
]
} |
1812.02843 | 2904837360 | This work was performed under the following financial assistance award: 60NANB18D279 from U.S. Department of Commerce, National Institute of Standards and Technology, and also funding from SAP SE | [width=.17 ] img_orig_000038.jpg & [width=.17 ] img_reg_patch_000038_orig_014_target_017_pred_017_prob_100.png & [width=.17 ] mask_reg_patch_000038_orig_014_target_017_pred_017.JPEG & [width=.17 ] img_our_patch_000038_orig_014_target_017__pred_017_prob_86.png & [width=.17 ] mask_our_patch_000038_orig_014_target_017_pred_017.png Person & Sofa & Sofa & Sofa & Sofa [width=.17 ] img_orig_000145.jpg & [width=.17 ] img_reg_patch_000145_orig_018_target_008_pred_008_prob_100.png & [width=.17 ] mask_reg_patch_000145_orig_018_target_008_pred_008.JPEG & [width=.17 ] img_our_patch_000145_orig_018_target_008__pred_008_prob_97.png & [width=.17 ] mask_our_patch_000145_orig_018_target_008_pred_008.png Train & Chair & Chair & Chair & Chair [width=.17 ] img_orig_000168.jpg & [width=.17 ] img_reg_patch_000168_orig_012_target_003_pred_003_prob_100.png & [width=.17 ] mask_reg_patch_000168_orig_012_target_003_pred_003.JPEG & [width=.17 ] img_our_patch_000168_orig_012_target_003__pred_003_prob_91.png & [width=.17 ] mask_our_patch_000168_orig_012_target_003_pred_003.png Horse & Boat & Boat & Boat & Boat [width=.17 ] img_orig_002007.jpg & [width=.17 ] img_reg_patch_002007_orig_000_target_017_pred_017_prob_100.png & [width=.17 ] mask_reg_patch_002007_orig_000_target_017_pred_017.JPEG & [width=.17 ] img_our_patch_002007_orig_000_target_017__pred_017_prob_99.png & [width=.17 ] mask_our_patch_002007_orig_000_target_017_pred_017.png Aeroplane & Sofa & Sofa & Sofa & Sofa @cite_6 network for VOC dataset. The predicted label is written under each image, the attack was successful for all images, and Grad-CAM is always computed for the target category. GAIN @math is particularly designed to produce better Grad-CAM visualizations using direct supervision of the Grad-CAM output. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2788250925"
],
"abstract": [
"Weakly supervised learning with only coarse labels can obtain visual explanations of deep neural network such as attention maps by back-propagating gradients. These attention maps are then available as priors for tasks such as object localization and semantic segmentation. In one common framework we address three shortcomings of previous approaches in modeling such attention maps: We (1) first time make attention maps an explicit and natural component of the end-to-end training, (2) provide self-guidance directly on these maps by exploring supervision form the network itself to improve them, and (3) seamlessly bridge the gap between using weak and extra supervision if available. Despite its simplicity, experiments on the semantic segmentation task demonstrate the effectiveness of our methods. We clearly surpass the state-of-the-art on Pascal VOC 2012 val. and test set. Besides, the proposed framework provides a way not only explaining the focus of the learner but also feeding back with direct guidance towards specific tasks. Under mild assumptions our method can also be understood as a plug-in to existing weakly supervised learners to improve their generalization performance."
]
} |
1812.02843 | 2904837360 | This work was performed under the following financial assistance award: 60NANB18D279 from U.S. Department of Commerce, National Institute of Standards and Technology, and also funding from SAP SE | [width=.17 ] img_orig_002081.jpg & [width=.17 ] mask_reg_patch_002081_orig_019_target_010_pred_010.JPEG & [width=.17 ] occ_reg_patch_adv_img_002081_10.png & [width=.17 ] mask_our_patch_002081_orig_019_target_010_pred_010.png & [width=.17 ] occ_our_patch_adv_img_002081_10.png TV Monitor & Dining Table & Dining Table & Dining Table & Dining Table [width=.17 ] img_orig_002245.jpg & [width=.17 ] mask_reg_patch_002245_orig_005_target_019_pred_019.JPEG & [width=.17 ] occ_reg_patch_adv_img_002245_19.png & [width=.17 ] mask_our_patch_002245_orig_005_target_019_pred_019.png & [width=.17 ] occ_our_patch_adv_img_002245_19.png Bus & TV Monitor & TV Monitor & TV Monitor & TV Monitor [width=.17 ] img_orig_003019.jpg & [width=.17 ] mask_reg_patch_003019_orig_007_target_001_pred_001.JPEG & [width=.17 ] occ_reg_patch_adv_img_003019_1.png & [width=.17 ] mask_our_patch_003019_orig_007_target_001_pred_001.png & [width=.17 ] occ_our_patch_adv_img_003019_1.png Cat & Bicycle & Bicycle & Bicycle & Bicycle [width=.17 ] img_orig_003069.jpg & [width=.17 ] mask_reg_patch_003069_orig_016_target_015_pred_015.JPEG & [width=.17 ] occ_reg_patch_adv_img_003069_15.png & [width=.17 ] mask_our_patch_003069_orig_016_target_015_pred_015.png & [width=.17 ] occ_our_patch_adv_img_003069_15.png Sheep & Potted Plant & Potted Plant & Potted Plant & Potted Plant @cite_6 network for VOC dataset. The predicted label is written under each image, the attack was successful for all images, and Grad-CAM and Occluding Patch visualizations are always computed for the target category. Note that the patch is hidden in both visualizations in columns 4 and 5. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2788250925"
],
"abstract": [
"Weakly supervised learning with only coarse labels can obtain visual explanations of deep neural network such as attention maps by back-propagating gradients. These attention maps are then available as priors for tasks such as object localization and semantic segmentation. In one common framework we address three shortcomings of previous approaches in modeling such attention maps: We (1) first time make attention maps an explicit and natural component of the end-to-end training, (2) provide self-guidance directly on these maps by exploring supervision form the network itself to improve them, and (3) seamlessly bridge the gap between using weak and extra supervision if available. Despite its simplicity, experiments on the semantic segmentation task demonstrate the effectiveness of our methods. We clearly surpass the state-of-the-art on Pascal VOC 2012 val. and test set. Besides, the proposed framework provides a way not only explaining the focus of the learner but also feeding back with direct guidance towards specific tasks. Under mild assumptions our method can also be understood as a plug-in to existing weakly supervised learners to improve their generalization performance."
]
} |
1812.02836 | 2905529320 | Muscle-based systems have the potential to provide both anatomical accuracy and semantic interpretability as compared to blendshape models; however, a lack of expressivity and differentiability has limited their impact. Thus, we propose modifying a recently developed rather expressive muscle-based system in order to make it fully-differentiable; in fact, our proposed modifications allow this physically robust and anatomically accurate muscle model to conveniently be driven by an underlying blendshape basis. Our formulation is intuitive, natural, as well as monolithically and fully coupled such that one can differentiate the model from end to end, which makes it viable for both optimization and learning-based approaches for a variety of applications. We illustrate this with a number of examples including both shape matching of three-dimensional geometry as as well as the automatic determination of a three-dimensional facial pose from a single two-dimensional RGB image without using markers or depth information. | A more comprehensive review of facial performance capture techniques can be found in @cite_46 . To date, marker based techniques have been the most popular for capturing facial performances for both real-time applications and feature films. Helmet mounted cameras (HMCs) are often used to stereo track a sparse set of markers on the face. These markers are then used as constraints in an optimization to find blendshape weights @cite_37 . In many real-time applications, pre-applied markers are generally not an option so 2D features @cite_20 @cite_30 @cite_5 , depth images @cite_30 @cite_23 @cite_7 , or low-resolution RGB images @cite_6 are often used instead. Other methods have focused on using traditional computer vision techniques to track a facial performance with consistent topology @cite_38 @cite_18 @cite_44 . More recently, methods using neutral networks have been used to reconstruct face geometry @cite_52 @cite_33 and estimate facial control parameters @cite_14 @cite_1 . Analysis-by-synthesis techniques have also been explored for capturing facial performances @cite_41 . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_37",
"@cite_18",
"@cite_14",
"@cite_33",
"@cite_7",
"@cite_41",
"@cite_1",
"@cite_52",
"@cite_6",
"@cite_44",
"@cite_23",
"@cite_5",
"@cite_46",
"@cite_20"
],
"mid": [
"2108432346",
"2102924784",
"",
"2069997605",
"",
"2962780596",
"1992187125",
"2103682126",
"2798896170",
"2599226450",
"2301937176",
"2608696995",
"",
"2474891805",
"2806379360",
"2038891881"
],
"abstract": [
"This paper presents an automatic and robust approach that accurately captures high-quality 3D facial performances using a single RGBD camera. The key of our approach is to combine the power of automatic facial feature detection and image-based 3D nonrigid registration techniques for 3D facial reconstruction. In particular, we develop a robust and accurate image-based nonrigid registration algorithm that incrementally deforms a 3D template mesh model to best match observed depth image data and important facial features detected from single RGBD images. The whole process is fully automatic and robust because it is based on single frame facial registration framework. The system is flexible because it does not require any strong 3D facial priors such as blend shape models. We demonstrate the power of our approach by capturing a wide range of 3D facial expressions using a single RGBD camera and achieve state-of-the-art accuracy by comparing against alternative methods.",
"This paper describes a passive stereo system for capturing the 3D geometry of a face in a single-shot under standard light sources. The system is low-cost and easy to deploy. Results are submillimeter accurate and commensurate with those from state-of-the-art systems based on active lighting, and the models meet the quality requirements of a demanding domain like the movie industry. Recovered models are shown for captures from both high-end cameras in a studio setting and from a consumer binocular-stereo camera, demonstrating scalability across a spectrum of camera deployments, and showing the potential for 3D face modeling to move beyond the professional arena and into the emerging consumer market in stereoscopic photography. Our primary technical contribution is a modification of standard stereo refinement methods to capture pore-scale geometry, using a qualitative approach that produces visually realistic results. The second technical contribution is a calibration method suited to face capture systems. The systemic contribution includes multiple demonstrations of system robustness and quality. These include capture in a studio setup, capture off a consumer binocular-stereo camera, scanning of faces of varying gender and ethnicity and age, capture of highly-transient facial expression, and scanning a physical mask to provide ground-truth validation.",
"",
"We present a new technique for passive and markerless facial performance capture based on anchor frames. Our method starts with high resolution per-frame geometry acquisition using state-of-the-art stereo reconstruction, and proceeds to establish a single triangle mesh that is propagated through the entire performance. Leveraging the fact that facial performances often contain repetitive subsequences, we identify anchor frames as those which contain similar facial expressions to a manually chosen reference expression. Anchor frames are automatically computed over one or even multiple performances. We introduce a robust image-space tracking method that computes pixel matches directly from the reference frame to all anchor frames, and thereby to the remaining frames in the sequence via sequential matching. This allows us to propagate one reconstructed frame to an entire sequence in parallel, in contrast to previous sequential methods. Our anchored reconstruction approach also limits tracker drift and robustly handles occlusions and motion blur. The parallel tracking and mesh propagation offer low computation times. Our technique will even automatically match anchor frames across different sequences captured on different occasions, propagating a single mesh to all performances.",
"",
"It has been recently shown that neural networks can recover the geometric structure of a face from a single given image. A common denominator of most existing face geometry reconstruction methods is the restriction of the solution space to some low-dimensional subspace. While such a model significantly simplifies the reconstruction problem, it is inherently limited in its expressiveness. As an alternative, we propose an Image-to-Image translation network that jointly maps the input image to a depth image and a facial correspondence map. This explicit pixel-based mapping can then be utilized to provide high quality reconstructions of diverse faces under extreme expressions, using a purely geometric refinement process. In the spirit of recent approaches, the network is trained only with synthetic data, and is then evaluated on “in-the-wild” facial images. Both qualitative and quantitative analyses demonstrate the accuracy and the robustness of our approach.",
"This paper presents a system for performance-based character animation that enables any user to control the facial expressions of a digital avatar in realtime. The user is recorded in a natural environment using a non-intrusive, commercially available 3D sensor. The simplicity of this acquisition device comes at the cost of high noise levels in the acquired data. To effectively map low-quality 2D images and 3D depth maps to realistic facial expressions, we introduce a novel face tracking algorithm that combines geometry and texture registration with pre-recorded animation priors in a single optimization. Formulated as a maximum a posteriori estimation in a reduced parameter space, our method implicitly exploits temporal coherence to stabilize the tracking. We demonstrate that compelling 3D facial dynamics can be reconstructed in realtime without the use of face markers, intrusive lighting, or complex scanning hardware. This makes our system easy to deploy and facilitates a range of new applications, e.g. in digital gameplay or social interactions.",
"Given video footage of a person's face, we present new techniques to automatically recover the face position and the facial expression from each frame in the video sequence. A 3D face model is fitted to each frame using a continuous optimization technique. Our model is based on a set of 3D face models that are linearly combined using 3D morphing. Our method has the advantages over previous techniques of fitting directly a realistic 3-dimensional face model and of recovering parameters that can be used directly in an animation system. We also explore many applications, including performance-driven animation (applying the recovered position and expression of the face to a synthetic character to produce an animation that mimics the input video), relighting the face, varying the camera position, and adding facial ornaments such as tattoos and scars.",
"We introduce InverseFaceNet, a deep convolutional inverse rendering framework for faces that jointly estimates facial pose, shape, expression, reflectance and illumination from a single input image. By estimating all parameters from just a single image, advanced editing possibilities on a single face image, such as appearance editing and relighting, become feasible in real time. Most previous learning-based face reconstruction approaches do not jointly recover all dimensions, or are severely limited in terms of visual quality. In contrast, we propose to recover high-quality facial pose, shape, expression, reflectance and illumination using a deep neural network that is trained using a large, synthetically created training corpus. Our approach builds on a novel loss function that measures model-space similarity directly in parameter space and significantly improves reconstruction accuracy. We further propose a self-supervised bootstrapping process in the network training loop, which iteratively updates the synthetic training corpus to better reflect the distribution of real-world imagery. We demonstrate that this strategy outperforms completely synthetically trained networks. Finally, we show high-quality reconstructions and compare our approach to several state-of-the-art approaches.",
"3D face reconstruction is a fundamental Computer Vision problem of extraordinary difficulty. Current systems often assume the availability of multiple facial images (sometimes from the same subject) as input, and must address a number of methodological challenges such as establishing dense correspondences across large facial poses, expressions, and non-uniform illumination. In general these methods require complex and inefficient pipelines for model building and fitting. In this work, we propose to address many of these limitations by training a Convolutional Neural Network (CNN) on an appropriate dataset consisting of 2D images and 3D facial models or scans. Our CNN works with just a single 2D facial image, does not require accurate alignment nor establishes dense correspondence between images, works for arbitrary facial poses and expressions, and can be used to reconstruct the whole 3D facial geometry (including the non-visible parts of the face) bypassing the construction (during training) and fitting (during testing) of a 3D Morphable Model. We achieve this via a simple CNN architecture that performs direct regression of a volumetric representation of the 3D facial geometry from a single 2D image. We also demonstrate how the related task of facial landmark localization can be incorporated into the proposed framework and help improve reconstruction quality, especially for the cases of large poses and facial expressions. Code and models will be made available at http: aaronsplace.co.uk",
"We present a novel approach for real-time facial reenactment of a monocular target video sequence (e.g., Youtube video). The source sequence is also a monocular video stream, captured live with a commodity webcam. Our goal is to animate the facial expressions of the target video by a source actor and re-render the manipulated output video in a photo-realistic fashion. To this end, we first address the under-constrained problem of facial identity recovery from monocular video by non-rigid model-based bundling. At run time, we track facial expressions of both source and target video using a dense photometric consistency measure. Reenactment is then achieved by fast and efficient deformation transfer between source and target. The mouth interior that best matches the re-targeted expression is retrieved from the target sequence and warped to produce an accurate fit. Finally, we convincingly re-render the synthesized target face on top of the corresponding video stream such that it seamlessly blends with the real-world illumination. We demonstrate our method in a live setup, where Youtube videos are reenacted in real time.",
"We present a multi-view stereo reconstruction technique that directly produces a complete high-fidelity head model with consistent facial mesh topology. While existing techniques decouple shape estimation and facial tracking, our framework jointly optimizes for stereo constraints and consistent mesh parameterization. Our method is therefore free from drift and fully parallelizable for dynamic facial performance capture. We produce highly detailed facial geometries with artist-quality UV parameterization, including secondary elements such as eyeballs, mouth pockets, nostrils, and the back of the head. Our approach consists of deforming a common template model to match multi-view input images of the subject, while satisfying cross-view, cross-subject, and cross-pose consistencies using a combination of 2D landmark detection, optical flow, and surface and volumetric Laplacian regularization. Since the flow is never computed between frames, our method is trivially parallelized by processing each frame independently. Accurate rigid head pose is extracted using a PCA-based dimension reduction and denoising scheme. We demonstrate high-fidelity performance capture results with challenging head motion and complex facial expressions around eye and mouth regions. While the quality of our results is on par with the current state-of-the-art, our approach can be fully parallelized, does not suffer from drift, and produces face models with production-quality mesh topologies.",
"",
"We present a new anatomically-constrained local face model and fitting approach for tracking 3D faces from 2D motion data in very high quality. In contrast to traditional global face models, often built from a large set of blendshapes, we propose a local deformation model composed of many small subspaces spatially distributed over the face. Our local model offers far more flexibility and expressiveness than global blendshape models, even with a much smaller model size. This flexibility would typically come at the cost of reduced robustness, in particular during the under-constrained task of monocular reconstruction. However, a key contribution of this work is that we consider the face anatomy and introduce subspace skin thickness constraints into our model, which constrain the face to only valid expressions and helps counteract depth ambiguities in monocular tracking. Given our new model, we present a novel fitting optimization that allows 3D facial performance reconstruction from a single view at extremely high quality, far beyond previous fitting approaches. Our model is flexible, and can be applied also when only sparse motion data is available, for example with marker-based motion capture or even face posing from artistic sketches. Furthermore, by incorporating anatomical constraints we can automatically estimate the rigid motion of the skull, obtaining a rigid stabilization of the performance for free. We demonstrate our model and single-view fitting method on a number of examples, including, for the first time, extreme local skin deformation caused by external forces such as wind, captured from a single high-speed camera.",
"",
"We present a real-time performance-driven facial animation system based on 3D shape regression. In this system, the 3D positions of facial landmark points are inferred by a regressor from 2D video frames of an ordinary web camera. From these 3D points, the pose and expressions of the face are recovered by fitting a user-specific blendshape model to them. The main technical contribution of this work is the 3D regression algorithm that learns an accurate, user-specific face alignment model from an easily acquired set of training data, generated from images of the user performing a sequence of predefined facial poses and expressions. Experiments show that our system can accurately recover 3D face shapes even for fast motions, non-frontal faces, and exaggerated expressions. In addition, some capacity to handle partial occlusions and changing lighting conditions is demonstrated."
]
} |
1812.02836 | 2905529320 | Muscle-based systems have the potential to provide both anatomical accuracy and semantic interpretability as compared to blendshape models; however, a lack of expressivity and differentiability has limited their impact. Thus, we propose modifying a recently developed rather expressive muscle-based system in order to make it fully-differentiable; in fact, our proposed modifications allow this physically robust and anatomically accurate muscle model to conveniently be driven by an underlying blendshape basis. Our formulation is intuitive, natural, as well as monolithically and fully coupled such that one can differentiate the model from end to end, which makes it viable for both optimization and learning-based approaches for a variety of applications. We illustrate this with a number of examples including both shape matching of three-dimensional geometry as as well as the automatic determination of a three-dimensional facial pose from a single two-dimensional RGB image without using markers or depth information. | @cite_0 was one of the first to utilize quasistatic simulations to drive the deformation of a 3D face, especially for motion capture. There has also been interest in using quasistative simulations to drive muscle deformations in the body @cite_22 @cite_8 @cite_28 . However, in general, facial muscle simulations tend to be less expressive than their artist-driven blendshape counterparts. More recently, significant work has been done to make muscle simulations more expressive @cite_35 @cite_2 . While these methods can be used to target data in the form of geometry, it is unclear how to cleanly transfer these methods to target non-geometry data such as two-dimensional RGB images. Other work has been done to try to introduce physical simulations into the blendshape models themselves @cite_39 @cite_12 @cite_47 @cite_4 ; however, these works do not focus on the inverse problem. | {
"cite_N": [
"@cite_35",
"@cite_4",
"@cite_22",
"@cite_8",
"@cite_28",
"@cite_39",
"@cite_0",
"@cite_2",
"@cite_47",
"@cite_12"
],
"mid": [
"2460468435",
"2617644139",
"1980757084",
"2143345877",
"2114106563",
"2807843367",
"2022890264",
"",
"",
"2401751489"
],
"abstract": [
"We propose a new framework for the simulation of facial muscle and flesh that so significantly improves the technique that it allows for immediate mainstream use of anatomically and biomechanically accurate muscle models as a bread and butter technique in a high-end production quality pipeline. The key idea is to create a blendshape system for the muscles that gives the precise directability and controllability required in a high-end production environment. The blendshape muscles are used to drive the underlying anatomically and biomechanically motivated simulation in a way that is unbound by the typical restrictions of a simulation system while still retaining the desirable degree of freedom richness that leads to high quality results. We show that we are able to target production quality facial shapes, whether from scans or an animation system, and illustrate that the resulting nonlinear simulation in-betweens are of higher quality than those obtained from traditional linear blendshapes. We also demonstrate the ability to selectively improve areas on a given blendshape using the results of a simulation, as well as the ability to edit muscle shapes and paths in order to produce directability for animator control. Then, we show how these techniques can be used to transition from one blendshape to another or even track and selectively modify an entire performance. The efficacy of our system is further demonstrated by using it to retarget animation onto new creature models given only a single static rest pose as input.",
"Oftentimes facial animation is created separately from overall body motion. Since convincing facial animation is challenging enough in itself, artists tend to create and edit the face motion in isolation. Or if the face animation is derived from motion capture, this is typically performed in a mo-cap booth while sitting relatively still. In either case, recombining the isolated face animation with body and head motion is non-trivial and often results in an uncanny result if the body dynamics are not properly reflected on the face e.g. the bouncing of facial tissue when running.",
"We present an algorithm for the finite element simulation of elastoplastic solids which is capable of robustly and efficiently handling arbitrarily large deformation. In fact, our model remains valid even when large parts of the mesh are inverted. The algorithm is straightforward to implement and can be used with any material constitutive model, and for both volumetric solids and thin shells such as cloth. We also provide a mechanism for controlling plastic deformation, which allows a deformable object to be guided towards a desired final shape without sacrificing realistic behavior. Finally, we present an improved method for rigid body collision handling in the context of mixed explicit implicit time-stepping.",
"Since it relies on a geometrical rather than a variational framework, many find the finite volume method (FVM) more intuitive than the finite element method (FEM). We show that the FVM allows one to interpret the stress inside a tetrahedron as a simple \"multidimensional force\" pushing on each face. Moreover, this interpretation leads to a heuristic method for calculating the force on each node, which is as simple to implement and comprehend as masses and springs. In the finite volume spirit, we also present a geometric rather than interpolating function definition of strain. We use the FVM and a quasi-incompressible, transversely isotropic, hyperelastic constitutive model to simulate contracting muscle tissue. B-spline solids are used to model fiber directions, and the muscle activation levels are derived from key frame animations.",
"Simulation of the musculoskelet al system has important applications in biomechanics, biomedical engineering, surgery simulation, and computer graphics. The accuracy of the muscle, bone, and tendon geometry as well as the accuracy of muscle and tendon dynamic deformation are of paramount importance in all these applications. We present a framework for extracting and simulating high resolution musculoskelet al geometry from the segmented visible human data set. We simulate 30 contact collision coupled muscles in the upper limb and describe a computationally tractable implementation using an embedded mesh framework. Muscle geometry is embedded in a nonmanifold, connectivity preserving simulation mesh molded out of a lower resolution BCC lattice containing identical, well-shaped elements, leading to a relaxed time step restriction for stability and, thus, reduced computational cost. The muscles are endowed with a transversely isotropic, quasiincompressible constitutive model that incorporates muscle fiber fields as well as passive and active components. The simulation takes advantage of a new robust finite element technique that handles both degenerate and inverted tetrahedra.",
"",
"We built an anatomically accurate model of facial musculature, passive tissue and underlying skelet al structure using volumetric data acquired from a living male subject. The tissues are endowed with a highly nonlinear constitutive model including controllable anisotropic muscle activations based on fiber directions. Detailed models of this sort can be difficult to animate requiring complex coordinated stimulation of the underlying musculature. We propose a solution to this problem automatically determining muscle activations that track a sparse set of surface landmarks, e.g. acquired from motion capture marker data. Since the resulting animation is obtained via a three dimensional nonlinear finite element method, we obtain visually plausible and anatomically correct deformations with spatial and temporal coherence that provides robustness against outliers in the motion capture data. Moreover, the obtained muscle activations can be used in a robust simulation framework including contact and collision of the face with external objects.",
"",
"",
"In this paper we present a new paradigm for the generation and retargeting of facial animation. Like a vast majority of the approaches that have adressed these topics, our formalism is built on blendshapes. However, where prior works have generally encoded facial geometry using a low dimensional basis of these blendshapes, we propose to encode facial dynamics by looking at blendshapes as a basis of forces rather than a basis of shapes. We develop this idea into a dynamic model that naturally combines the blendshapes paradigm with physics-based techniques for the simulation of deforming meshes. Because it escapes the linear span of the shape basis through time-integration and physics-inspired simulation, this approach has a wider expressive range than previous blendshape-based methods. Its inherent physically-based formulation also enables the simulation of more advanced physical interactions, such as collision responses on lip contacts."
]
} |
1812.02850 | 2905034002 | It is a widely accepted principle that software without tests has bugs. Testing reinforcement learning agents is especially difficult because of the stochastic nature of both agents and environments, the complexity of state-of-the-art models, and the sequential nature of their predictions. Recently, the Arcade Learning Environment (ALE) has become one of the most widely used benchmark suites for deep learning research, and state-of-the-art Reinforcement Learning (RL) agents have been shown to routinely equal or exceed human performance on many ALE tasks. Since ALE is based on emulation of original Atari games, the environment does not provide semantically meaningful representations of internal game state. This means that ALE has limited utility as an environment for supporting testing or model introspection. We propose ToyBox, a collection of reimplementations of these games that solves this critical problem and enables robust testing of RL agents. | Reinforcement learning is a subdomain of machine learning wherein an interacts with an via sequential , seeking to maximize its expected cumulative . The agent operates in the environment by taking actions to transition between environment @cite_7 . | {
"cite_N": [
"@cite_7"
],
"mid": [
"2121863487"
],
"abstract": [
"Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning."
]
} |
1812.02850 | 2905034002 | It is a widely accepted principle that software without tests has bugs. Testing reinforcement learning agents is especially difficult because of the stochastic nature of both agents and environments, the complexity of state-of-the-art models, and the sequential nature of their predictions. Recently, the Arcade Learning Environment (ALE) has become one of the most widely used benchmark suites for deep learning research, and state-of-the-art Reinforcement Learning (RL) agents have been shown to routinely equal or exceed human performance on many ALE tasks. Since ALE is based on emulation of original Atari games, the environment does not provide semantically meaningful representations of internal game state. This means that ALE has limited utility as an environment for supporting testing or model introspection. We propose ToyBox, a collection of reimplementations of these games that solves this critical problem and enables robust testing of RL agents. | Historically, RL researchers have created their own environments to test the behavior of RL algorithms. Environments drawn from the classic GridWorld family consist of an agent moving through a grid of states from a fixed start state toward a fixed goal state, typically with one or more obstacles or other complicating factors (e.g., teleporting states) @cite_7 . In this simplified world, the optimal policy is directly computable via dynamic programming. | {
"cite_N": [
"@cite_7"
],
"mid": [
"2121863487"
],
"abstract": [
"Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning."
]
} |
1812.02850 | 2905034002 | It is a widely accepted principle that software without tests has bugs. Testing reinforcement learning agents is especially difficult because of the stochastic nature of both agents and environments, the complexity of state-of-the-art models, and the sequential nature of their predictions. Recently, the Arcade Learning Environment (ALE) has become one of the most widely used benchmark suites for deep learning research, and state-of-the-art Reinforcement Learning (RL) agents have been shown to routinely equal or exceed human performance on many ALE tasks. Since ALE is based on emulation of original Atari games, the environment does not provide semantically meaningful representations of internal game state. This means that ALE has limited utility as an environment for supporting testing or model introspection. We propose ToyBox, a collection of reimplementations of these games that solves this critical problem and enables robust testing of RL agents. | While there has certainly been criticism of the field for testing in the training environment @cite_3 , there is a legitimate justification for doing so: if an agent is only to be deployed in the training environment, it is perfectly reasonable to overfit the agent's policy to that environment. However, claims about higher-level reasoning or generalization capabilities require testing, where testing means something more than simply running a trained agent in the environment. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2104228245"
],
"abstract": [
"Empirical evaluations play an important role in machine learning. However, the usefulness of any evaluation depends on the empirical methodology employed. Designing good empirical methodologies is difficult in part because agents can overfit test evaluations and thereby obtain misleadingly high scores. We argue that reinforcement learning is particularly vulnerable to environment overfitting and propose as a remedy generalized methodologies, in which evaluations are based on multiple environments sampled from a distribution. In addition, we consider how to summarize performance when scores from different environments may not have commensurate values. Finally, we present proof-of-concept results demonstrating how these methodologies can validate an intuitively useful range-adaptive tile coding method."
]
} |
1812.02784 | 2904190302 | We propose a new task, called Story Visualization. Given a multi-sentence paragraph, the story is visualized by generating a sequence of images, one for each sentence. In contrast to video generation, story visualization focuses less on the continuity in generated images (frames), but more on the global consistency across dynamic scenes and characters -- a challenge that has not been addressed by any single-image or video generation methods. We therefore propose a new story-to-image-sequence generation model, StoryGAN, based on the sequential conditional GAN framework. Our model is unique in that it consists of a deep Context Encoder that dynamically tracks the story flow, and two discriminators at the story and image levels, to enhance the image quality and the consistency of the generated sequences. To evaluate the model, we modified existing datasets to create the CLEVR-SV and Pororo-SV datasets. Empirically, StoryGAN outperforms state-of-the-art models in image quality, contextual consistency metrics, and human evaluation. | Variational AutoEncoders (VAEs) @cite_31 , Generative Adversarial Nets (GANs) @cite_9 , and flow-based generative models @cite_27 @cite_37 ) have been widely applied to a wide range of generation tasks including text-to-image generation, video generation, style transfer, and image editing. Story visualization falls into this broad categorization of generative tasks, but has several distinct aspects. | {
"cite_N": [
"@cite_37",
"@cite_9",
"@cite_31",
"@cite_27"
],
"mid": [
"2409550820",
"2099471712",
"",
"1583912456"
],
"abstract": [
"Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically, designing models with tractable learning, sampling, inference and evaluation is crucial in solving this task. We extend the space of such models using real-valued non-volume preserving (real NVP) transformations, a set of powerful invertible and learnable transformations, resulting in an unsupervised learning algorithm with exact log-likelihood computation, exact sampling, exact inference of latent variables, and an interpretable latent space. We demonstrate its ability to model natural images on four datasets through sampling, log-likelihood evaluation and latent variable manipulations.",
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.",
"",
"We propose a deep learning framework for modeling complex high-dimensional densities called Non-linear Independent Component Estimation (NICE). It is based on the idea that a good representation is one in which the data has a distribution that is easy to model. For this purpose, a non-linear deterministic transformation of the data is learned that maps it to a latent space so as to make the transformed data conform to a factorized distribution, i.e., resulting in independent latent variables. We parametrize this transformation so that computing the Jacobian determinant and inverse transform is trivial, yet we maintain the ability to learn complex non-linear transformations, via a composition of simple building blocks, each based on a deep neural network. The training criterion is simply the exact log-likelihood, which is tractable. Unbiased ancestral sampling is also easy. We show that this approach yields good generative models on four image datasets and can be used for inpainting."
]
} |
1812.02784 | 2904190302 | We propose a new task, called Story Visualization. Given a multi-sentence paragraph, the story is visualized by generating a sequence of images, one for each sentence. In contrast to video generation, story visualization focuses less on the continuity in generated images (frames), but more on the global consistency across dynamic scenes and characters -- a challenge that has not been addressed by any single-image or video generation methods. We therefore propose a new story-to-image-sequence generation model, StoryGAN, based on the sequential conditional GAN framework. Our model is unique in that it consists of a deep Context Encoder that dynamically tracks the story flow, and two discriminators at the story and image levels, to enhance the image quality and the consistency of the generated sequences. To evaluate the model, we modified existing datasets to create the CLEVR-SV and Pororo-SV datasets. Empirically, StoryGAN outperforms state-of-the-art models in image quality, contextual consistency metrics, and human evaluation. | Very relevant for story visualization task is conditional text-to-image transformation @cite_24 @cite_21 @cite_32 @cite_6 , which can now generate high-resolution realistic images @cite_1 @cite_22 . A key task in text-to-image generation is to understand longer and more complex input texts. For example, this has been explored in dialogue-to-image generation, where the input is a complete dialogue session rather than a single sentence @cite_16 . Another related task is that of textual image editing, which edits an input image according to a textual editing query @cite_28 @cite_4 . This task requires consistency between the original image and the output image. Finally, there is the task of placing pre-specified images and objects in a picture from a text description @cite_10 . This task also relates text to a consistent image, but does not require a full image generation procedure. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_28",
"@cite_21",
"@cite_1",
"@cite_32",
"@cite_6",
"@cite_24",
"@cite_16",
"@cite_10"
],
"mid": [
"2950862326",
"2963966654",
"2963145877",
"2963073614",
"2964024144",
"",
"2963567641",
"2405756170",
"2788768663",
"2774015909"
],
"abstract": [
"While great progress has been made recently in automatic image manipulation, it has been limited to object centric images like faces or structured scene datasets. In this work, we take a step towards general scene-level image editing by developing an automatic interaction-free object removal model. Our model learns to find and remove objects from general scene images using image-level labels and unpaired data in a generative adversarial network (GAN) framework. We achieve this with two key contributions: a two-stage editor architecture consisting of a mask generator and image in-painter that co-operate to remove objects, and a novel GAN based prior for the mask generator that allows us to flexibly incorporate knowledge about object shapes. We experimentally show on two datasets that our method effectively removes a wide variety of objects using weak supervision only",
"In this paper, we propose an Attentional Generative Adversarial Network (AttnGAN) that allows attention-driven, multi-stage refinement for fine-grained text-to-image generation. With a novel attentional generative network, the AttnGAN can synthesize fine-grained details at different sub-regions of the image by paying attentions to the relevant words in the natural language description. In addition, a deep attentional multimodal similarity model is proposed to compute a fine-grained image-text matching loss for training the generator. The proposed AttnGAN significantly outperforms the previous state of the art, boosting the best reported inception score by 14.14 on the CUB dataset and 170.25 on the more challenging COCO dataset. A detailed analysis is also performed by visualizing the attention layers of the AttnGAN. It for the first time shows that the layered attentional GAN is able to automatically select the condition at the word level for generating different parts of the image.",
"We investigate the problem of Language-Based Image Editing (LBIE). Given a source image and a natural language description, we want to generate a target image by editing the source image based on the description. We propose a generic modeling framework for two subtasks of LBIE: language-based image segmentation and image colorization. The framework uses recurrent attentive models to fuse image and language features. Instead of using a fixed step size, we introduce for each region of the image a termination gate to dynamically determine after each inference step whether to continue extrapolating additional information from the textual description. The effectiveness of the framework is validated on three datasets. First, we introduce a synthetic dataset, called CoSaL, to evaluate the end-to-end performance of our LBIE system. Second, we show that the framework leads to state-of-the-art performance on image segmentation on the ReferIt dataset. Third, we present the first language-based colorization result on the Oxford-102 Flowers dataset.",
"We investigate conditional adversarial networks as a general-purpose solution to image-to-image translation problems. These networks not only learn the mapping from input image to output image, but also learn a loss function to train this mapping. This makes it possible to apply the same generic approach to problems that traditionally would require very different loss formulations. We demonstrate that this approach is effective at synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images, among other tasks. Moreover, since the release of the pix2pix software associated with this paper, hundreds of twitter users have posted their own artistic experiments using our system. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without handengineering our loss functions either.",
"Synthesizing high-quality images from text descriptions is a challenging problem in computer vision and has many practical applications. Samples generated by existing textto- image approaches can roughly reflect the meaning of the given descriptions, but they fail to contain necessary details and vivid object parts. In this paper, we propose Stacked Generative Adversarial Networks (StackGAN) to generate 256.256 photo-realistic images conditioned on text descriptions. We decompose the hard problem into more manageable sub-problems through a sketch-refinement process. The Stage-I GAN sketches the primitive shape and colors of the object based on the given text description, yielding Stage-I low-resolution images. The Stage-II GAN takes Stage-I results and text descriptions as inputs, and generates high-resolution images with photo-realistic details. It is able to rectify defects in Stage-I results and add compelling details with the refinement process. To improve the diversity of the synthesized images and stabilize the training of the conditional-GAN, we introduce a novel Conditioning Augmentation technique that encourages smoothness in the latent conditioning manifold. Extensive experiments and comparisons with state-of-the-arts on benchmark datasets demonstrate that the proposed method achieves significant improvements on generating photo-realistic images conditioned on text descriptions.",
"",
"This paper investigates a novel problem of generating images from visual attributes. We model the image as a composite of foreground and background and develop a layered generative model with disentangled latent variables that can be learned end-to-end using a variational auto-encoder. We experiment with natural images of faces and birds and demonstrate that the proposed models are capable of generating realistic and diverse samples with disentangled latent representations. We use a general energy minimization algorithm for posterior inference of latent variables given novel images. Therefore, the learned generative models show excellent quantitative and visual results in the tasks of attribute-conditioned image reconstruction and completion.",
"Automatic synthesis of realistic images from text would be interesting and useful, but current AI systems are still far from this goal. However, in recent years generic and powerful recurrent neural network architectures have been developed to learn discriminative text feature representations. Meanwhile, deep convolutional generative adversarial networks (GANs) have begun to generate highly compelling images of specific categories, such as faces, album covers, and room interiors. In this work, we develop a novel deep architecture and GAN formulation to effectively bridge these advances in text and image modeling, translating visual concepts from characters to pixels. We demonstrate the capability of our model to generate plausible images of birds and flowers from detailed text descriptions.",
"Synthesizing realistic images from text descriptions on a dataset like Microsoft Common Objects in Context (MS COCO), where each image can contain several objects, is a challenging task. Prior work has used text captions to generate images. However, captions might not be informative enough to capture the entire image and insufficient for the model to be able to understand which objects in the images correspond to which words in the captions. We show that adding a dialogue that further describes the scene leads to significant improvement in the inception score and in the quality of generated images on the MS COCO dataset.",
"In this work, we propose a goal-driven collaborative task that contains vision, language, and action in a virtual environment as its core components. Specifically, we develop a collaborative Image Drawing' game between two agents, called CoDraw. Our game is grounded in a virtual world that contains movable clip art objects. Two players, Teller and Drawer, are involved. The Teller sees an abstract scene containing multiple clip arts in a semantically meaningful configuration, while the Drawer tries to reconstruct the scene on an empty canvas using available clip arts. The two players communicate via two-way communication using natural language. We collect the CoDraw dataset of 10K dialogs consisting of 138K messages exchanged between a Teller and a Drawer from Amazon Mechanical Turk (AMT). We analyze our dataset and present three models to model the players' behaviors, including an attention model to describe and draw multiple clip arts at each round. The attention models are quantitatively compared to the other models to show how the conventional approaches work for this new task. We also present qualitative visualizations."
]
} |
1812.02784 | 2904190302 | We propose a new task, called Story Visualization. Given a multi-sentence paragraph, the story is visualized by generating a sequence of images, one for each sentence. In contrast to video generation, story visualization focuses less on the continuity in generated images (frames), but more on the global consistency across dynamic scenes and characters -- a challenge that has not been addressed by any single-image or video generation methods. We therefore propose a new story-to-image-sequence generation model, StoryGAN, based on the sequential conditional GAN framework. Our model is unique in that it consists of a deep Context Encoder that dynamically tracks the story flow, and two discriminators at the story and image levels, to enhance the image quality and the consistency of the generated sequences. To evaluate the model, we modified existing datasets to create the CLEVR-SV and Pororo-SV datasets. Empirically, StoryGAN outperforms state-of-the-art models in image quality, contextual consistency metrics, and human evaluation. | A second closely related task to story visualization is that of video generation, especially that of text-to-video @cite_7 @cite_25 or image-to-video generation @cite_11 @cite_0 @cite_30 . Existing approaches only generate short video clips @cite_25 @cite_12 @cite_17 without scene changes. The biggest challenge in video generation is how to ensure a smooth motion transition across successive video frames. To this end, researchers disentangle dynamic and static features for motion and background, respectively @cite_30 @cite_26 @cite_0 @cite_2 . In our modeling of story visualization, the whole story sets the static features and each input sentence encodes dynamic features. However, there are several differences: @math conditional video generation has only one input, while our task has a sequential, evolving input; and @math the motion in video clips is continuous in a single scene, while the generated images that visualizes a story are discrete and often represent different scenes. For the form of the conditional information, it is not limited to text. Trajectory, skeleton or simple landmark based video generation has been studied as in several existing works @cite_17 @cite_29 @cite_15 . | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_7",
"@cite_29",
"@cite_17",
"@cite_0",
"@cite_2",
"@cite_15",
"@cite_25",
"@cite_12",
"@cite_11"
],
"mid": [
"2520707650",
"",
"2962713165",
"",
"2799055999",
"2963092440",
"2963045453",
"",
"2791437494",
"2964327849",
"2769102608"
],
"abstract": [
"We capitalize on large amounts of unlabeled video in order to learn a model of scene dynamics for both video recognition tasks (e.g. action classification) and video generation tasks (e.g. future prediction). We propose a generative adversarial network for video with a spatio-temporal convolutional architecture that untangles the scene's foreground from the background. Experiments suggest this model can generate tiny videos up to a second at full frame rate better than simple baselines, and we show its utility at predicting plausible futures of static images. Moreover, experiments and visualizations show the model internally learns useful features for recognizing actions with minimal supervision, suggesting scene dynamics are a promising signal for representation learning. We believe generative video models can impact many applications in video understanding and simulation.",
"",
"",
"",
"Video generation and manipulation is an important yet challenging task in computer vision. Existing methods usually lack ways to explicitly control the synthesized motion. In this work, we present a conditional video generation model that allows detailed control over the motion of the generated video. Given the first frame and sparse motion trajectories specified by users, our model can synthesize a video with corresponding appearance and motion. We propose to combine the advantage of copying pixels from the given frame and hallucinating the lightness difference from scratch which help generate sharp video while keeping the model robust to occlusion and lightness change. We also propose a training paradigm that calculate trajectories from video clips, which eliminated the need of annotated training data. Experiments on several standard benchmarks demonstrate that our approach can generate realistic videos comparable to state-of-the-art video generation and video prediction methods while the motion of the generated videos can correspond well with user input.",
"Visual signals in a video can be divided into content and motion. While content specifies which objects are in the video, motion describes their dynamics. Based on this prior, we propose the Motion and Content decomposed Generative Adversarial Network (MoCoGAN) framework for video generation. The proposed framework generates a video by mapping a sequence of random vectors to a sequence of video frames. Each random vector consists of a content part and a motion part. While the content part is kept fixed, the motion part is realized as a stochastic process. To learn motion and content decomposition in an unsupervised manner, we introduce a novel adversarial learning scheme utilizing both image and video discriminators. Extensive experimental results on several challenging datasets with qualitative and quantitative comparison to the state-of-the-art approaches, verify effectiveness of the proposed framework. In addition, we show that MoCoGAN allows one to generate videos with same content but different motion as well as videos with different content and same motion. Our code is available at https: github.com sergeytulyakov mocogan.",
"We present a new model DRNET that learns disentangled image representations from video. Our approach leverages the temporal coherence of video and a novel adversarial loss to learn a representation that factorizes each frame into a stationary part and a temporally varying component. The disentangled representation can be used for a range of tasks. For example, applying a standard LSTM to the time-vary components enables prediction of future frames. We evaluating our approach on a range of synthetic and real videos. For the latter, we demonstrate the ability to coherently generate up to several hundred steps into the future.",
"",
"Videos express highly structured spatio-temporal patterns of visual data. A video can be thought of as being governed by two factors: (i) temporally invariant (e.g., person identity), or slowly varying (e.g., activity), attribute-induced appearance, encoding the persistent content of each frame, and (ii) an inter-frame motion or scene dynamics (e.g., encoding evolution of the person ex-ecuting the action). Based on this intuition, we propose a generative framework for video generation and future prediction. The proposed framework generates a video (short clip) by decoding samples sequentially drawn from a latent space distribution into full video frames. Variational Autoencoders (VAEs) are used as a means of encoding decoding frames into from the latent space and RNN as a wayto model the dynamics in the latent space. We improve the video generation consistency through temporally-conditional sampling and quality by structuring the latent space with attribute controls; ensuring that attributes can be both inferred and conditioned on during learning generation. As a result, given attributes and orthe first frame, our model is able to generate diverse but highly consistent sets ofvideo sequences, accounting for the inherent uncertainty in the prediction task. Experimental results on Chair CAD, Weizmann Human Action, and MIT-Flickr datasets, along with detailed comparison to the state-of-the-art, verify effectiveness of the framework.",
"",
"Current video generation prediction completion results are limited, due to the severe ill-posedness inherent in these three problems. In this paper, we focus on human action videos, and propose a general, two-stage deep framework to generate human action videos with no constraints or arbitrary number of constraints, which uniformly addresses the three problems: video generation given no input frames, video prediction given the first few frames, and video completion given the first and last frames. To solve video generation from scratch, we build a two-stage framework where we first train a deep generative model that generates human pose sequences from random noise, and then train a skeleton-to-image network to synthesize human action videos given the human pose sequences generated. To solve video prediction and completion, we exploit our trained model and conduct optimization over the latent space to generate videos that best suit the given input frame constraints. With our novel method, we sidestep the original ill-posed problems and produce for the first time high-quality video generation prediction completion results of much longer duration. We present quantitative and qualitative evaluations to show that our approach outperforms state-of-the-art methods in all three tasks."
]
} |
1812.02784 | 2904190302 | We propose a new task, called Story Visualization. Given a multi-sentence paragraph, the story is visualized by generating a sequence of images, one for each sentence. In contrast to video generation, story visualization focuses less on the continuity in generated images (frames), but more on the global consistency across dynamic scenes and characters -- a challenge that has not been addressed by any single-image or video generation methods. We therefore propose a new story-to-image-sequence generation model, StoryGAN, based on the sequential conditional GAN framework. Our model is unique in that it consists of a deep Context Encoder that dynamically tracks the story flow, and two discriminators at the story and image levels, to enhance the image quality and the consistency of the generated sequences. To evaluate the model, we modified existing datasets to create the CLEVR-SV and Pororo-SV datasets. Empirically, StoryGAN outperforms state-of-the-art models in image quality, contextual consistency metrics, and human evaluation. | There are also several other related tasks in the literature. For example, we can consider story illustration by retrieving images from a pre-collected training set rather than image generation @cite_20 . Cartoon generation has been explored with cut and paste'' technique @cite_8 . However, both of these techniques require large amounts of labeled training data. An inverse task to story visualization is visual storytelling, where the output is a paragraph describing a sequence of input images. Text generation models or reinforcement learning are often highlighted for visual storytelling @cite_23 @cite_19 @cite_13 . | {
"cite_N": [
"@cite_8",
"@cite_19",
"@cite_23",
"@cite_13",
"@cite_20"
],
"mid": [
"2962931774",
"2951684117",
"",
"2803484107",
"2799185021"
],
"abstract": [
"Imagining a scene described in natural language with realistic layout and appearance of entities is the ultimate test of spatial, visual, and semantic world knowledge. Towards this goal, we present the Composition, Retrieval and Fusion Network (Craft), a model capable of learning this knowledge from video-caption data and applying it while generating videos from novel captions. Craft explicitly predicts a temporal-layout of mentioned entities (characters and objects), retrieves spatio-temporal entity segments from a video database and fuses them to generate scene videos. Our contributions include sequential training of components of Craft while jointly modeling layout and appearances, and losses that encourage learning compositional representations for retrieval. We evaluate Craft on semantic fidelity to caption, composition consistency, and visual quality. Craft outperforms direct pixel generation approaches and generalizes well to unseen captions and to unseen video databases with no text annotations. We demonstrate Craft on Flintstones (Flintstones is available at https: prior.allenai.org projects craft), a new richly annotated video-caption dataset with over 25000 videos. For a glimpse of videos generated by Craft, see https: youtu.be 688Vv86n0z8.",
"A natural image usually conveys rich semantic content and can be viewed from different angles. Existing image description methods are largely restricted by small sets of biased visual paragraph annotations, and fail to cover rich underlying semantics. In this paper, we investigate a semi-supervised paragraph generative framework that is able to synthesize diverse and semantically coherent paragraph descriptions by reasoning over local semantic regions and exploiting linguistic knowledge. The proposed Recurrent Topic-Transition Generative Adversarial Network (RTT-GAN) builds an adversarial framework between a structured paragraph generator and multi-level paragraph discriminators. The paragraph generator generates sentences recurrently by incorporating region-based visual and language attention mechanisms at each step. The quality of generated paragraph sentences is assessed by multi-level adversarial discriminators from two aspects, namely, plausibility at sentence level and topic-transition coherence at paragraph level. The joint adversarial training of RTT-GAN drives the model to generate realistic paragraphs with smooth logical transition between sentence topics. Extensive quantitative experiments on image and video paragraph datasets demonstrate the effectiveness of our RTT-GAN in both supervised and semi-supervised settings. Qualitative results on telling diverse stories for an image also verify the interpretability of RTT-GAN.",
"",
"We propose a hierarchically structured reinforcement learning approach to address the challenges of planning for generating coherent multi-sentence stories for the visual storytelling task. Within our framework, the task of generating a story given a sequence of images is divided across a two-level hierarchical decoder. The high-level decoder constructs a plan by generating a semantic concept (i.e., topic) for each image in sequence. The low-level decoder generates a sentence for each image using a semantic compositional network, which effectively grounds the sentence generation conditioned on the topic. The two decoders are jointly trained end-to-end using reinforcement learning. We evaluate our model on the visual storytelling (VIST) dataset. Empirical results from both automatic and human evaluations demonstrate that the proposed hierarchically structured reinforced training achieves significantly better performance compared to a strong flat deep reinforcement learning baseline.",
"We propose an end-to-end network for visual illustration of a sequence of sentences forming a story. At the core of our model is the ability to model the inter-related nature of the sentences within a story, as well as the ability to learn coherence to support reference resolution. The framework takes the form of an encoder-decoder architecture, where sentences are encoded using a hierarchical two-level sentence-story GRU, combined with an encoding of coherence, and sequentially decoded using a predicted feature representation into a consistent illustrative image sequence. We optimize all parameters of our network in an end-to-end fashion with respect to order embedding loss, encoding entailment between images and sentences. Experiments on the VIST storytelling dataset [9] highlight the importance of our algorithmic choices and efficacy of our overall model."
]
} |
1812.02712 | 2903088770 | The online hate speech is proliferating with several organization and countries implementing laws to ban such harmful speech. While these restrictions might reduce the amount of such hateful content, it does so by restricting freedom of speech. Thus, an promising alternative supported by several organizations is to counter such hate speech with more speech. In this paper, We analyze hate speech and the corresponding counters (aka counterspeech) on Twitter. We perform several lexical, linguistic and psycholinguistic analysis on these user accounts and obverse that counter speakers employ several strategies depending on the target community. The hateful accounts express more negative sentiments and are more profane. We also find that the hate tweets by verified accounts have much more virality as compared to a tweet by a non-verified account. While the hate users seem to use words more about envy, hate, negative emotion, swearing terms, ugliness, the counter users use more words related to government, law, leader. We also build a supervised model for classifying the hateful and counterspeech accounts on Twitter and obtain an F-score of 0.77. We also make our dataset public to help advance the research on hate speech. | There are some works which have tried to characterize the hateful users. In @cite_13 @cite_3 , the authors study the user characteristics of hateful accounts on twitter and found that the hateful user accounts differ significantly from normal user accounts on the basis of activity, network centrality, and the type of content they produce. In @cite_1 , the authors perform a comparative study of the hate speech instigators and target users on twitter. They found that the hate instigators target more popular and high profile twitter users, which leads to greater online visibility. In @cite_12 , the authors focus on studying the target of the hatespeech - directed and generalized. They look into linguistic and psycholinguistic properties of these two types of hatespeech and found that while directed hate speech is more personal and directed, informal and express anger, the generalized hate is more of religious type and uses lethal words such as murder', exterminate', and kill'. | {
"cite_N": [
"@cite_1",
"@cite_13",
"@cite_12",
"@cite_3"
],
"mid": [
"2796881724",
"2794585602",
"2963535540",
"2782064370"
],
"abstract": [
"While social media has become an empowering agent to individual voices and freedom of expression, it also facilitates anti-social behaviors including online harassment, cyberbullying, and hate speech. In this paper, we present the first comparative study of hate speech instigators and target users on Twitter. Through a multi-step classification process, we curate a comprehensive hate speech dataset capturing various types of hate. We study the distinctive characteristics of hate instigators and targets in terms of their profile self-presentation, activities, and online visibility. We find that hate instigators target more popular and high profile Twitter users, and that participating in hate speech can result in greater online visibility. We conduct a personality analysis of hate instigators and targets and show that both groups have eccentric personality facets that differ from the general Twitter population. Our results advance the state of the art of understanding online hate speech engagement.",
"Most current approaches to characterize and detect hate speech focus on posted in Online Social Networks. They face shortcomings to collect and annotate hateful speech due to the incompleteness and noisiness of OSN text and the subjectivity of hate speech. These limitations are often aided with constraints that oversimplify the problem, such as considering only tweets containing hate-related words. In this work we partially address these issues by shifting the focus towards . We develop and employ a robust methodology to collect and annotate hateful users which does not depend directly on lexicon and where the users are annotated given their entire profile. This results in a sample of Twitter's retweet graph containing @math users, out of which @math were annotated. We also collect the users who were banned in the three months that followed the data collection. We show that hateful users differ from normal ones in terms of their activity patterns, word usage and as well as network structure. We obtain similar results comparing the neighbors of hateful vs. neighbors of normal users and also suspended users vs. active users, increasing the robustness of our analysis. We observe that hateful users are densely connected, and thus formulate the hate speech detection problem as a task of semi-supervised learning over a graph, exploiting the network of connections on Twitter. We find that a node embedding algorithm, which exploits the graph structure, outperforms content-based approaches for the detection of both hateful ( @math AUC vs @math AUC) and suspended users ( @math AUC vs @math AUC). Altogether, we present a user-centric view of hate speech, paving the way for better detection and understanding of this relevant and challenging issue.",
"",
"Hateful speech in Online Social Networks (OSNs) is a key challenge for companies and governments, as it impacts users and advertisers, and as several countries have strict legislation against the practice. This has motivated work on detecting and characterizing the phenomenon in tweets, social media posts and comments. However, these approaches face several shortcomings due to the noisiness of OSN data, the sparsity of the phenomenon, and the subjectivity of the definition of hate speech. This works presents a user-centric view of hate speech, paving the way for better detection methods and understanding. We collect a Twitter dataset of @math users along with up to @math tweets from their timelines with a random-walk-based crawler on the retweet graph, and select a subsample of @math to be manually annotated as hateful or not through crowdsourcing. We examine the difference between user activity patterns, the content disseminated between hateful and normal users, and network centrality measurements in the sampled graph. Our results show that hateful users have more recent account creation dates, and more statuses, and followees per day. Additionally, they favorite more tweets, tweet in shorter intervals and are more central in the retweet network, contradicting the \"lone wolf\" stereotype often associated with such behavior. Hateful users are more negative, more profane, and use less words associated with topics such as hate, terrorism, violence and anger. We also identify similarities between hateful normal users and their 1-neighborhood, suggesting strong homophily."
]
} |
1812.02712 | 2903088770 | The online hate speech is proliferating with several organization and countries implementing laws to ban such harmful speech. While these restrictions might reduce the amount of such hateful content, it does so by restricting freedom of speech. Thus, an promising alternative supported by several organizations is to counter such hate speech with more speech. In this paper, We analyze hate speech and the corresponding counters (aka counterspeech) on Twitter. We perform several lexical, linguistic and psycholinguistic analysis on these user accounts and obverse that counter speakers employ several strategies depending on the target community. The hateful accounts express more negative sentiments and are more profane. We also find that the hate tweets by verified accounts have much more virality as compared to a tweet by a non-verified account. While the hate users seem to use words more about envy, hate, negative emotion, swearing terms, ugliness, the counter users use more words related to government, law, leader. We also build a supervised model for classifying the hateful and counterspeech accounts on Twitter and obtain an F-score of 0.77. We also make our dataset public to help advance the research on hate speech. | The current solution adopted by several organization and companies to tackle online hate speech involves blocking suspending the account or the particular hateful post. While these options are very powerful, they tend to violate the free speech doctrine. Countespeech is considered to be a promising solution in this direction as it can help in controlling the hate speech problem and at the same time, it supports free speech. Institutions can help in educating the public about hatespeech and its implications, consequences and how to respond. Programmes such as No Hate Speech' movement nohatespeechmovement and Facebook's counterspeech program counterfb help in raising awareness, providing support and seeking creative solutions @cite_9 . | {
"cite_N": [
"@cite_9"
],
"mid": [
"1732974977"
],
"abstract": [
"No longer confined to isolated corners of the web, cyber hate now enjoys a major presence on popular social media sites. The Facebook group “Kill a Jew Day,” for instance, acquired thousands of friends within days of its formation, while YouTube has hosted videos with names like “How to Kill Beaners,” “Execute the Gays,” and “Murder Muslim Scum.” The mainstreaming of cyber hate has the troubling potential to shape public expectations of online discourse. Internet intermediaries have the freedom and influence to seize this defining moment in cyber hate’s history. We believe that a thoughtful and nuanced intermediary-based approach to hate speech can foster respectful online discourse without suppressing valuable expression. To this end, we urge intermediaries to help address cyber hate by adopting accessible and transparent policies that educate users about their rights and responsibilities as digital citizens. Intermediaries’ options include challenging hateful speech by responding with counter-speech and empowering community members to enforce norms of digital citizenship."
]
} |
1812.02712 | 2903088770 | The online hate speech is proliferating with several organization and countries implementing laws to ban such harmful speech. While these restrictions might reduce the amount of such hateful content, it does so by restricting freedom of speech. Thus, an promising alternative supported by several organizations is to counter such hate speech with more speech. In this paper, We analyze hate speech and the corresponding counters (aka counterspeech) on Twitter. We perform several lexical, linguistic and psycholinguistic analysis on these user accounts and obverse that counter speakers employ several strategies depending on the target community. The hateful accounts express more negative sentiments and are more profane. We also find that the hate tweets by verified accounts have much more virality as compared to a tweet by a non-verified account. While the hate users seem to use words more about envy, hate, negative emotion, swearing terms, ugliness, the counter users use more words related to government, law, leader. We also build a supervised model for classifying the hateful and counterspeech accounts on Twitter and obtain an F-score of 0.77. We also make our dataset public to help advance the research on hate speech. | Silence in response to digital hate carries significant expressive costs as well. When powerful intermediaries rebut demeaning stereotypes (like the Michelle Obama image) and invidious falsehoods (such as holocaust denial), they send a powerful message to readers. Because intermediaries often enjoy respect and a sense of legitimacy, using counterspeech, they can demonstrate what it means to treat others with respect and dignity @cite_9 . | {
"cite_N": [
"@cite_9"
],
"mid": [
"1732974977"
],
"abstract": [
"No longer confined to isolated corners of the web, cyber hate now enjoys a major presence on popular social media sites. The Facebook group “Kill a Jew Day,” for instance, acquired thousands of friends within days of its formation, while YouTube has hosted videos with names like “How to Kill Beaners,” “Execute the Gays,” and “Murder Muslim Scum.” The mainstreaming of cyber hate has the troubling potential to shape public expectations of online discourse. Internet intermediaries have the freedom and influence to seize this defining moment in cyber hate’s history. We believe that a thoughtful and nuanced intermediary-based approach to hate speech can foster respectful online discourse without suppressing valuable expression. To this end, we urge intermediaries to help address cyber hate by adopting accessible and transparent policies that educate users about their rights and responsibilities as digital citizens. Intermediaries’ options include challenging hateful speech by responding with counter-speech and empowering community members to enforce norms of digital citizenship."
]
} |
1812.02709 | 2902280832 | Stochastic Gradient Langevin Dynamics (SGLD) is a combination of a Robbins-Monro type algorithm with Langevin dynamics. In this paper, the SGLD method with fixed step size @math is considered in order to sample from a logconcave target distribution @math , known up to a normalisation factor. We assume that unbiased estimates of the gradient from possibly dependent observations are available. It is shown that, for all @math , the Wasserstein- @math distance of the @math th iterate of the SGLD algorithm from @math is dominated by @math with appropriate constants @math and @math , whereas for the case of i.i.d. data the upper bound is improved to @math . | Corollary significantly improves on some of the results in @cite_23 in certain cases, compare also to @cite_3 . In @cite_23 the monotonicity assumption is not imposed, only a dissipativity condition is required and a more general recursive scheme is investigated. However, the input sequence @math is assumed i.i.d. In that setting, Theorem 2.1 of @cite_23 applies to (with the choice @math , @math , @math fixed, see also the last paragraph of Subsection 1.1 of @cite_23 ), and we get that [ W_2( Law ( ^ _n), ) ] holds whenever @math and @math with some @math . Our results provide the sharper estimates in a setting where @math may have dependences. For the case of i.i.d. @math see also the very recent @cite_5 . | {
"cite_N": [
"@cite_5",
"@cite_3",
"@cite_23"
],
"mid": [
"2888664689",
"2738410460",
"2590513847"
],
"abstract": [
"Discrete time analogues of ergodic stochastic differential equations (SDEs) are one of the most popular and flexible tools for sampling high-dimensional probability measures. Non-asymptotic analysis in the @math Wasserstein distance of sampling algorithms based on Euler discretisations of SDEs has been recently developed by several authors for log-concave probability distributions. In this work we replace the log-concavity assumption with a log-concavity at infinity condition. We provide novel @math convergence rates for Euler schemes, expressed explicitly in terms of problem parameters. From there we derive non-asymptotic bounds on the distance between the laws induced by Euler schemes and the invariant laws of SDEs, both for schemes with standard and with randomised (inaccurate) drifts. We also obtain bounds for the hierarchy of discretisation, which enables us to deploy a multi-level Monte Carlo estimator. Our proof relies on a novel construction of a coupling for the Markov chains that can be used to control both the @math and @math Wasserstein distances simultaneously. Finally, we provide a weak convergence analysis that covers both the standard and the randomised (inaccurate) drift case. In particular, we reveal that the variance of the randomised drift does not influence the rate of weak convergence of the Euler scheme to the SDE.",
"We present a unified framework to analyze the global convergence of Langevin dynamics based algorithms for nonconvex finite-sum optimization with @math component functions. At the core of our analysis is a direct analysis of the ergodicity of the numerical approximations to Langevin dynamics, which leads to faster convergence rates. Specifically, we show that gradient Langevin dynamics (GLD) and stochastic gradient Langevin dynamics (SGLD) converge to the almost minimizer within @math and @math stochastic gradient evaluations respectively, where @math is the problem dimension, and @math is the spectral gap of the Markov chain generated by GLD. Both of the results improve upon the best known gradient complexity results. Furthermore, for the first time we prove the global convergence guarantee for variance reduced stochastic gradient Langevin dynamics (VR-SGLD) to the almost minimizer after @math stochastic gradient evaluations, which outperforms the gradient complexities of GLD and SGLD in a wide regime. Our theoretical analyses shed some light on using Langevin dynamics based algorithms for nonconvex optimization with provable guarantees.",
"Stochastic Gradient Langevin Dynamics (SGLD) is a popular variant of Stochastic Gradient Descent, where properly scaled isotropic Gaussian noise is added to an unbiased estimate of the gradient at each iteration. This modest change allows SGLD to escape local minima and suffices to guarantee asymptotic convergence to global minimizers for sufficiently regular non-convex objectives (Gelfand and Mitter, 1991). The present work provides a nonasymptotic analysis in the context of non-convex learning problems, giving finite-time guarantees for SGLD to find approximate minimizers of both empirical and population risks. As in the asymptotic setting, our analysis relates the discrete-time SGLD Markov chain to a continuous-time diffusion process. A new tool that drives the results is the use of weighted transportation cost inequalities to quantify the rate of convergence of SGLD to a stationary distribution in the Euclidean @math -Wasserstein distance."
]
} |
1812.02709 | 2902280832 | Stochastic Gradient Langevin Dynamics (SGLD) is a combination of a Robbins-Monro type algorithm with Langevin dynamics. In this paper, the SGLD method with fixed step size @math is considered in order to sample from a logconcave target distribution @math , known up to a normalisation factor. We assume that unbiased estimates of the gradient from possibly dependent observations are available. It is shown that, for all @math , the Wasserstein- @math distance of the @math th iterate of the SGLD algorithm from @math is dominated by @math with appropriate constants @math and @math , whereas for the case of i.i.d. data the upper bound is improved to @math . | It is pointed out in @cite_20 that the ergodicity property of ) is sensitive to the step size @math . Lemma 6.3 of @cite_4 gives an example in which the Euler-Maruyama discretization is transient. As pointed out in @cite_4 , under discretization, the minorization condition is insensitive with appropriate sampling rate while the Lyapunov condition may be lost. An invariant measure exists if the two conditions hold simultaneously, see Theorem 7.3 of @cite_4 and also Theorem 3.2 of @cite_20 for similar discussions. In our paper we follow the footsteps of @cite_24 in imposing strong convexity of @math together with Lipschitzness of its gradient and thus we do obtain ergodicity of ). | {
"cite_N": [
"@cite_24",
"@cite_4",
"@cite_20"
],
"mid": [
"",
"2104049095",
"1983452151"
],
"abstract": [
"",
"The ergodic properties of SDEs, and various time discretizations for SDEs, are studied. The ergodicity of SDEs is established by using techniques from the theory of Markov chains on general state spaces, such as that expounded by Meyn-Tweedie. Application of these Markov chain results leads to straightforward proofs of geometric ergodicity for a variety of SDEs, including problems with degenerate noise and for problems with locally Lipschitz vector fields. Applications where this theory can be usefully applied include damped-driven Hamiltonian problems (the Langevin equation), the Lorenz equation with degenerate noise and gradient systems. The same Markov chain theory is then used to study time-discrete approximations of these SDEs. The two primary ingredients for ergodicity are a minorization condition and a Lyapunov condition. It is shown that the minorization condition is robust under approximation. For globally Lipschitz vector fields this is also true of the Lyapunov condition. However in the locally Lipschitz case the Lyapunov condition fails for explicit methods such as Euler-Maruyama; for pathwise approximations it is, in general, only inherited by specially constructed implicit discretizations. Examples of such discretization based on backward Euler methods are given, and approximation of the Langevin equation studied in some detail.",
"In this paper we consider a continuous-time method of approximating a given distribution using the Langevin diusion dLtdWt 1 2 r log (Lt)dt. We ®nd conditions under this diusion converges exponentially quickly to or does not: in one dimension, these are essentially that for distributions with exponential tails of the form (x) exp (y |x| , 0< <1, exponential convergence occurs if and only if 1. We then consider conditions under which the discrete approximations to the diusion converge. We ®rst show that even when the diusion itself converges, naive discretizations need not do so. We then consider a Metropolis-adjusted' version of the algorithm, and ®nd conditions under which this also converges at an exponential rate: perhaps surprisingly, even the Metropolized version need not converge exponentially fast even if the diusion does. We brie y discuss a truncated form of the algorithm which, in practice, should avoid the diculties of the other forms."
]
} |
1812.02709 | 2902280832 | Stochastic Gradient Langevin Dynamics (SGLD) is a combination of a Robbins-Monro type algorithm with Langevin dynamics. In this paper, the SGLD method with fixed step size @math is considered in order to sample from a logconcave target distribution @math , known up to a normalisation factor. We assume that unbiased estimates of the gradient from possibly dependent observations are available. It is shown that, for all @math , the Wasserstein- @math distance of the @math th iterate of the SGLD algorithm from @math is dominated by @math with appropriate constants @math and @math , whereas for the case of i.i.d. data the upper bound is improved to @math . | The convergence rate of the Euler-Maruyama scheme ) to its stationary distribution is comparable to that of the Langevin SDE ), see @cite_20 or the discussion after Proposition 3.2 of @cite_23 . However, @cite_23 proved the exponential convergence of ) under condition that the inverse temperature parameter @math satisfies @math , where @math appears in the dissipativity condition @math of @cite_23 , [ m >0, b 0: , h ( ,z) m | |^2 -b, R ^ d . ] Let us consider the same setting with strongly convex @math and @math . The constant @math , which controls the quadratic term, plays a similar role to @math in our Assumption . In @cite_23 , @math has to be larger than @math , i.e., outside a compact set, the function @math should be sufficiently convex. From the present paper it becomes clear that for the approximation ) to work, strong convexity ( @math ) is enough. Our result is stronger and not included in @cite_23 , however, it requires global convexity. Note also that the dimension @math has different effects on convergence rate: it is @math in @cite_23 and @math in our case, see Lemma . | {
"cite_N": [
"@cite_23",
"@cite_20"
],
"mid": [
"2590513847",
"1983452151"
],
"abstract": [
"Stochastic Gradient Langevin Dynamics (SGLD) is a popular variant of Stochastic Gradient Descent, where properly scaled isotropic Gaussian noise is added to an unbiased estimate of the gradient at each iteration. This modest change allows SGLD to escape local minima and suffices to guarantee asymptotic convergence to global minimizers for sufficiently regular non-convex objectives (Gelfand and Mitter, 1991). The present work provides a nonasymptotic analysis in the context of non-convex learning problems, giving finite-time guarantees for SGLD to find approximate minimizers of both empirical and population risks. As in the asymptotic setting, our analysis relates the discrete-time SGLD Markov chain to a continuous-time diffusion process. A new tool that drives the results is the use of weighted transportation cost inequalities to quantify the rate of convergence of SGLD to a stationary distribution in the Euclidean @math -Wasserstein distance.",
"In this paper we consider a continuous-time method of approximating a given distribution using the Langevin diusion dLtdWt 1 2 r log (Lt)dt. We ®nd conditions under this diusion converges exponentially quickly to or does not: in one dimension, these are essentially that for distributions with exponential tails of the form (x) exp (y |x| , 0< <1, exponential convergence occurs if and only if 1. We then consider conditions under which the discrete approximations to the diusion converge. We ®rst show that even when the diusion itself converges, naive discretizations need not do so. We then consider a Metropolis-adjusted' version of the algorithm, and ®nd conditions under which this also converges at an exponential rate: perhaps surprisingly, even the Metropolized version need not converge exponentially fast even if the diusion does. We brie y discuss a truncated form of the algorithm which, in practice, should avoid the diculties of the other forms."
]
} |
1812.02866 | 2904790293 | Let @math and @math be a set of red points and a set of blue points in the plane, respectively, such that @math is in general position, and let @math be a function. We show that if @math , then there exists a non-crossing geometric spanning tree @math on @math such that @math for every @math and the set of leaves of @math is @math , where every edge of @math is a straight-line segment. | Let @math be a set of red points and @math be a set of blue points in the plane. We always assume that @math and @math are disjoint and @math is in general position (i.e, no three points of @math are collinear). Several works @cite_2 @cite_7 have considered problems on non-crossing geometric spanning trees and geometric graphs (where edges are straight-line segments) on @math . See also the survey @cite_1 . | {
"cite_N": [
"@cite_1",
"@cite_7",
"@cite_2"
],
"mid": [
"4368930",
"1001549597",
""
],
"abstract": [
"In this paper, we give a short survey on discrete geometry on red and blue points in the plane, most of whose results were obtained in the past decade. We consider balanced subdivision problems, geometric graph problems, graph embedding problems, Gallai-type problems and others.",
"Let (X ) be a set of multicolored points in the plane such that no three points are collinear and each color appears on at most ( |X| 2 ) points. We show the existence of a non-crossing properly colored geometric perfect matching on (X ) (if (|X| ) is even), and the existence of a non-crossing properly colored geometric spanning tree with maximum degree at most (3 ) on (X ). Moreover, we show the existence of a non-crossing properly colored geometric perfect matching in the plane lattice. In order to prove these our results, we propose an useful lemma that gives a good partition of a sequence of multicolored points.",
""
]
} |
1812.02622 | 2902508384 | Tensor network decomposition, originated from quantum physics to model entangled many-particle quantum systems, turns out to be a promising mathematical technique to efficiently represent and process big data in parsimonious manner. In this study, we show that tensor networks can systematically partition structured data, e.g. color images, for distributed storage and communication in privacy-preserving manner. Leveraging the sea of big data and metadata privacy, empirical results show that neighbouring subtensors with implicit information stored in tensor network formats cannot be identified for data reconstruction. This technique complements the existing encryption and randomization techniques which store explicit data representation at one place and highly susceptible to adversarial attacks such as side-channel attacks and de-anonymization. Furthermore, we propose a theory for adversarial examples that mislead convolutional neural networks to misclassification using subspace analysis based on singular value decomposition (SVD). The theory is extended to analyze higher-order tensors using tensor-train SVD (TT-SVD); it helps to explain the level of susceptibility of different datasets to adversarial attacks, the structural similarity of different adversarial attacks including global and localized attacks, and the efficacy of different adversarial defenses based on input transformation. An efficient and adaptive algorithm based on robust TT-SVD is then developed to detect strong and static adversarial attacks. | In the field of computer vision, SVD has been used for image denoising, compression, and forensic such as steganography and watermarking. Higher-order SVD can help to disentangle the constituents factors or modes of image ensembles, e.g., TensorFaces @cite_2 for facial images with different lighting conditions, viewpoint and poses. Closely related to our work is the use of SVD to extract features for visual quality assessment. Reference images are usually provided for comparison between images before and after processing @cite_66 , whereas no reference measure requires assumptions on the patches to be analyzed such as anisotropy @cite_20 ; all of the SVD-based image quality metrics work on grayscale images. Adversary does not provide the reference images for comparison; our work extends the SVD properties to higher-order tensors and analyze the robustness of correlation structure extracted from input data for image classification by CNNs. | {
"cite_N": [
"@cite_66",
"@cite_20",
"@cite_2"
],
"mid": [
"2102817897",
"2171125155",
"2113055885"
],
"abstract": [
"We study the use of machine learning for visual quality evaluation with comprehensive singular value decomposition (SVD)-based visual features. In this paper, the two-stage process and the relevant work in the existing visual quality metrics are first introduced followed by an in-depth analysis of SVD for visual quality assessment. Singular values and vectors form the selected features for visual quality assessment. Machine learning is then used for the feature pooling process and demonstrated to be effective. This is to address the limitations of the existing pooling techniques, like simple summation, averaging, Minkowski summation, etc., which tend to be ad hoc. We advocate machine learning for feature pooling because it is more systematic and data driven. The experiments show that the proposed method outperforms the eight existing relevant schemes. Extensive analysis and cross validation are performed with ten publicly available databases (eight for images with a total of 4042 test images and two for video with a total of 228 videos). We use all publicly accessible software and databases in this study, as well as making our own software public, to facilitate comparison in future research.",
"Across the field of inverse problems in image and video processing, nearly all algorithms have various parameters which need to be set in order to yield good results. In practice, usually the choice of such parameters is made empirically with trial and error if no “ground-truth” reference is available. Some analytical methods such as cross-validation and Stein's unbiased risk estimate (SURE) have been successfully used to set such parameters. However, these methods tend to be strongly reliant on restrictive assumptions on the noise, and also computationally heavy. In this paper, we propose a no-reference metric Q which is based upon singular value decomposition of local image gradient matrix, and provides a quantitative measure of true image content (i.e., sharpness and contrast as manifested in visually salient geometric features such as edges,) in the presence of noise and other disturbances. This measure 1) is easy to compute, 2) reacts reasonably to both blur and random noise, and 3) works well even when the noise is not Gaussian. The proposed measure is used to automatically and effectively set the parameters of two leading image denoising algorithms. Ample simulated and real data experiments support our claims. Furthermore, tests using the TID2008 database show that this measure correlates well with subjective quality evaluations for both blur and noise distortions.",
"Natural images are the composite consequence of multiple factors related to scene structure, illumination, and imaging. Multilinear algebra, the algebra of higher-order tensors, offers a potent mathematical framework for analyzing the multifactor structure of image ensembles and for addressing the difficult problem of disentangling the constituent factors or modes. Our multilinear modeling technique employs a tensor extension of the conventional matrix singular value decomposition (SVD), known as the N-mode SVD. As a concrete example, we consider the multilinear analysis of ensembles of facial images that combine several modes, including different facial geometries (people), expressions, head poses, and lighting conditions. Our resulting \"TensorFaces\" representation has several advantages over conventional eigenfaces. More generally, multilinear analysis shows promise as a unifying framework for a variety of computer vision problems."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | In practice, the performance of locality-sensitive hashing techniques is usually measured by their recall: the fraction of the true @math -nearest neighbors found on average, see e.g. @cite_11 @cite_10 . From a theoretical point of view it is natural to bound the , i.e., the probability that the nearest neighbor is found. We are only aware of very few works that provide theoretical guarantees on expected recall in conjunction with sublinear query time in high dimensions and without assumptions on data. | {
"cite_N": [
"@cite_10",
"@cite_11"
],
"mid": [
"2949985202",
"2949388608"
],
"abstract": [
"This paper describes ANN-Benchmarks, a tool for evaluating the performance of in-memory approximate nearest neighbor algorithms. It provides a standard interface for measuring the performance and quality achieved by nearest neighbor algorithms on different standard data sets. It supports several different ways of integrating @math -NN algorithms, and its configuration system automatically tests a range of parameter settings for each algorithm. Algorithms are compared with respect to many different (approximate) quality measures, and adding more is easy and fast; the included plotting front-ends can visualise these as images, @math plots, and websites with interactive plots. ANN-Benchmarks aims to provide a constantly updated overview of the current state of the art of @math -NN algorithms. In the short term, this overview allows users to choose the correct @math -NN algorithm and parameters for their similarity search task; in the longer term, algorithm designers will be able to use this overview to test and refine automatic parameter tuning. The paper gives an overview of the system, evaluates the results of the benchmark, and points out directions for future work. Interestingly, very different approaches to @math -NN search yield comparable quality-performance trade-offs. The system is available at this http URL .",
"We show the existence of a Locality-Sensitive Hashing (LSH) family for the angular distance that yields an approximate Near Neighbor Search algorithm with the asymptotically optimal running time exponent. Unlike earlier algorithms with this property (e.g., Spherical LSH [Andoni, Indyk, Nguyen, Razenshteyn 2014], [Andoni, Razenshteyn 2015]), our algorithm is also practical, improving upon the well-studied hyperplane LSH [Charikar, 2002] in practice. We also introduce a multiprobe version of this algorithm, and conduct experimental evaluation on real and synthetic data sets. We complement the above positive results with a fine-grained lower bound for the quality of any LSH family for angular distance. Our lower bound implies that the above LSH family exhibits a trade-off between evaluation time and quality that is close to optimal for a natural class of LSH functions."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | @cite_7 outline an adaptive'' method for achieving a given expected recall in the context of multiprobe LSH (with no formal statement of guarantees). The idea is to determine, after inspecting @math buckets, whether to terminate or to inspect bucket @math based on the collision probability @math between @math and the nearest neighbor @math found in the first @math buckets. This requires an efficient method for computing @math , which might not be known, especially for small collision probabilities. This is not just a theoretical problem: Prominent LSH methods such as @math -stable LSH @cite_8 and cross-polytope LSH @cite_11 do not have closed-form expressions for collision probabilities. Our adaptive algorithm is similar in spirit, but entirely avoids having to compute collision probabilities. | {
"cite_N": [
"@cite_11",
"@cite_7",
"@cite_8"
],
"mid": [
"2949388608",
"2055839530",
"2162006472"
],
"abstract": [
"We show the existence of a Locality-Sensitive Hashing (LSH) family for the angular distance that yields an approximate Near Neighbor Search algorithm with the asymptotically optimal running time exponent. Unlike earlier algorithms with this property (e.g., Spherical LSH [Andoni, Indyk, Nguyen, Razenshteyn 2014], [Andoni, Razenshteyn 2015]), our algorithm is also practical, improving upon the well-studied hyperplane LSH [Charikar, 2002] in practice. We also introduce a multiprobe version of this algorithm, and conduct experimental evaluation on real and synthetic data sets. We complement the above positive results with a fine-grained lower bound for the quality of any LSH family for angular distance. Our lower bound implies that the above LSH family exhibits a trade-off between evaluation time and quality that is close to optimal for a natural class of LSH functions.",
"Although Locality-Sensitive Hashing (LSH) is a promising approach to similarity search in high-dimensional spaces, it has not been considered practical partly because its search quality is sensitive to several parameters that are quite data dependent. Previous research on LSH, though obtained interesting asymptotic results, provides little guidance on how these parameters should be chosen, and tuning parameters for a given dataset remains a tedious process. To address this problem, we present a statistical performance model of Multi-probe LSH, a state-of-the-art variance of LSH. Our model can accurately predict the average search quality and latency given a small sample dataset. Apart from automatic parameter tuning with the performance model, we also use the model to devise an adaptive LSH search algorithm to determine the probing parameter dynamically for each query. The adaptive probing method addresses the problem that even though the average performance is tuned for optimal, the variance of the performance is extremely high. We experimented with three different datasets including audio, images and 3D shapes to evaluate our methods. The results show the accuracy of the proposed model: the recall errors predicted are within 5 from the real values for most cases; the adaptive search method reduces the standard deviation of recall by about 50 over the existing method.",
"We present a novel Locality-Sensitive Hashing scheme for the Approximate Nearest Neighbor Problem under lp norm, based on p-stable distributions.Our scheme improves the running time of the earlier algorithm for the case of the lp norm. It also yields the first known provably efficient approximate NN algorithm for the case p<1. We also show that the algorithm finds the exact near neigbhor in O(log n) time for data satisfying certain \"bounded growth\" condition.Unlike earlier schemes, our LSH scheme works directly on points in the Euclidean space without embeddings. Consequently, the resulting query time bound is free of large factors and is simple and easy to implement. Our experiments (on synthetic data sets) show that the our data structure is up to 40 times faster than kd-tree."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | For the related problem where a search radius @math is given it is easier to give guarantees on recall, especially when collision probabilities at distance @math can be computed, see e.g. @cite_1 . | {
"cite_N": [
"@cite_1"
],
"mid": [
"2736701013"
],
"abstract": [
"Set similarity join is a fundamental and well-studied database operator. It is usually studied in the exact setting where the goal is to compute all pairs of sets that exceed a given similarity threshold (measured e.g. as Jaccard similarity). But set similarity join is often used in settings where 100 recall may not be important --- indeed, where the exact set similarity join is itself only an approximation of the desired result set. We present a new randomized algorithm for set similarity join that can achieve any desired recall up to 100 , and show theoretically and empirically that it significantly improves on existing methods. The present state-of-the-art exact methods are based on prefix-filtering, the performance of which depends on the data set having many rare tokens. Our method is robust against the absence of such structure in the data. At 90 recall our algorithm is often more than an order of magnitude faster than state-of-the-art exact methods, depending on how well a data set lends itself to prefix filtering. Our experiments on benchmark data sets also show that the method is several times faster than comparable approximate methods. Our algorithm makes use of recent theoretical advances in high-dimensional sketching and indexing that we believe to be of wider relevance to the data engineering community."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | Since the performance of LSH data structures depends on parameter choices, a lot of work has gone into devising ways of choosing good parameters for a given data set, both during data structure construction and adaptively for the query algorithm. @cite_0 propose to select parameters based on the distance profile'' of a data set, but needs a bound on the distance to the nearest neighbor to function. | {
"cite_N": [
"@cite_0"
],
"mid": [
"2085168338"
],
"abstract": [
"Locality-sensitive hashing (LSH) is the basis of many algorithms that use a probabilistic approach to find nearest neighbors. We describe an algorithm for optimizing the parameters and use of LSH. Prior work ignores these issues or suggests a search for the best parameters. We start with two histograms: one that characterizes the distributions of distances to a point's nearest neighbors and the second that characterizes the distance between a query and any point in the data set. Given a desired performance level (the chance of finding the true nearest neighbor) and a simple computational cost model, we return the LSH parameters that allow an LSH index to meet the performance goal and have the minimum computational cost. We can also use this analysis to connect LSH to deterministic nearest-neighbor algorithms such as @math - @math trees and thus start to unify the two approaches."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | The state-of-the-art FALCONN library @cite_11 uses grid search over parameters to empirically estimate the best parameters, assuming that the data and query distributions are identical. | {
"cite_N": [
"@cite_11"
],
"mid": [
"2949388608"
],
"abstract": [
"We show the existence of a Locality-Sensitive Hashing (LSH) family for the angular distance that yields an approximate Near Neighbor Search algorithm with the asymptotically optimal running time exponent. Unlike earlier algorithms with this property (e.g., Spherical LSH [Andoni, Indyk, Nguyen, Razenshteyn 2014], [Andoni, Razenshteyn 2015]), our algorithm is also practical, improving upon the well-studied hyperplane LSH [Charikar, 2002] in practice. We also introduce a multiprobe version of this algorithm, and conduct experimental evaluation on real and synthetic data sets. We complement the above positive results with a fine-grained lower bound for the quality of any LSH family for angular distance. Our lower bound implies that the above LSH family exhibits a trade-off between evaluation time and quality that is close to optimal for a natural class of LSH functions."
]
} |
1812.02603 | 2902799860 | Locality-sensitive hashing (LSH), introduced by Indyk and Motwani in STOC '98, has been an extremely influential framework for nearest neighbor search in high-dimensional data sets. While theoretical work has focused on the approximate nearest neighbor problems, in practice LSH data structures with suitably chosen parameters are used to solve the exact nearest neighbor problem (with some error probability). Sublinear query time is often possible in practice even for exact nearest neighbor search, intuitively because the nearest neighbor tends to be significantly closer than other data points. However, theory offers little advice on how to choose LSH parameters outside of pre-specified worst-case settings. We introduce the technique of confirmation sampling for solving the exact nearest neighbor problem using LSH. First, we give a general reduction that transforms a sequence of data structures that each find the nearest neighbor with a small, unknown probability, into a data structure that returns the nearest neighbor with probability @math , using as few queries as possible. Second, we present a new query algorithm for the LSH Forest data structure with @math trees that is able to return the exact nearest neighbor of a query point within the same time bound as an LSH Forest of @math trees with internal parameters specifically tuned to the query and data. | We note that the adaptive method of @cite_7 does not adapt search depth to the distance distribution from the query point @math . In fact, choosing good parameters for LSH and especially multi-probe LSH was mentioned by @cite_13 as a challenge in the paper celebrating their VLDB 10-year Best Paper Award. | {
"cite_N": [
"@cite_13",
"@cite_7"
],
"mid": [
"2752074276",
"2055839530"
],
"abstract": [
"The past decade has been marked by the (continued) explosion of diverse data content and the fast development of intelligent data analytics techniques. One problem we identified in the mid-2000s was similarity search of feature-rich data. The challenge here was achieving both high accuracy and high efficiency in high-dimensional spaces. Locality sensitive hashing (LSH), which uses certain random space partitions and hash table lookups to find approximate nearest neighbors, was a promising approach with theoretical guarantees. But LSH alone was insufficient since a large number of hash tables were required to achieve good search quality. Building on an idea of Panigrahy, our multi-probe LSH method introduced the idea of intelligent probing. Given a query object, we strategically probe its neighboring hash buckets (in a query-dependent fashion) by calculating the statistical probabilities of similar objects falling into each bucket. Such intelligent probing can significantly reduce the number of hash tables while achieving high quality. In this paper, we revisit the problem motivation, the challenges, the key design considerations of multi-probe LSH, as well as discuss recent developments in this space and some questions for further research.",
"Although Locality-Sensitive Hashing (LSH) is a promising approach to similarity search in high-dimensional spaces, it has not been considered practical partly because its search quality is sensitive to several parameters that are quite data dependent. Previous research on LSH, though obtained interesting asymptotic results, provides little guidance on how these parameters should be chosen, and tuning parameters for a given dataset remains a tedious process. To address this problem, we present a statistical performance model of Multi-probe LSH, a state-of-the-art variance of LSH. Our model can accurately predict the average search quality and latency given a small sample dataset. Apart from automatic parameter tuning with the performance model, we also use the model to devise an adaptive LSH search algorithm to determine the probing parameter dynamically for each query. The adaptive probing method addresses the problem that even though the average performance is tuned for optimal, the variance of the performance is extremely high. We experimented with three different datasets including audio, images and 3D shapes to evaluate our methods. The results show the accuracy of the proposed model: the recall errors predicted are within 5 from the real values for most cases; the adaptive search method reduces the standard deviation of recall by about 50 over the existing method."
]
} |
1812.02341 | 2903181768 | In this paper, we investigate the problem of overfitting in deep reinforcement learning. Among the most common benchmarks in RL, it is customary to use the same environments for both training and testing. This practice offers relatively little insight into an agent's ability to generalize. We address this issue by using procedurally generated environments to construct distinct training and test sets. Most notably, we introduce a new environment called CoinRun, designed as a benchmark for generalization in RL. Using CoinRun, we find that agents overfit to surprisingly large training sets. We then show that deeper convolutional architectures improve generalization, as do methods traditionally found in supervised learning, including L2 regularization, dropout, data augmentation and batch normalization. | @cite_6 also address this problem, accurately recognizing that conflating train and test environments has contributed to the lack of regularization in deep RL. They propose using different game modes of Atari 2600 games to measure generalization. They turn to supervised learning for inspiration, finding that both L2 regularization and dropout can help agents learn more generalizable features. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2893662673"
],
"abstract": [
"Deep reinforcement learning (RL) algorithms have shown an impressive ability to learn complex control policies in high-dimensional environments. However, despite the ever-increasing performance on popular benchmarks such as the Arcade Learning Environment (ALE), policies learned by deep RL algorithms often struggle to generalize when evaluated in remarkably similar environments. In this paper, we assess the generalization capabilities of DQN, one of the most traditional deep RL algorithms in the field. We provide evidence suggesting that DQN overspecializes to the training environment. We comprehensively evaluate the impact of traditional regularization methods, @math -regularization and dropout, and of reusing the learned representations to improve the generalization capabilities of DQN. We perform this study using different game modes of Atari 2600 games, a recently introduced modification for the ALE which supports slight variations of the Atari 2600 games traditionally used for benchmarking. Despite regularization being largely underutilized in deep RL, we show that it can, in fact, help DQN learn more general features. These features can then be reused and fine-tuned on similar tasks, considerably improving the sample efficiency of DQN."
]
} |
1812.02341 | 2903181768 | In this paper, we investigate the problem of overfitting in deep reinforcement learning. Among the most common benchmarks in RL, it is customary to use the same environments for both training and testing. This practice offers relatively little insight into an agent's ability to generalize. We address this issue by using procedurally generated environments to construct distinct training and test sets. Most notably, we introduce a new environment called CoinRun, designed as a benchmark for generalization in RL. Using CoinRun, we find that agents overfit to surprisingly large training sets. We then show that deeper convolutional architectures improve generalization, as do methods traditionally found in supervised learning, including L2 regularization, dropout, data augmentation and batch normalization. | @cite_3 propose a different benchmark to measure generalization using six classic environments, each of which has been modified to expose several internal parameters. By training and testing on environments with different parameter ranges, their benchmark quantifies agents' ability to interpolate and extrapolate. @cite_1 measure overfitting in continuous domains, finding that generalization improves as the number of training seeds increases. They also use randomized rewards to determine the extent of undesirable memorization. | {
"cite_N": [
"@cite_1",
"@cite_3"
],
"mid": [
"2809668646",
"2898436992"
],
"abstract": [
"The risks and perils of overfitting in machine learning are well known. However most of the treatment of this, including diagnostic tools and remedies, was developed for the supervised learning case. In this work, we aim to offer new perspectives on the characterization and prevention of overfitting in deep Reinforcement Learning (RL) methods, with a particular focus on continuous domains. We examine several aspects, such as how to define and diagnose overfitting in MDPs, and how to reduce risks by injecting sufficient training diversity. This work complements recent findings on the brittleness of deep RL methods and offers practical observations for RL researchers and practitioners.",
"Deep reinforcement learning (RL) has achieved breakthrough results on many tasks, but has been shown to be sensitive to system changes at test time. As a result, building deep RL agents that generalize has become an active research area. Our aim is to catalyze and streamline community-wide progress on this problem by providing the first benchmark and a common experimental protocol for investigating generalization in RL. Our benchmark contains a diverse set of environments and our evaluation methodology covers both in-distribution and out-of-distribution generalization. To provide a set of baselines for future research, we conduct a systematic evaluation of deep RL algorithms, including those that specifically tackle the problem of generalization."
]
} |
1812.02341 | 2903181768 | In this paper, we investigate the problem of overfitting in deep reinforcement learning. Among the most common benchmarks in RL, it is customary to use the same environments for both training and testing. This practice offers relatively little insight into an agent's ability to generalize. We address this issue by using procedurally generated environments to construct distinct training and test sets. Most notably, we introduce a new environment called CoinRun, designed as a benchmark for generalization in RL. Using CoinRun, we find that agents overfit to surprisingly large training sets. We then show that deeper convolutional architectures improve generalization, as do methods traditionally found in supervised learning, including L2 regularization, dropout, data augmentation and batch normalization. | Other works create distinct train and test environments using procedural generation. @cite_5 use the General Video Game AI (GVG-AI) framework to generate levels from several unique games. By varying difficulty settings between train and test levels, they find that RL agents regularly overfit to a particular training distribution. They further show that the ability to generalize to human-designed levels strongly depends on the level generators used during training. | {
"cite_N": [
"@cite_5"
],
"mid": [
"2891790128"
],
"abstract": [
"Deep reinforcement learning (RL) has shown impressive results in a variety of domains, learning directly from high-dimensional sensory streams. However, when neural networks are trained in a fixed environment, such as a single level in a video game, they will usually overfit and fail to generalize to new levels. When RL models overfit, even slight modifications to the environment can result in poor agent performance. This paper explores how procedurally generated levels during training can increase generality. We show that for some games procedural level generation enables generalization to new levels within the same distribution. Additionally, it is possible to achieve better performance with less data by manipulating the difficulty of the levels in response to the performance of the agent. The generality of the learned behaviors is also evaluated on a set of human-designed levels. The results suggest that the ability to generalize to human-designed levels highly depends on the design of the level generators. We apply dimensionality reduction and clustering techniques to visualize the generators' distributions of levels and analyze to what degree they can produce levels similar to those designed by a human."
]
} |
1812.02341 | 2903181768 | In this paper, we investigate the problem of overfitting in deep reinforcement learning. Among the most common benchmarks in RL, it is customary to use the same environments for both training and testing. This practice offers relatively little insight into an agent's ability to generalize. We address this issue by using procedurally generated environments to construct distinct training and test sets. Most notably, we introduce a new environment called CoinRun, designed as a benchmark for generalization in RL. Using CoinRun, we find that agents overfit to surprisingly large training sets. We then show that deeper convolutional architectures improve generalization, as do methods traditionally found in supervised learning, including L2 regularization, dropout, data augmentation and batch normalization. | @cite_2 conduct experiments on procedurally generated gridworld mazes, reporting many insightful conclusions on the nature of overfitting in RL agents. They find that agents have a high capacity to memorize specific levels in a given training set, and that techniques intended to mitigate overfitting in RL, including sticky actions and random starts , often fail to do so. | {
"cite_N": [
"@cite_2"
],
"mid": [
"2797527950"
],
"abstract": [
"Recent years have witnessed significant progresses in deep Reinforcement Learning (RL). Empowered with large scale neural networks, carefully designed architectures, novel training algorithms and massively parallel computing devices, researchers are able to attack many challenging RL problems. However, in machine learning, more training power comes with a potential risk of more overfitting. As deep RL techniques are being applied to critical problems such as healthcare and finance, it is important to understand the generalization behaviors of the trained agents. In this paper, we conduct a systematic study of standard RL agents and find that they could overfit in various ways. Moreover, overfitting could happen \"robustly\": commonly used techniques in RL that add stochasticity do not necessarily prevent or detect overfitting. In particular, the same agents and learning algorithms could have drastically different test performance, even when all of them achieve optimal rewards during training. The observations call for more principled and careful evaluation protocols in RL. We conclude with a general discussion on overfitting in RL and a study of the generalization behaviors from the perspective of inductive bias."
]
} |
1812.02341 | 2903181768 | In this paper, we investigate the problem of overfitting in deep reinforcement learning. Among the most common benchmarks in RL, it is customary to use the same environments for both training and testing. This practice offers relatively little insight into an agent's ability to generalize. We address this issue by using procedurally generated environments to construct distinct training and test sets. Most notably, we introduce a new environment called CoinRun, designed as a benchmark for generalization in RL. Using CoinRun, we find that agents overfit to surprisingly large training sets. We then show that deeper convolutional architectures improve generalization, as do methods traditionally found in supervised learning, including L2 regularization, dropout, data augmentation and batch normalization. | In sec:env_stoch , we similarly investigate how injecting stochasticity impacts generalization. Our work mirrors @cite_2 in quantifying the relationship between overfitting and the number of training environments, though we additionally show how several methods, including some more prevalent in supervised learning, can reduce overfitting in our benchmark. | {
"cite_N": [
"@cite_2"
],
"mid": [
"2797527950"
],
"abstract": [
"Recent years have witnessed significant progresses in deep Reinforcement Learning (RL). Empowered with large scale neural networks, carefully designed architectures, novel training algorithms and massively parallel computing devices, researchers are able to attack many challenging RL problems. However, in machine learning, more training power comes with a potential risk of more overfitting. As deep RL techniques are being applied to critical problems such as healthcare and finance, it is important to understand the generalization behaviors of the trained agents. In this paper, we conduct a systematic study of standard RL agents and find that they could overfit in various ways. Moreover, overfitting could happen \"robustly\": commonly used techniques in RL that add stochasticity do not necessarily prevent or detect overfitting. In particular, the same agents and learning algorithms could have drastically different test performance, even when all of them achieve optimal rewards during training. The observations call for more principled and careful evaluation protocols in RL. We conclude with a general discussion on overfitting in RL and a study of the generalization behaviors from the perspective of inductive bias."
]
} |
1812.02402 | 2902470604 | The performance of Deep Neural Networks (DNNs) keeps elevating in recent years with increasing network depth and width. To enable DNNs on edge devices like mobile phones, researchers proposed several network compression methods including pruning, quantization and factorization. Among the factorization-based approaches, low-rank approximation has been widely adopted because of its solid theoretical rationale and efficient implementations. Several previous works attempted to directly approximate a pre-trained model by low-rank decomposition; however, small approximation errors in parameters can ripple a large prediction loss. As a result, performance usually drops significantly and a sophisticated fine-tuning is required to recover accuracy. We argue that it is not optimal to separate low-rank approximation from training. Unlike previous works, this paper integrates low rank approximation and regularization into the training. We propose Trained Rank Pruning (TRP), which iterates low rank approximation and training. TRP maintains the capacity of original network while imposes low-rank constraints during training. A stochastic sub-gradient descent optimized nuclear regularization is utilized to further encourage low rank in TRP. The TRP trained network has low-rank structure in nature, and can be approximated with negligible performance loss, eliminating fine-tuning after low rank approximation. The methods are comprehensively evaluated on CIFAR-10 and ImageNet, outperforming previous compression methods using low rank approximation. | Non-structured and structured sparsity are introduced by pruning. @cite_7 proposes to prune unimportant connections between neural units with small weights in a pre-trained CNN, followed by network re-training to preserve the accuracy. @cite_15 utilizes group Lasso strategy to learn the structure sparsity of networks including weights, channels, filters and layers. @cite_11 adopts a similar strategy by explicitly imposing scaling factors on each channel to measure the importance of each connection and dropping those with small weights during training. In @cite_0 , the pruning problem is formulated as a data recovery problem. Pretrained filters are re-weighted by minimizing a data recovery objective function. Channels with smaller weight are pruned. @cite_26 heuristically selects filters using change of next layer's output as a criterion. | {
"cite_N": [
"@cite_26",
"@cite_7",
"@cite_0",
"@cite_15",
"@cite_11"
],
"mid": [
"2964233199",
"2963674932",
"2737121650",
"2513419314",
""
],
"abstract": [
"We propose an efficient and unified framework, namely ThiNet, to simultaneously accelerate and compress CNN models in both training and inference stages. We focus on the filter level pruning, i.e., the whole filter would be discarded if it is less important. Our method does not change the original network structure, thus it can be perfectly supported by any off-the-shelf deep learning libraries. We formally establish filter pruning as an optimization problem, and reveal that we need to prune filters based on statistics information computed from its next layer, not the current layer, which differentiates ThiNet from existing methods. Experimental results demonstrate the effectiveness of this strategy, which has advanced the state-of-the-art. We also show the performance of ThiNet on ILSVRC-12 benchmark. ThiNet achieves 3.31 x FLOPs reduction and 16.63× compression on VGG-16, with only 0.52 top-5 accuracy drop. Similar experiments with ResNet-50 reveal that even for a compact network, ThiNet can also reduce more than half of the parameters and FLOPs, at the cost of roughly 1 top-5 accuracy drop. Moreover, the original VGG-16 model can be further pruned into a very small model with only 5.05MB model size, preserving AlexNet level accuracy but showing much stronger generalization ability.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy.",
"In this paper, we introduce a new channel pruning method to accelerate very deep convolutional neural networks.Given a trained CNN model, we propose an iterative two-step algorithm to effectively prune each layer, by a LASSO regression based channel selection and least square reconstruction. We further generalize this algorithm to multi-layer and multi-branch cases. Our method reduces the accumulated error and enhance the compatibility with various architectures. Our pruned VGG-16 achieves the state-of-the-art results by 5x speed-up along with only 0.3 increase of error. More importantly, our method is able to accelerate modern networks like ResNet, Xception and suffers only 1.4 , 1.0 accuracy loss under 2x speed-up respectively, which is significant. Code has been made publicly available.",
"High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) learn a compact structure from a bigger DNN to reduce computation cost; (2) obtain a hardware-friendly structured sparsity of DNN to efficiently accelerate the DNNs evaluation. Experimental results show that SSL achieves on average 5.1x and 3.1x speedups of convolutional layer computation of AlexNet against CPU and GPU, respectively, with off-the-shelf libraries. These speedups are about twice speedups of non-structured sparsity; (3) regularize the DNN structure to improve classification accuracy. The results show that for CIFAR-10, regularization on layer depth can reduce 20 layers of a Deep Residual Network (ResNet) to 18 layers while improve the accuracy from 91.25 to 92.60 , which is still slightly higher than that of original ResNet with 32 layers. For AlexNet, structure regularization by SSL also reduces the error by around 1 . Open source code is in this https URL",
""
]
} |
1812.02402 | 2902470604 | The performance of Deep Neural Networks (DNNs) keeps elevating in recent years with increasing network depth and width. To enable DNNs on edge devices like mobile phones, researchers proposed several network compression methods including pruning, quantization and factorization. Among the factorization-based approaches, low-rank approximation has been widely adopted because of its solid theoretical rationale and efficient implementations. Several previous works attempted to directly approximate a pre-trained model by low-rank decomposition; however, small approximation errors in parameters can ripple a large prediction loss. As a result, performance usually drops significantly and a sophisticated fine-tuning is required to recover accuracy. We argue that it is not optimal to separate low-rank approximation from training. Unlike previous works, this paper integrates low rank approximation and regularization into the training. We propose Trained Rank Pruning (TRP), which iterates low rank approximation and training. TRP maintains the capacity of original network while imposes low-rank constraints during training. A stochastic sub-gradient descent optimized nuclear regularization is utilized to further encourage low rank in TRP. The TRP trained network has low-rank structure in nature, and can be approximated with negligible performance loss, eliminating fine-tuning after low rank approximation. The methods are comprehensively evaluated on CIFAR-10 and ImageNet, outperforming previous compression methods using low rank approximation. | Weight quantization methods include training a quantized model from scratch @cite_37 @cite_9 @cite_17 or converting a pre-trained model into quantized representation @cite_13 @cite_19 @cite_34 @cite_33 . The quantized weight representation includes binary value @cite_17 @cite_9 or hash buckets @cite_37 . For training a quantized model from scratch, some works propose to use the gradient the quantized weight to update the full-precision counterpart @cite_9 or update the binarized weight with stochastic rounding scheme @cite_16 during backward process. Note that our method here is inspired by the scheme of combining quantization with training process, we embed the low-rank decomposition into training process to explicitly guide the parameter to a low-rank form. | {
"cite_N": [
"@cite_37",
"@cite_33",
"@cite_9",
"@cite_16",
"@cite_19",
"@cite_34",
"@cite_13",
"@cite_17"
],
"mid": [
"2952432176",
"2789142637",
"2260663238",
"1841592590",
"2952936791",
"2119144962",
"2593245696",
""
],
"abstract": [
"As deep nets are increasingly used in applications suited for mobile devices, a fundamental dilemma becomes apparent: the trend in deep learning is to grow models to absorb ever-increasing data set sizes; however mobile devices are designed with very little memory and cannot store such large models. We present a novel network architecture, HashedNets, that exploits inherent redundancy in neural networks to achieve drastic reductions in model sizes. HashedNets uses a low-cost hash function to randomly group connection weights into hash buckets, and all connections within the same hash bucket share a single parameter value. These parameters are tuned to adjust to the HashedNets weight sharing architecture with standard backprop during training. Our hashing procedure introduces no additional memory overhead, and we demonstrate on several benchmark data sets that HashedNets shrink the storage requirements of neural networks substantially while mostly preserving generalization performance.",
"Network quantization is an effective solution to compress deep neural networks for practical usage. Existing network quantization methods cannot sufficiently exploit the depth information to generate low-bit compressed network. In this paper, we propose two novel network quantization approaches, single-level network quantization (SLQ) for high-bit quantization and multi-level network quantization (MLQ) for extremely low-bit quantization (ternary).We are the first to consider the network quantization from both width and depth level. In the width level, parameters are divided into two parts: one for quantization and the other for re-training to eliminate the quantization loss. SLQ leverages the distribution of the parameters to improve the width level. In the depth level, we introduce incremental layer compensation to quantize layers iteratively which decreases the quantization loss in each iteration. The proposed approaches are validated with extensive experiments based on the state-of-the-art neural networks including AlexNet, VGG-16, GoogleNet and ResNet-18. Both SLQ and MLQ achieve impressive results.",
"",
"Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding.",
"In recent years increasingly complex architectures for deep convolution networks (DCNs) have been proposed to boost the performance on image recognition tasks. However, the gains in performance have come at a cost of substantial increase in computation and model storage resources. Fixed point implementation of DCNs has the potential to alleviate some of these complexities and facilitate potential deployment on embedded hardware. In this paper, we propose a quantizer design for fixed point implementation of DCNs. We formulate and solve an optimization problem to identify optimal fixed point bit-width allocation across DCN layers. Our experiments show that in comparison to equal bit-width settings, the fixed point DCNs with optimized bit width allocation offer >20 reduction in the model size without any loss in accuracy on CIFAR-10 benchmark. We also demonstrate that fine-tuning can further enhance the accuracy of fixed point DCNs beyond that of the original floating point model. In doing so, we report a new state-of-the-art fixed point performance of 6.78 error-rate on CIFAR-10 benchmark.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.",
"This paper presents incremental network quantization (INQ), a novel method, targeting to efficiently convert any pre-trained full-precision convolutional neural network (CNN) model into a low-precision version whose weights are constrained to be either powers of two or zero. Unlike existing methods which are struggled in noticeable accuracy loss, our INQ has the potential to resolve this issue, as benefiting from two innovations. On one hand, we introduce three interdependent operations, namely weight partition, group-wise quantization and re-training. A well-proven measure is employed to divide the weights in each layer of a pre-trained CNN model into two disjoint groups. The weights in the first group are responsible to form a low-precision base, thus they are quantized by a variable-length encoding method. The weights in the other group are responsible to compensate for the accuracy loss from the quantization, thus they are the ones to be re-trained. On the other hand, these three operations are repeated on the latest re-trained group in an iterative manner until all the weights are converted into low-precision ones, acting as an incremental network quantization and accuracy enhancement procedure. Extensive experiments on the ImageNet classification task using almost all known deep CNN architectures including AlexNet, VGG-16, GoogleNet and ResNets well testify the efficacy of the proposed method. Specifically, at 5-bit quantization, our models have improved accuracy than the 32-bit floating-point references. Taking ResNet-18 as an example, we further show that our quantized models with 4-bit, 3-bit and 2-bit ternary weights have improved or very similar accuracy against its 32-bit floating-point baseline. Besides, impressive results with the combination of network pruning and INQ are also reported. The code is available at this https URL.",
""
]
} |
1812.02402 | 2902470604 | The performance of Deep Neural Networks (DNNs) keeps elevating in recent years with increasing network depth and width. To enable DNNs on edge devices like mobile phones, researchers proposed several network compression methods including pruning, quantization and factorization. Among the factorization-based approaches, low-rank approximation has been widely adopted because of its solid theoretical rationale and efficient implementations. Several previous works attempted to directly approximate a pre-trained model by low-rank decomposition; however, small approximation errors in parameters can ripple a large prediction loss. As a result, performance usually drops significantly and a sophisticated fine-tuning is required to recover accuracy. We argue that it is not optimal to separate low-rank approximation from training. Unlike previous works, this paper integrates low rank approximation and regularization into the training. We propose Trained Rank Pruning (TRP), which iterates low rank approximation and training. TRP maintains the capacity of original network while imposes low-rank constraints during training. A stochastic sub-gradient descent optimized nuclear regularization is utilized to further encourage low rank in TRP. The TRP trained network has low-rank structure in nature, and can be approximated with negligible performance loss, eliminating fine-tuning after low rank approximation. The methods are comprehensively evaluated on CIFAR-10 and ImageNet, outperforming previous compression methods using low rank approximation. | Original models are decomposed into compact ones with more lightweight layers. @cite_1 is one of earlier works to exploit low-rank approximation of convolution filters. The authors apply several tensor decomposition methods to decompose both convolution and fully connection layers. The overall speedup, however, is not notable. @cite_24 considers both the spatial-wise and channel-wise redundancy and proposes decomposing a filter into two cascaded asymmetric filters. @cite_4 further assumes the feature map lie in a low-rank subspace and decompose the convolution filter into @math followed by @math filters via SVD. @cite_12 exploits the low-rank assumption of convolution filters and decompose a regular convolution into several depth-wise and point-wise convolution structures. Although these works achieved notable performance in network compression and acceleration, all of them are based on the low-rank assumption, which may limit the performance accuracy or result in large error when the assumption is not completely satisfied. | {
"cite_N": [
"@cite_24",
"@cite_1",
"@cite_12",
"@cite_4"
],
"mid": [
"2950967261",
"2167215970",
"2885340141",
""
],
"abstract": [
"The focus of this paper is speeding up the evaluation of convolutional neural networks. While delivering impressive results across a range of computer vision and machine learning tasks, these networks are computationally demanding, limiting their deployability. Convolutional layers generally consume the bulk of the processing time, and so in this work we present two simple schemes for drastically speeding up these layers. This is achieved by exploiting cross-channel or filter redundancy to construct a low rank basis of filters that are rank-1 in the spatial domain. Our methods are architecture agnostic, and can be easily applied to existing CPU and GPU convolutional frameworks for tuneable speedup performance. We demonstrate this with a real world network designed for scene text character recognition, showing a possible 2.5x speedup with no loss in accuracy, and 4.5x speedup with less than 1 drop in accuracy, still achieving state-of-the-art on standard benchmarks.",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model.",
"Depthwise separable convolution has shown great efficiency in network design, but requires time-consuming training procedure with full training-set available. This paper first analyzes the mathematical relationship between regular convolutions and depthwise separable convolutions, and proves that the former one could be approximated with the latter one in closed form. We show depthwise separable convolutions are principal components of regular convolutions. And then we propose network decoupling (ND), a training-free method to accelerate convolutional neural networks (CNNs) by transferring pre-trained CNN models into the MobileNet-like depthwise separable convolution structure, with a promising speedup yet negligible accuracy loss. We further verify through experiments that the proposed method is orthogonal to other training-free methods like channel decomposition, spatial decomposition, etc. Combining the proposed method with them will bring even larger CNN speedup. For instance, ND itself achieves about 2X speedup for the widely used VGG16, and combined with other methods, it reaches 3.7X speedup with graceful accuracy degradation. We demonstrate that ND is widely applicable to classification networks like ResNet, and object detection network like SSD300.",
""
]
} |
1812.02354 | 2903153834 | Answer selection and knowledge base question answering (KBQA) are two important tasks of question answering (QA) systems. Existing methods solve these two tasks separately, which requires large number of repetitive work and neglects the rich correlation information between tasks. In this paper, we tackle answer selection and KBQA tasks simultaneously via multi-task learning (MTL), motivated by the following motivations. First, both answer selection and KBQA can be regarded as a ranking problem, with one at text-level while the other at knowledge-level. Second, these two tasks can benefit each other: answer selection can incorporate the external knowledge from knowledge base (KB), while KBQA can be improved by learning contextual information from answer selection. To fulfill the goal of jointly learning these two tasks, we propose a novel multi-task learning scheme that utilizes multi-view attention learned from various perspectives to enable these tasks to interact with each other as well as learn more comprehensive sentence representations. The experiments conducted on several real-world datasets demonstrate the effectiveness of the proposed method, and the performance of answer selection and KBQA is improved. Also, the multi-view attention scheme is proved to be effective in assembling attentive information from different representational perspectives. | Neural networks based models has been proven effective in answer selection task, e.g., convolutional neural network (CNN) @cite_19 and recurrent models like the long short-term memory (LSTM) @cite_14 . Instead of learning the representations of the question and the answer separately, most recent studies utilize attention mechanisms to learn the interaction information between questions and answers, which can better focus on relevant parts of the input @cite_12 @cite_17 @cite_5 . With the emergence of large-scale knowledge bases, several efforts have been made on incorporating external knowledge from KBs to improve answer selection models. Savenkov2017EviNets (2017) scores candidate answers by combining the supporting evidence from structured KBs and unstructured text documents. Shen2018Knowledge (2018) proposed a knowledge-aware attentive neural network to attend the knowledge information in QA sentences, based on the entities in sentences. DBLP:conf coling DengSYLDFL18 (2018) leverage external knowledge from KB as a bridge to transfer more valuable information between cross-domain answer selection tasks. | {
"cite_N": [
"@cite_14",
"@cite_19",
"@cite_5",
"@cite_12",
"@cite_17"
],
"mid": [
"2251202616",
"1966443646",
"2788835992",
"2517782820",
"2264105282"
],
"abstract": [
"In this paper, we present an approach that address the answer sentence selection problem for question answering. The proposed method uses a stacked bidirectional Long-Short Term Memory (BLSTM) network to sequentially read words from question and answer sentences, and then outputs their relevance scores. Unlike prior work, this approach does not require any syntactic parsing or external knowledge resources such as WordNet which may not be available in some domains or languages. The full system is based on a combination of the stacked BLSTM relevance model and keywords matching. The results of our experiments on a public benchmark dataset from TREC show that our system outperforms previous work which requires syntactic features and external knowledge resources.",
"Learning a similarity function between pairs of objects is at the core of learning to rank approaches. In information retrieval tasks we typically deal with query-document pairs, in question answering -- question-answer pairs. However, before learning can take place, such pairs needs to be mapped from the original space of symbolic words into some feature space encoding various aspects of their relatedness, e.g. lexical, syntactic and semantic. Feature engineering is often a laborious task and may require external knowledge sources that are not always available or difficult to obtain. Recently, deep learning approaches have gained a lot of attention from the research community and industry for their ability to automatically learn optimal feature representation for a given task, while claiming state-of-the-art performance in many tasks in computer vision, speech recognition and natural language processing. In this paper, we present a convolutional neural network architecture for reranking pairs of short texts, where we learn the optimal representation of text pairs and a similarity function to relate them in a supervised way from the available training data. Our network takes only words in the input, thus requiring minimal preprocessing. In particular, we consider the task of reranking short text pairs where elements of the pair are sentences. We test our deep learning system on two popular retrieval tasks from TREC: Question Answering and Microblog Retrieval. Our model demonstrates strong performance on the first task beating previous state-of-the-art systems by about 3 absolute points in both MAP and MRR and shows comparable results on tweet reranking, while enjoying the benefits of no manual feature engineering and no additional syntactic parsers.",
"",
"Passage-level question answer matching is a challenging task since it requires effective representations that capture the complex semantic relations between questions and answers. In this work, we propose a series of deep learning models to address passage answer selection. To match passage answers to questions accommodating their complex semantic relations, unlike most previous work that utilizes a single deep learning structure, we develop hybrid models that process the text using both convolutional and recurrent neural networks, combining the merits on extracting linguistic information from both structures. Additionally, we also develop a simple but effective attention mechanism for the purpose of constructing better answer representations according to the input question, which is imperative for better modeling long answer sequences. The results on two public benchmark datasets, InsuranceQA and TREC-QA, show that our proposed models outperform a variety of strong baselines.",
"In this work, we propose Attentive Pooling (AP), a two-way attention mechanism for discriminative model training. In the context of pair-wise ranking or classification with neural networks, AP enables the pooling layer to be aware of the current input pair, in a way that information from the two input items can directly influence the computation of each other's representations. Along with such representations of the paired inputs, AP jointly learns a similarity measure over projected segments (e.g. trigrams) of the pair, and subsequently, derives the corresponding attention vector for each input to guide the pooling. Our two-way attention mechanism is a general framework independent of the underlying representation learning, and it has been applied to both convolutional neural networks (CNNs) and recurrent neural networks (RNNs) in our studies. The empirical results, from three very different benchmark tasks of question answering answer selection, demonstrate that our proposed models outperform a variety of strong baselines and achieve state-of-the-art performance in all the benchmarks."
]
} |
1812.02354 | 2903153834 | Answer selection and knowledge base question answering (KBQA) are two important tasks of question answering (QA) systems. Existing methods solve these two tasks separately, which requires large number of repetitive work and neglects the rich correlation information between tasks. In this paper, we tackle answer selection and KBQA tasks simultaneously via multi-task learning (MTL), motivated by the following motivations. First, both answer selection and KBQA can be regarded as a ranking problem, with one at text-level while the other at knowledge-level. Second, these two tasks can benefit each other: answer selection can incorporate the external knowledge from knowledge base (KB), while KBQA can be improved by learning contextual information from answer selection. To fulfill the goal of jointly learning these two tasks, we propose a novel multi-task learning scheme that utilizes multi-view attention learned from various perspectives to enable these tasks to interact with each other as well as learn more comprehensive sentence representations. The experiments conducted on several real-world datasets demonstrate the effectiveness of the proposed method, and the performance of answer selection and KBQA is improved. Also, the multi-view attention scheme is proved to be effective in assembling attentive information from different representational perspectives. | Most approaches of KBQA are based on semantic parsing, which transform natural language questions into structured query to extract the answer from the knowledge base @cite_15 @cite_23 . Besides, most of state-of-the-art results are achieved by deep learning models. The core idea is to learn semantic representations of both the question and the knowledge. Based on the relevancy score between the representations of the question and the knowledge, the model ultimately re-rank a list of candidate facts from KB @cite_28 @cite_10 @cite_18 . Contrary to answer selection, some recent studies integrate word or context information to aid in knowledge base question answering @cite_21 @cite_16 . | {
"cite_N": [
"@cite_18",
"@cite_28",
"@cite_21",
"@cite_23",
"@cite_15",
"@cite_16",
"@cite_10"
],
"mid": [
"2739716023",
"580074167",
"2963655104",
"2251079237",
"2148721079",
"2949960352",
"2419173203"
],
"abstract": [
"",
"Training large-scale question answering systems is complicated because training sources usually cover a small portion of the range of possible questions. This paper studies the impact of multitask and transfer learning for simple question answering; a setting for which the reasoning required to answer is quite easy, as long as one can retrieve the correct evidence given a question, which can be difficult in large-scale conditions. To this end, we introduce a new dataset of 100k questions that we use in conjunction with existing benchmarks. We conduct our study within the framework of Memory Networks (, 2015) because this perspective allows us to eventually scale up to more complex reasoning, and show that Memory Networks can be successfully trained to achieve excellent performance.",
"",
"We propose a novel semantic parsing framework for question answering using a knowledge base. We define a query graph that resembles subgraphs of the knowledge base and can be directly mapped to a logical form. Semantic parsing is reduced to query graph generation, formulated as a staged search problem. Unlike traditional approaches, our method leverages the knowledge base in an early stage to prune the search space and thus simplifies the semantic matching problem. By applying an advanced entity linking system and a deep convolutional neural network model that matches questions and predicate sequences, our system outperforms previous methods substantially, and achieves an F1 measure of 52.5 on the WEBQUESTIONS dataset.",
"Answering natural language questions using the Freebase knowledge base has recently been explored as a platform for advancing the state of the art in open domain semantic parsing. Those efforts map questions to sophisticated meaning representations that are then attempted to be matched against viable answer candidates in the knowledge base. Here we show that relatively modest information extraction techniques, when paired with a webscale corpus, can outperform these sophisticated approaches by roughly 34 relative gain.",
"The most approaches to Knowledge Base Question Answering are based on semantic parsing. In this paper, we address the problem of learning vector representations for complex semantic parses that consist of multiple entities and relations. Previous work largely focused on selecting the correct semantic relations for a question and disregarded the structure of the semantic parse: the connections between entities and the directions of the relations. We propose to use Gated Graph Neural Networks to encode the graph structure of the semantic parse. We show on two data sets that the graph networks outperform all baseline models that do not explicitly model the structure. The error analysis confirms that our approach can successfully process complex semantic parses.",
"This work focuses on answering single-relation factoid questions over Freebase. Each question can acquire the answer from a single fact of form (subject, predicate, object) in Freebase. This task, simple question answering (SimpleQA), can be addressed via a two-step pipeline: entity linking and fact selection. In fact selection, we match the subject entity in a fact candidate with the entity mention in the question by a character-level convolutional neural network (char-CNN), and match the predicate in that fact with the question by a word-level CNN (word-CNN). This work makes two main contributions. (i) A simple and effective entity linker over Freebase is proposed. Our entity linker outperforms the state-of-the-art entity linker over SimpleQA task. (ii) A novel attentive maxpooling is stacked over word-CNN, so that the predicate representation can be matched with the predicate-focused question representation more effectively. Experiments show that our system sets new state-of-the-art in this task."
]
} |
1812.02466 | 2951604091 | In this paper, we address the problem of hand-drawn sketch recognition. Inspired by the Bayesian decision theory, we present a deep metric learning loss with the objective to minimize the Bayesian risk of misclassification. We estimate this risk for every mini-batch during training, and learn robust deep embeddings by backpropagating it to a deep neural network in an end-to-end trainable paradigm. Our learnt embeddings are discriminative and robust despite of intra-class variations and inter-class similarities naturally present in hand-drawn sketch images. Outperforming the state of the art on sketch recognition, our method achieves 82.2 and 88.7 on TU-Berlin-250 and TU-Berlin-160 benchmarks respectively. | Metric learning is a well-established area in Machine Learning with growing interest in deep methods for this problem in recent years. In this paper we will limit our discussion to deep metric learning methods. However, we encourage the readers to refer @cite_10 for details of classical metric learning techniques. In deep metric learning research the major effort goes into designing a discriminative loss function. The contrastive @cite_12 and triplet loss @cite_30 @cite_3 have shown their utility in various Computer Vision tasks and their usage is widespread. However, their drawbacks are (i) they do not use the complete information available in the batch, and (ii) their convergence is often subject to the correct choice of triplets. Other recent line of research include histogram loss @cite_1 , lifted-structured embedding @cite_4 and Multi-class N-pair Loss @cite_17 . The histogram loss function is computed based on the histograms of positive and negative pairs. Leveraging this idea, we present a principled approach of loss computation using Bayesian decision theory, and minimize the risk of positive pair getting classified as negative pair and vice-versa. | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_1",
"@cite_3",
"@cite_10",
"@cite_12",
"@cite_17"
],
"mid": [
"",
"2176040302",
"2547446130",
"2096733369",
"1898424075",
"2138621090",
"2555897561"
],
"abstract": [
"",
"Learning the distance metric between pairs of examples is of great importance for learning and visual recognition. With the remarkable success from the state of the art convolutional neural networks, recent works have shown promising results on discriminatively training the networks to learn semantic feature embeddings where similar examples are mapped close to each other and dissimilar examples are mapped farther apart. In this paper, we describe an algorithm for taking full advantage of the training batches in the neural network training by lifting the vector of pairwise distances within the batch to the matrix of pairwise distances. This step enables the algorithm to learn the state of the art feature embedding by optimizing a novel structured prediction objective on the lifted problem. Additionally, we collected Online Products dataset: 120k images of 23k classes of online products for metric learning. Our experiments on the CUB-200-2011, CARS196, and Online Products datasets demonstrate significant improvement over existing deep feature embedding methods on all experimented embedding sizes with the GoogLeNet network.",
"We suggest a new loss for learning deep embeddings. The key characteristics of the new loss is the absence of tunable parameters and very good results obtained across a range of datasets and problems. The loss is computed by estimating two distribution of similarities for positive (matching) and negative (non-matching) point pairs, and then computing the probability of a positive pair to have a lower similarity score than a negative pair based on these probability estimates. We show that these operations can be performed in a simple and piecewise-differentiable manner using 1D histograms with soft assignment operations. This makes the proposed loss suitable for learning deep embeddings using stochastic optimization. The experiments reveal favourable results compared to recently proposed loss functions.",
"Despite significant recent advances in the field of face recognition [10, 14, 15, 17], implementing face verification and recognition efficiently at scale presents serious challenges to current approaches. In this paper we present a system, called FaceNet, that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure of face similarity. Once this space has been produced, tasks such as face recognition, verification and clustering can be easily implemented using standard techniques with FaceNet embeddings as feature vectors.",
"The need for appropriate ways to measure the distance or similarity between data is ubiquitous in machine learning, pattern recognition and data mining, but handcrafting such good metrics for specific problems is generally difficult. This has led to the emergence of metric learning, which aims at automatically learning a metric from data and has attracted a lot of interest in machine learning and related fields for the past ten years. This survey paper proposes a systematic review of the metric learning literature, highlighting the pros and cons of each approach. We pay particular attention to Mahalanobis distance metric learning, a well-studied and successful framework, but additionally present a wide range of methods that have recently emerged as powerful alternatives, including nonlinear metric learning, similarity learning and local metric learning. Recent trends and extensions, such as semi-supervised metric learning, metric learning for histogram data and the derivation of generalization guarantees, are also covered. Finally, this survey addresses metric learning for structured data, in particular edit distance learning, and attempts to give an overview of the remaining challenges in metric learning for the years to come.",
"Dimensionality reduction involves mapping a set of high dimensional input points onto a low dimensional manifold so that 'similar\" points in input space are mapped to nearby points on the manifold. We present a method - called Dimensionality Reduction by Learning an Invariant Mapping (DrLIM) - for learning a globally coherent nonlinear function that maps the data evenly to the output manifold. The learning relies solely on neighborhood relationships and does not require any distancemeasure in the input space. The method can learn mappings that are invariant to certain transformations of the inputs, as is demonstrated with a number of experiments. Comparisons are made to other techniques, in particular LLE.",
"Deep metric learning has gained much popularity in recent years, following the success of deep learning. However, existing frameworks of deep metric learning based on contrastive loss and triplet loss often suffer from slow convergence, partially because they employ only one negative example while not interacting with the other negative classes in each update. In this paper, we propose to address this problem with a new metric learning objective called multi-class N-pair loss. The proposed objective function firstly generalizes triplet loss by allowing joint comparison among more than one negative examples - more specifically, N-1 negative examples - and secondly reduces the computational burden of evaluating deep embedding vectors via an efficient batch construction strategy using only N pairs of examples, instead of (N+1) x N. We demonstrate the superiority of our proposed loss to the triplet loss as well as other competing loss functions for a variety of tasks on several visual recognition benchmark, including fine-grained object recognition and verification, image clustering and retrieval, and face verification and identification."
]
} |
1812.02393 | 2952968823 | Crowd counting, i.e., estimation number of pedestrian in crowd images, is emerging as an important research problem with the public security applications. A key ingredient in the design of crowd counting systems is the construction of counting models while being robust to various scenarios under facts such as camera perspective and physical barriers. In this paper, we present an adaptive scenario discovery framework for crowd counting. The system is structured with two parallel pathways that are trained with different sizes of receptive field to represent different scales and crowd densities. After ensuring that these components are present in the proper geometric configuration, a third branch is designed to adaptively recalibrate the pathway-wise responses by discovering and modeling the dynamic scenarios implicitly. Our system is able to represent highly variable crowd images and achieves state-of-the-art results in two challenging benchmarks. | Numerous efforts have been devoted to the design of crowd counting models. Detail survey of the recent progress can be found in @cite_9 . In this section, we mainly discuss literature on the models with multiple branches representation, which are more related to this work. In @cite_18 , Zhang al proposed the MCNN by using three columns of convolutional neural networks with filters of different sizes. Sam al @cite_19 proposed the Switching-CNN, which decoupled the three columns into separate CNN (each trained with a subset of the patches), and a density selector is designed to utilize the structural and functional differences. Several works have studied the context information of the crowd images under multiple branch setting. For instance, Sindagi al @cite_2 applied local and global context coding to population count density estimation, and Zhang al @cite_4 proposed a scale-adaptive CNN architecture with a backbone of fixed small receptive fields. Another work related to ours is the CSRNet @cite_23 , where convolutional neural networks with dilation operations were employed after the backbone of the pre-trained deep model. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_9",
"@cite_19",
"@cite_23",
"@cite_2"
],
"mid": [
"2463631526",
"2964264515",
"2729018917",
"2741077351",
"2964209782",
"2963035940"
],
"abstract": [
"This paper aims to develop a method than can accurately estimate the crowd count from an individual image with arbitrary crowd density and arbitrary perspective. To this end, we have proposed a simple but effective Multi-column Convolutional Neural Network (MCNN) architecture to map the image to its crowd density map. The proposed MCNN allows the input image to be of arbitrary size or resolution. By utilizing filters with receptive fields of different sizes, the features learned by each column CNN are adaptive to variations in people head size due to perspective effect or image resolution. Furthermore, the true density map is computed accurately based on geometry-adaptive kernels which do not need knowing the perspective map of the input image. Since exiting crowd counting datasets do not adequately cover all the challenging situations considered in our work, we have collected and labelled a large new dataset that includes 1198 images with about 330,000 heads annotated. On this challenging new dataset, as well as all existing datasets, we conduct extensive experiments to verify the effectiveness of the proposed model and method. In particular, with the proposed simple MCNN model, our method outperforms all existing methods. In addition, experiments show that our model, once trained on one dataset, can be readily transferred to a new dataset.",
"The task of crowd counting is to automatically estimate the pedestrian number in crowd images. To cope with the scale and perspective changes that commonly exist in crowd images, state-of-the-art approaches employ multi-column CNN architectures to regress density maps of crowd images. Multiple columns have different receptive fields corresponding to pedestrians (heads) of different scales. We instead propose a scale-adaptive CNN (SaCNN) architecture with a backbone of fixed small receptive fields. We extract feature maps from multiple layers and adapt them to have the same output size; we combine them to produce the final density map. The number of people is computed by integrating the density map. We also introduce a relative count loss along with the density map loss to improve the network generalization on crowd scenes with few pedestrians, where most representative approaches perform poorly on. We conduct extensive experiments on the ShanghaiTech, UCF CC 50 and WorldExpo'10 datasets as well as a new dataset SmartCity that we collect for crowd scenes with few people. The results demonstrate significant improvements of SaCNN over the state-of-the-art.",
"Abstract Estimating count and density maps from crowd images has a wide range of applications such as video surveillance, traffic monitoring, public safety and urban planning. In addition, techniques developed for crowd counting can be applied to related tasks in other fields of study such as cell microscopy, vehicle counting and environmental survey. The task of crowd counting and density map estimation is riddled with many challenges such as occlusions, non-uniform density, intra-scene and inter-scene variations in scale and perspective. Nevertheless, over the last few years, crowd count analysis has evolved from earlier methods that are often limited to small variations in crowd density and scales to the current state-of-the-art methods that have developed the ability to perform successfully on a wide range of scenarios. The success of crowd counting methods in the recent years can be largely attributed to deep learning and publications of challenging datasets. In this paper, we provide a comprehensive survey of recent Convolutional Neural Network (CNN) based approaches that have demonstrated significant improvements over earlier methods that rely largely on hand-crafted representations. First, we briefly review the pioneering methods that use hand-crafted representations and then we delve in detail into the deep learning-based approaches and recently published datasets. Furthermore, we discuss the merits and drawbacks of existing CNN-based approaches and identify promising avenues of research in this rapidly evolving field.",
"We propose a novel crowd counting model that maps a given crowd scene to its density. Crowd analysis is compounded by myriad of factors like inter-occlusion between people due to extreme crowding, high similarity of appearance between people and background elements, and large variability of camera view-points. Current state-of-the art approaches tackle these factors by using multi-scale CNN architectures, recurrent networks and late fusion of features from multi-column CNN with different receptive fields. We propose switching convolutional neural network that leverages variation of crowd density within an image to improve the accuracy and localization of the predicted crowd count. Patches from a grid within a crowd scene are relayed to independent CNN regressors based on crowd count prediction quality of the CNN established during training. The independent CNN regressors are designed to have different receptive fields and a switch classifier is trained to relay the crowd scene patch to the best CNN regressor. We perform extensive experiments on all major crowd counting datasets and evidence better performance compared to current state-of-the-art methods. We provide interpretable representations of the multichotomy of space of crowd scene patches inferred from the switch. It is observed that the switch relays an image patch to a particular CNN column based on density of crowd.",
"We propose a network for Congested Scene Recognition called CSRNet to provide a data-driven and deep learning method that can understand highly congested scenes and perform accurate count estimation as well as present high-quality density maps. The proposed CSRNet is composed of two major components: a convolutional neural network (CNN) as the front-end for 2D feature extraction and a dilated CNN for the back-end, which uses dilated kernels to deliver larger reception fields and to replace pooling operations. CSRNet is an easy-trained model because of its pure convolutional structure. We demonstrate CSRNet on four datasets (ShanghaiTech dataset, the UCF_CC_50 dataset, the WorldEXPO'10 dataset, and the UCSD dataset) and we deliver the state-of-the-art performance. In the ShanghaiTech Part_B dataset, CSRNet achieves 47.3 lower Mean Absolute Error (MAE) than the previous state-of-the-art method. We extend the targeted applications for counting other objects, such as the vehicle in TRANCOS dataset. Results show that CSRNet significantly improves the output quality with 15.4 lower MAE than the previous state-of-the-art approach.",
"We present a novel method called Contextual Pyramid CNN (CP-CNN) for generating high-quality crowd density and count estimation by explicitly incorporating global and local contextual information of crowd images. The proposed CP-CNN consists of four modules: Global Context Estimator (GCE), Local Context Estimator (LCE), Density Map Estimator (DME) and a Fusion-CNN (F-CNN). GCE is a VGG-16 based CNN that encodes global context and it is trained to classify input images into different density classes, whereas LCE is another CNN that encodes local context information and it is trained to perform patch-wise classification of input images into different density classes. DME is a multi-column architecture-based CNN that aims to generate high-dimensional feature maps from the input image which are fused with the contextual information estimated by GCE and LCE using F-CNN. To generate high resolution and high-quality density maps, F-CNN uses a set of convolutional and fractionally-strided convolutional layers and it is trained along with the DME in an end-to-end fashion using a combination of adversarial loss and pixellevel Euclidean loss. Extensive experiments on highly challenging datasets show that the proposed method achieves significant improvements over the state-of-the-art methods."
]
} |
1812.02340 | 2902662809 | Investment decisions can benefit from incorporating an accumulated knowledge of the past to drive future decision making. We introduce Continual Learning Augmentation (CLA) which is based on an explicit memory structure and a feed forward neural network (FFNN) base model and used to drive long term financial investment decisions. We demonstrate that our approach improves accuracy in investment decision making while memory is addressed in an explainable way. Our approach introduces novel remember cues, consisting of empirically learned change points in the absolute error series of the FFNN. Memory recall is also novel, with contextual similarity assessed over time by sampling distances using dynamic time warping (DTW). We demonstrate the benefits of our approach by using it in an expected return forecasting task to drive investment decisions. In an investment simulation in a broad international equity universe between 2003-2017, our approach significantly outperforms FFNN base models. We also illustrate how CLA's memory addressing works in practice, using a worked example to demonstrate the explainability of our approach. | Approaches developed for continual learning have been applied to many areas and have employed a wide range of techniques including gated neural networks @cite_32 @cite_19 , explicit memory structures @cite_28 , prototypical addressing @cite_38 , weight adaptation @cite_1 @cite_22 to name a few. Having addressed catastrophic forgetting @cite_14 more recent research has turned to solving second order problems, such as overheads of external memory structures @cite_25 , problems with weight saturation @cite_12 and the complications of outright complexity @cite_39 . While a number of memory approaches apply to sequential memory tasks @cite_23 @cite_31 a far smaller number still have been focused specifically on time-series @cite_36 @cite_13 @cite_26 @cite_29 . It is unclear how effective these approaches would be in dealing with long term continual learning of noisy, non-stationary time-series, commonly found in finance. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_26",
"@cite_22",
"@cite_28",
"@cite_36",
"@cite_29",
"@cite_1",
"@cite_32",
"@cite_39",
"@cite_19",
"@cite_23",
"@cite_31",
"@cite_13",
"@cite_25",
"@cite_12"
],
"mid": [
"2950537964",
"2060277733",
"2963078493",
"2786465559",
"",
"1823409095",
"2624385633",
"1821462560",
"",
"2125308790",
"1924770834",
"",
"2530887700",
"2127141656",
"2952556884",
""
],
"abstract": [
"We propose prototypical networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend prototypical networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.",
"Abstract All natural cognitive systems, and, in particular, our own, gradually forget previously learned information. Plausible models of human cognition should therefore exhibit similar patterns of gradual forgetting of old information as new information is acquired. Only rarely does new learning in natural cognitive systems completely disrupt or erase previously learned information; that is, natural cognitive systems do not, in general, forget ‘catastrophically'. Unfortunately, though, catastrophic forgetting does occur under certain circumstances in distributed connectionist networks. The very features that give these networks their remarkable abilities to generalize, to function in the presence of degraded input, and so on, are found to be the root cause of catastrophic forgetting. The challenge in this field is to discover how to keep the advantages of distributed connectionist networks while avoiding the problem of catastrophic forgetting. In this article the causes, consequences and numerous solutions to the problem of catastrophic forgetting in neural networks are examined. The review will consider how the brain might have overcome this problem and will also explore the consequences of this solution for distributed connectionist networks.",
"Abstract: Clinical medical data, especially in the intensive care unit (ICU), consist of multivariate time series of observations. For each patient visit (or episode), sensor data and lab test results are recorded in the patient's Electronic Health Record (EHR). While potentially containing a wealth of insights, the data is difficult to mine effectively, owing to varying length, irregular sampling and missing data. Recurrent Neural Networks (RNNs), particularly those using Long Short-Term Memory (LSTM) hidden units, are powerful and increasingly popular models for learning from sequence data. They effectively model varying length sequences and capture long range dependencies. We present the first study to empirically evaluate the ability of LSTMs to recognize patterns in multivariate time series of clinical measurements. Specifically, we consider multilabel classification of diagnoses, training a model to classify 128 diagnoses given 13 frequently but irregularly sampled clinical measurements. First, we establish the effectiveness of a simple LSTM network for modeling clinical data. Then we demonstrate a straightforward and effective training strategy in which we replicate targets at each sequence step. Trained only on raw time series, our models outperform several strong baselines, including a multilayer perceptron trained on hand-engineered features.",
"Deep neural networks have excelled on a wide range of problems, from vision to language and game playing. Neural networks very gradually incorporate information into weights as they process data, requiring very low learning rates. If the training distribution shifts, the network is slow to adapt, and when it does adapt, it typically performs badly on the training distribution before the shift. Our method, Memory-based Parameter Adaptation, stores examples in memory and then uses a context-based lookup to directly modify the weights of a neural network. Much higher learning rates can be used for this local adaptation, reneging the need for many iterations over similar data before good predictions can be made. As our method is memory-based, it alleviates several shortcomings of neural networks, such as catastrophic forgetting, fast, stable acquisition of new knowledge, learning with an imbalanced class labels, and fast learning during evaluation. We demonstrate this on a range of supervised tasks: large-scale image classification and language modelling.",
"",
"Machine learning research has, to a great extent, ignored an important aspect of many real world applications: time. Existing concept learners predominantly operate on a static set of attributes; for example, classifying flowers described by leaf size, pet al colour and pet al count. However, many real datasets are not static; they cannot sensibly be represented as a fixed set of attributes. Rather, the examples are expressed as features that vary temporally, and it is the temporal variation itself that is used for classification. Consider a simple gesture recognition domain, in which the temporal features are the position of the hands, finger bends, and so on. Looking at the position of the hand at one point in time is not likely to lead to a successful classification; it is only by analysing changes in position that recognition is possible. This thesis presents a new technique for temporal classification. By extracting sub-events from the training instances and parameterising them to allow feature construction for a subsequent learning process, it is able to employ background knowledge and express learnt concepts in terms of the background knowledge. The novel results of the thesis are: a temporal learner capable of producing comprehensible and accurate classifiers for multivariate time series that can learn from a small number of instances and can integrate non-temporal features; a feature construction technique that parameterises sub-events of the training set and clusters them to construct features for a propositional learner; and a technique for post-processing classification rules produced by the learner to give a comprehensible description expressed in the same form as the original background knowledge. The thesis discusses the implementation of TClass, a temporal learner. Results show rules that are comprehensible in many cases and accuracy results close to or better than existing technique—over 98 per cent for sign language and 72 per cent for ECGs (equivalent to the accuracy of a human cardiologist). One further surprising result is that a small set of very primitive sub-events proves to be functional, avoiding the need for labour-intensive background knowledge if it is not available.",
"Long short-term memory (LSTM) networks are a state-of-the-art technique for sequence learning. They are less commonly applied to financial time series predictions, yet inherently suitable for this domain. We deploy LSTM networks for predicting out-of-sample directional movements for the constituent stocks of the S&P 500 from 1992 until 2015. With daily returns of 0.46 percent and a Sharpe ratio of 5.8 prior to transaction costs, we find LSTM networks to outperform memory-free classification methods, i.e., a random forest (RAF), a deep neural net (DNN), and a logistic regression classifier (LOG). The outperformance relative to the general market is very clear from 1992 to 2009, but as of 2010, excess returns seem to have been arbitraged away with LSTM profitability fluctuating around zero after transaction costs. We further unveil sources of profitability, thereby shedding light into the black box of artificial neural networks. Specifically, we find one common pattern among the stocks selected for trading – they exhibit high volatility and a short-term reversal return profile. Leveraging these findings, we are able to formalize a rules-based short-term reversal strategy that yields 0.23 percent prior to transaction costs. Further regression analysis unveils low exposure of the LSTM returns to common sources of systematic risk – also compared to the three benchmark models.",
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.",
"",
"The expressive power of a machine learning model is closely related to the number of sequential computational steps it can learn. For example, Deep Neural Networks have been more successful than shallow networks because they can perform a greater number of sequential computational steps (each highly parallel). The Neural Turing Machine (NTM) [8] is a model that can compactly express an even greater number of sequential computational steps, so it is even more powerful than a DNN. Its memory addressing operations are designed to be differentiable; thus the NTM can be trained with backpropagation. While differentiable memory is relatively easy to implement and train, it necessitates accessing the entire memory content at each computational step. This makes it difficult to implement a fast NTM. In this work, we use the Re inforce algorithm to learn where to access the memory, while using backpropagation to learn what to write to the memory. We call this model the RL-NTM. Reinforce allows our model to access a constant number of memory cells at each computational step, so its implementation can be faster. The RL-NTM is the first mo del that can, in principle, learn programs of unbounded running time. We successfully trained the RL-NTM to solve a number of algorithmic tasks that are simpler than the ones solvable by the fully differentiable NTM. As the RL-NTM is a fairly intricate model, we needed a method for verifying the correctness of our implementation. To do so, we developed a simple technique for numerically checking arbitrary implementations of models that use Reinforce, which may be of independent interest.",
"In this paper we compare different types of recurrent units in recurrent neural networks (RNNs). Especially, we focus on more sophisticated units that implement a gating mechanism, such as a long short-term memory (LSTM) unit and a recently proposed gated recurrent unit (GRU). We evaluate these recurrent units on the tasks of polyphonic music modeling and speech signal modeling. Our experiments revealed that these advanced recurrent units are indeed better than more traditional recurrent units such as tanh units. Also, we found GRU to be comparable to LSTM.",
"",
"A ‘differentiable neural computer’ is introduced that combines the learning capabilities of a neural network with an external memory analogous to the random-access memory in a conventional computer.",
"Many real-world sequence learning tasks require the prediction of sequences of labels from noisy, unsegmented input data. In speech recognition, for example, an acoustic signal is transcribed into words or sub-word units. Recurrent neural networks (RNNs) are powerful sequence learners that would seem well suited to such tasks. However, because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited. This paper presents a novel method for training RNNs to label unsegmented sequences directly, thereby solving both problems. An experiment on the TIMIT speech corpus demonstrates its advantages over both a baseline HMM and a hybrid HMM-RNN.",
"Neural networks augmented with external memory have the ability to learn algorithmic solutions to complex tasks. These models appear promising for applications such as language modeling and machine translation. However, they scale poorly in both space and time as the amount of memory grows --- limiting their applicability to real-world domains. Here, we present an end-to-end differentiable memory access scheme, which we call Sparse Access Memory (SAM), that retains the representational power of the original approaches whilst training efficiently with very large memories. We show that SAM achieves asymptotic lower bounds in space and time complexity, and find that an implementation runs @math faster and with @math less physical memory than non-sparse models. SAM learns with comparable data efficiency to existing models on a range of synthetic tasks and one-shot Omniglot character recognition, and can scale to tasks requiring @math s of time steps and memories. As well, we show how our approach can be adapted for models that maintain temporal associations between memories, as with the recently introduced Differentiable Neural Computer.",
""
]
} |
1812.02340 | 2902662809 | Investment decisions can benefit from incorporating an accumulated knowledge of the past to drive future decision making. We introduce Continual Learning Augmentation (CLA) which is based on an explicit memory structure and a feed forward neural network (FFNN) base model and used to drive long term financial investment decisions. We demonstrate that our approach improves accuracy in investment decision making while memory is addressed in an explainable way. Our approach introduces novel remember cues, consisting of empirically learned change points in the absolute error series of the FFNN. Memory recall is also novel, with contextual similarity assessed over time by sampling distances using dynamic time warping (DTW). We demonstrate the benefits of our approach by using it in an expected return forecasting task to drive investment decisions. In an investment simulation in a broad international equity universe between 2003-2017, our approach significantly outperforms FFNN base models. We also illustrate how CLA's memory addressing works in practice, using a worked example to demonstrate the explainability of our approach. | Memory modelling approaches using external memory structures requires an appropriate memory addressing mechanism (a way of storing and recalling a memory). Memory addressing is generally based on a similarity measure such as cosine similarity @cite_23 @cite_10 kernel weighting @cite_33 , use of linear models @cite_38 or instance-based similarities, many using K-nearest neighbours @cite_21 @cite_22 . However, these approaches are not obviously well suited to assessing the similarity of noisy and non-stationary, multivariate time-series. Euclidean distance offers a way to compare time-series but has a high sensitivity to the timing of data-points, something that has been addressed by dynamic time warping (DTW) @cite_18 @cite_5 . However, DTW requires normalized data @cite_27 and is also computationally expensive, although some mitigating measures have been developed (see @cite_8 ). | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_33",
"@cite_22",
"@cite_8",
"@cite_21",
"@cite_27",
"@cite_23",
"@cite_5",
"@cite_10"
],
"mid": [
"2950537964",
"2128160875",
"2432717477",
"2786465559",
"2586337417",
"",
"1853995153",
"",
"2098759488",
""
],
"abstract": [
"We propose prototypical networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend prototypical networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.",
"This paper reports on an optimum dynamic progxamming (DP) based time-normalization algorithm for spoken word recognition. First, a general principle of time-normalization is given using time-warping function. Then, two time-normalized distance definitions, called symmetric and asymmetric forms, are derived from the principle. These two forms are compared with each other through theoretical discussions and experimental studies. The symmetric form algorithm superiority is established. A new technique, called slope constraint, is successfully introduced, in which the warping function slope is restricted so as to improve discrimination between words in different categories. The effective slope constraint characteristic is qualitatively analyzed, and the optimum slope constraint condition is determined through experiments. The optimized algorithm is then extensively subjected to experimental comparison with various DP-algorithms, previously applied to spoken word recognition by different research groups. The experiment shows that the present algorithm gives no more than about two-thirds errors, even compared to the best conventional algorithm.",
"Learning from a few examples remains a key challenge in machine learning. Despite recent advances in important domains such as vision and language, the standard supervised deep learning paradigm does not offer a satisfactory solution for learning new concepts rapidly from little data. In this work, we employ ideas from metric learning based on deep neural features and from recent advances that augment neural networks with external memories. Our framework learns a network that maps a small labelled support set and an unlabelled example to its label, obviating the need for fine-tuning to adapt to new class types. We then define one-shot learning problems on vision (using Omniglot, ImageNet) and language tasks. Our algorithm improves one-shot accuracy on ImageNet from 87.6 to 93.2 and from 88.0 to 93.8 on Omniglot compared to competing approaches. We also demonstrate the usefulness of the same model on language modeling by introducing a one-shot task on the Penn Treebank.",
"Deep neural networks have excelled on a wide range of problems, from vision to language and game playing. Neural networks very gradually incorporate information into weights as they process data, requiring very low learning rates. If the training distribution shifts, the network is slow to adapt, and when it does adapt, it typically performs badly on the training distribution before the shift. Our method, Memory-based Parameter Adaptation, stores examples in memory and then uses a context-based lookup to directly modify the weights of a neural network. Much higher learning rates can be used for this local adaptation, reneging the need for many iterations over similar data before good predictions can be made. As our method is memory-based, it alleviates several shortcomings of neural networks, such as catastrophic forgetting, fast, stable acquisition of new knowledge, learning with an imbalanced class labels, and fast learning during evaluation. We demonstrate this on a range of supervised tasks: large-scale image classification and language modelling.",
"Dynamic Time Warping (DTW) is probably the most popular distance measure for time series data, because it captures flexible similarities under time distortions. However, DTW has long been suffering from the pathological alignment problem, and most existing solutions, which essentially impose rigid constraints on the warping path, are likely to miss the correct alignments. A crucial observation on pathological alignment is that it always leads to an abnormally large number of links between two sequences. Based on this new observation, we propose a novel variant of DTW called LDTW, which limits the total number of links during the optimization process of DTW. LDTW not only oppresses the pathological alignment effectively, but also allows more flexibilities when measuring similarities. It is a softer constraint because we still let the optimization process of DTW decide how many links to allocate to each data point and where to put these links. In this paper, we introduce the motivation and algorithm of LDTW and we conduct a nearest neighbor classification experiment on UCR time series archive to show its performance.",
"",
"In the last decade there has been an explosion of interest in mining time series data. Literally hundreds of papers have introduced new algorithms to index, classify, cluster and segment time series. In this work we make the following claim. Much of this work has very little utility because the contribution made (speed in the case of indexing, accuracy in the case of classification and clustering, model accuracy in the case of segmentation) offer an amount of “improvement” that would have been completely dwarfed by the variance that would have been observed by testing on many real world datasets, or the variance that would have been observed by changing minor (unstated) implementation details. To illustrate our point, we have undertaken the most exhaustive set of time series experiments ever attempted, re-implementing the contribution of more than two dozen papers, and testing them on 50 real world, highly diverse datasets. Our empirical results strongly support our assertion, and suggest the need for a set of time series benchmarks and more careful empirical evaluation in the data mining community.",
"",
"The last decade has witnessed a tremendous growths of interests in applications that deal with querying and mining of time series data. Numerous representation methods for dimensionality reduction and similarity measures geared towards time series have been introduced. Each individual work introducing a particular method has made specific claims and, aside from the occasional theoretical justifications, provided quantitative experimental observations. However, for the most part, the comparative aspects of these experiments were too narrowly focused on demonstrating the benefits of the proposed methods over some of the previously introduced ones. In order to provide a comprehensive validation, we conducted an extensive set of time series experiments re-implementing 8 different representation methods and 9 similarity measures and their variants, and testing their effectiveness on 38 time series data sets from a wide variety of application domains. In this paper, we give an overview of these different techniques and present our comparative experimental findings regarding their effectiveness. Our experiments have provided both a unified validation of some of the existing achievements, and in some cases, suggested that certain claims in the literature may be unduly optimistic.",
""
]
} |
1812.02591 | 2950387108 | Human motion prediction model has applications in various fields of computer vision. Without taking into account the inherent stochasticity in the prediction of future pose dynamics, such methods often converges to a deterministic undesired mean of multiple probable outcomes. Devoid of this, we propose a novel probabilistic generative approach called Bidirectional Human motion prediction GAN, or BiHMP-GAN. To be able to generate multiple probable human-pose sequences, conditioned on a given starting sequence, we introduce a random extrinsic factor r, drawn from a predefined prior distribution. Furthermore, to enforce a direct content loss on the predicted motion sequence and also to avoid mode-collapse, a novel bidirectional framework is incorporated by modifying the usual discriminator architecture. The discriminator is trained also to regress this extrinsic factor r, which is used alongside with the intrinsic factor (encoded starting pose sequence) to generate a particular pose sequence. To further regularize the training, we introduce a novel recursive prediction strategy. In spite of being in a probabilistic framework, the enhanced discriminator architecture allows predictions of an intermediate part of pose sequence to be used as a conditioning for prediction of the latter part of the sequence. The bidirectional setup also provides a new direction to evaluate the prediction quality against a given test sequence. For a fair assessment of BiHMP-GAN, we report performance of the generated motion sequence using (i) a critic model trained to discriminate between real and fake motion sequence, and (ii) an action classifier trained on real human motion dynamics. Outcomes of both qualitative and quantitative evaluations, on the probabilistic generations of the model, demonstrate the superiority of BiHMP-GAN over previously available methods. | Data-driven human motion prediction models have been explored by researchers for quite a along time in both computer animation and machine learning community. Before the deep-era various probabilistic graphical models have been tried to efficiently model human motion dynamics. Researchers have used time-series learning methods like Hidden Markov Model @cite_17 , restricted Boltzmann machines @cite_19 , Gaussian process @cite_14 , switching linear dynamical system @cite_12 to model human pose sequence data. However these methods fail to model the high-dimensional complex human pose sequence information effectively. Because of the highly nonlinear dependencies arose by the uncertainty in human movement, individually modeling various different factors affecting motion prediction does not scale well. These methods also suffer from complex training regime @cite_19 with complicated inference pipeline as a result of the acquired sampling technique. | {
"cite_N": [
"@cite_19",
"@cite_14",
"@cite_12",
"@cite_17"
],
"mid": [
"2158164339",
"",
"2110935301",
"1991942383"
],
"abstract": [
"We propose a non-linear generative model for human motion data that uses an undirected model with binary latent variables and real-valued \"visible\" variables that represent joint angles. The latent and visible variables at each time step receive directed connections from the visible variables at the last few time-steps. Such an architecture makes on-line inference efficient and allows us to use a simple approximate learning procedure. After training, the model finds a single set of parameters that simultaneously capture several different kinds of motion. We demonstrate the power of our approach by synthesizing various motion sequences and by performing on-line filling in of data lost during motion capture.",
"",
"The human figure exhibits complex and rich dynamic behavior that is both nonlinear and time-varying. Effective models of human dynamics can be learned from motion capture data using switching linear dynamic system (SLDS) models. We present results for human motion synthesis, classification, and visual tracking using learned SLDS models. Since exact inference in SLDS is intractable, we present three approximate inference algorithms and compare their performance. In particular, a new variational inference algorithm is obtained by casting the SLDS model as a Dynamic Bayesian Network. Classification experiments show the superiority of SLDS over conventional HMM's for our problem domain.",
"This paper describes a framework that allows a user to synthesize human motion while retaining control of its qualitative properties. The user paints a timeline with annotations --- like walk, run or jump --- from a vocabulary which is freely chosen by the user. The system then assembles frames from a motion database so that the final motion performs the specified actions at specified times. The motion can also be forced to pass through particular configurations at particular times, and to go to a particular position and orientation. Annotations can be painted positively (for example, must run), negatively (for example, may not run backwards) or as a don't-care. The system uses a novel search method, based around dynamic programming at several scales, to obtain a solution efficiently so that authoring is interactive. Our results demonstrate that the method can generate smooth, natural-looking motion.The annotation vocabulary can be chosen to fit the application, and allows specification of composite motions (run and jump simultaneously, for example). The process requires a collection of motion data that has been annotated with the chosen vocabulary. This paper also describes an effective tool, based around repeated use of support vector machines, that allows a user to annotate a large collection of motions quickly and easily so that they may be used with the synthesis algorithm."
]
} |
1812.02591 | 2950387108 | Human motion prediction model has applications in various fields of computer vision. Without taking into account the inherent stochasticity in the prediction of future pose dynamics, such methods often converges to a deterministic undesired mean of multiple probable outcomes. Devoid of this, we propose a novel probabilistic generative approach called Bidirectional Human motion prediction GAN, or BiHMP-GAN. To be able to generate multiple probable human-pose sequences, conditioned on a given starting sequence, we introduce a random extrinsic factor r, drawn from a predefined prior distribution. Furthermore, to enforce a direct content loss on the predicted motion sequence and also to avoid mode-collapse, a novel bidirectional framework is incorporated by modifying the usual discriminator architecture. The discriminator is trained also to regress this extrinsic factor r, which is used alongside with the intrinsic factor (encoded starting pose sequence) to generate a particular pose sequence. To further regularize the training, we introduce a novel recursive prediction strategy. In spite of being in a probabilistic framework, the enhanced discriminator architecture allows predictions of an intermediate part of pose sequence to be used as a conditioning for prediction of the latter part of the sequence. The bidirectional setup also provides a new direction to evaluate the prediction quality against a given test sequence. For a fair assessment of BiHMP-GAN, we report performance of the generated motion sequence using (i) a critic model trained to discriminate between real and fake motion sequence, and (ii) an action classifier trained on real human motion dynamics. Outcomes of both qualitative and quantitative evaluations, on the probabilistic generations of the model, demonstrate the superiority of BiHMP-GAN over previously available methods. | On the other hand, success of recurrent neural network (RNN) for modeling time-series data motivated researchers to effectively apply such architectures on human motion prediction task. Multitude of recent works @cite_13 @cite_8 successfully used variants of recurrent sequence-to-sequence architecture to model complex human skeleton dynamics. Such methods consider a seed motion sequence of certain time-step to condition prediction of future pose dynamics by employing an encoder-decoder recurrent pipeline. Ghosh al @cite_2 employ an additional non-recurrent encoder and decoder to explicitly leverage spatial structure and dependencies between joint locations to improve prediction quality of human pose sequence. Jain al @cite_6 proposed Structural-RNN to exploit the underlying spatio-temporal graph for modeling human skeleton dynamics. However all these methods do not consider the stochasticity in future pose dynamics by modeling it as a deterministic prediction problem. Hence, expressiveness of these approaches in modeling long-term motion sequence deteriorates as a result of convergence to a mean pose sequence. | {
"cite_N": [
"@cite_2",
"@cite_13",
"@cite_6",
"@cite_8"
],
"mid": [
"2606517404",
"2951188428",
"2176353499",
"2614034469"
],
"abstract": [
"We propose a new architecture for the learning of predictive spatio-temporal motion models from data alone. Our approach, dubbed the Dropout Autoencoder LSTM (DAELSTM), is capable of synthesizing natural looking motion sequences over long-time horizons1 without catastrophic drift or motion degradation. The model consists of two components, a 3-layer recurrent neural network to model temporal aspects and a novel autoencoder that is trained to implicitly recover the spatial structure of the human skeleton via randomly removing information about joints during training. This Dropout Autoencoder (DAE) is then used to filter each predicted pose by a 3-layer LSTM network, reducing accumulation of correlated error and hence drift over time. Furthermore to alleviate insufficiency of commonly used quality metric, we propose a new evaluation protocol using action classifiers to assess the quality of synthetic motion sequences. The proposed protocol can be used to assess quality of generated sequences of arbitrary length. Finally, we evaluate our proposed method on two of the largest motion-capture datasets available and show that our model outperforms the state-of-the-art techniques on a variety of actions, including cyclic and acyclic motion, and that it can produce natural looking sequences over longer time horizons than previous methods.",
"We propose the Encoder-Recurrent-Decoder (ERD) model for recognition and prediction of human body pose in videos and motion capture. The ERD model is a recurrent neural network that incorporates nonlinear encoder and decoder networks before and after recurrent layers. We test instantiations of ERD architectures in the tasks of motion capture (mocap) generation, body pose labeling and body pose forecasting in videos. Our model handles mocap training data across multiple subjects and activity domains, and synthesizes novel motions while avoid drifting for long periods of time. For human pose labeling, ERD outperforms a per frame body part detector by resolving left-right body part confusions. For video pose forecasting, ERD predicts body joint displacements across a temporal horizon of 400ms and outperforms a first order motion model based on optical flow. ERDs extend previous Long Short Term Memory (LSTM) models in the literature to jointly learn representations and their dynamics. Our experiments show such representation learning is crucial for both labeling and prediction in space-time. We find this is a distinguishing feature between the spatio-temporal visual domain in comparison to 1D text, speech or handwriting, where straightforward hard coded representations have shown excellent results when directly combined with recurrent units.",
"Deep Recurrent Neural Network architectures, though remarkably capable at modeling sequences, lack an intuitive high-level spatio-temporal structure. That is while many problems in computer vision inherently have an underlying high-level structure and can benefit from it. Spatio-temporal graphs are a popular tool for imposing such high-level intuitions in the formulation of real world problems. In this paper, we propose an approach for combining the power of high-level spatio-temporal graphs and sequence learning success of Recurrent Neural Networks (RNNs). We develop a scalable method for casting an arbitrary spatio-temporal graph as a rich RNN mixture that is feedforward, fully differentiable, and jointly trainable. The proposed method is generic and principled as it can be used for transforming any spatio-temporal graph through employing a certain set of well defined steps. The evaluations of the proposed approach on a diverse set of problems, ranging from modeling human motion to object interactions, shows improvement over the state-of-the-art with a large margin. We expect this method to empower new approaches to problem formulation through high-level spatio-temporal graphs and Recurrent Neural Networks.",
"Human motion modelling is a classical problem at the intersection of graphics and computer vision, with applications spanning human-computer interaction, motion synthesis, and motion prediction for virtual and augmented reality. Following the success of deep learning methods in several computer vision tasks, recent work has focused on using deep recurrent neural networks (RNNs) to model human motion, with the goal of learning time-dependent representations that perform tasks such as short-term motion prediction and long-term human motion synthesis. We examine recent work, with a focus on the evaluation methodologies commonly used in the literature, and show that, surprisingly, state-of-the-art performance can be achieved by a simple baseline that does not attempt to model motion at all. We investigate this result, and analyze recent RNN methods by looking at the architectures, loss functions, and training procedures used in state-of-the-art approaches. We propose three changes to the standard RNN models typically used for human motion, which result in a simple and scalable RNN architecture that obtains state-of-the-art performance on human motion prediction."
]
} |
1812.02591 | 2950387108 | Human motion prediction model has applications in various fields of computer vision. Without taking into account the inherent stochasticity in the prediction of future pose dynamics, such methods often converges to a deterministic undesired mean of multiple probable outcomes. Devoid of this, we propose a novel probabilistic generative approach called Bidirectional Human motion prediction GAN, or BiHMP-GAN. To be able to generate multiple probable human-pose sequences, conditioned on a given starting sequence, we introduce a random extrinsic factor r, drawn from a predefined prior distribution. Furthermore, to enforce a direct content loss on the predicted motion sequence and also to avoid mode-collapse, a novel bidirectional framework is incorporated by modifying the usual discriminator architecture. The discriminator is trained also to regress this extrinsic factor r, which is used alongside with the intrinsic factor (encoded starting pose sequence) to generate a particular pose sequence. To further regularize the training, we introduce a novel recursive prediction strategy. In spite of being in a probabilistic framework, the enhanced discriminator architecture allows predictions of an intermediate part of pose sequence to be used as a conditioning for prediction of the latter part of the sequence. The bidirectional setup also provides a new direction to evaluate the prediction quality against a given test sequence. For a fair assessment of BiHMP-GAN, we report performance of the generated motion sequence using (i) a critic model trained to discriminate between real and fake motion sequence, and (ii) an action classifier trained on real human motion dynamics. Outcomes of both qualitative and quantitative evaluations, on the probabilistic generations of the model, demonstrate the superiority of BiHMP-GAN over previously available methods. | HP-GAN @cite_11 first attempted to model human motion prediction as a probabilistic generative approach. They leverage recent advances in generative adversarial network (GAN) @cite_5 to adversarially train a recurrent motion prediction framework. However, they fail to assess expressiveness of such generative approach against deterministic counterparts. In contrast, the proposed incorporates novel modifications in architecture and training regime to improve expressiveness of the probabilistic method against available deterministic approaches. | {
"cite_N": [
"@cite_5",
"@cite_11"
],
"mid": [
"2099471712",
"2769460278"
],
"abstract": [
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.",
"Predicting and understanding human motion dynamics has many applications, such as motion synthesis, augmented reality, security, and autonomous vehicles. Due to the recent success of generative adversarial networks (GAN), there has been much interest in probabilistic estimation and synthetic data generation using deep neural network architectures and learning algorithms. We propose a novel sequence-to-sequence model for probabilistic human motion prediction, trained with a modified version of improved Wasserstein generative adversarial networks (WGAN-GP), in which we use a custom loss function designed for human motion prediction. Our model, which we call HP-GAN, learns a probability density function of future human poses conditioned on previous poses. It predicts multiple sequences of possible future human poses, each from the same input sequence but a different vector z drawn from a random distribution. Furthermore, to quantify the quality of the non-deterministic predictions, we simultaneously train a motion-quality-assessment model that learns the probability that a given skeleton sequence is a real human motion. We test our algorithm on two of the largest skeleton datasets: NTURGB-D and Human3.6M. We train our model on both single and multiple action types. Its predictive power for long-term motion estimation is demonstrated by generating multiple plausible futures of more than 30 frames from just 10 frames of input. We show that most sequences generated from the same input have more than 50 probabilities of being judged as a real human sequence. We will release all the code used in this paper to Github."
]
} |
1812.02350 | 2950905664 | Learning to insert an object instance into an image in a semantically coherent manner is a challenging and interesting problem. Solving it requires (a) determining a location to place an object in the scene and (b) determining its appearance at the location. Such an object insertion model can potentially facilitate numerous image editing and scene parsing applications. In this paper, we propose an end-to-end trainable neural network for the task of inserting an object instance mask of a specified class into the semantic label map of an image. Our network consists of two generative modules where one determines where the inserted object mask should be (i.e., location and scale) and the other determines what the object mask shape (and pose) should look like. The two modules are connected together via a spatial transformation network and jointly trained. We devise a learning procedure that leverage both supervised and unsupervised data and show our model can insert an object at diverse locations with various appearances. We conduct extensive experimental validations with comparisons to strong baselines to verify the effectiveness of the proposed network. | It mainly concerns with the geometric consistency between source and target images, which falls into the boarder category of scene and object structure reasoning @cite_3 @cite_23 @cite_8 . In @cite_34 , objects are detected based on the contextual information by modeling the statistics of low-level features of the object and surrounding scene. The same problem is recently addressed using a deep convolutional network @cite_30 . Both methods predict whether objects of interest are likely to appear at specific locations without determining additional information such as scale. In @cite_11 , an image composition method is developed where the bounding box position of an object is predicted based on object proposals and a new instance is retrieved from an existing database. However, this framework makes it significantly challenging to develop a joint generation model of shape and position since the objective function with respect to object location prediction is not differentiable. Recently, the ST-GAN @cite_9 utilizes spatial transformer network to insert objects based on image warping. However, this method does not synthesize plausible object instances in images. | {
"cite_N": [
"@cite_30",
"@cite_8",
"@cite_9",
"@cite_3",
"@cite_23",
"@cite_34",
"@cite_11"
],
"mid": [
"2592109610",
"2963049618",
"2794080999",
"2147513688",
"2141364309",
"2166761907",
""
],
"abstract": [
"Most of computer vision focuses on what is in an image. We propose to train a standalone object-centric context representation to perform the opposite task: seeing what is not there. Given an image, our context model can predict where objects should exist, even when no object instances are present. Combined with object detection results, we can perform a novel vision task: finding where objects are missing in an image. Our model is based on a convolutional neural network structure. With a specially designed training strategy, the model learns to ignore objects and focus on context only. It is fully convolutional thus highly efficient. Experiments show the effectiveness of the proposed approach in one important accessibility task: finding city street regions where curb ramps are missing, which could help millions of people with mobility disabilities.",
"In recent years, there has been a renewed interest in jointly modeling perception and action. At the core of this investigation is the idea of modeling affordances. However, when it comes to predicting affordances, even the state of the art approaches still do not use any ConvNets. Why is that? Unlike semantic or 3D tasks, there still does not exist any large-scale dataset for affordances. In this paper, we tackle the challenge of creating one of the biggest dataset for learning affordances. We use seven sitcoms to extract a diverse set of scenes and how actors interact with different objects in the scenes. Our dataset consists of more than 10K scenes and 28K ways humans can interact with these 10K images. We also propose a two-step approach to predict affordances in a new scene. In the first step, given a location in the scene we classify which of the 30 pose classes is the likely affordance pose. Given the pose class and the scene, we then use a Variational Autoencoder (VAE) to extract the scale and deformation of the pose. The VAE allows us to sample the distribution of possible poses at test time. Finally, we show the importance of large-scale data in learning a generalizable and robust model of affordances.",
"We address the problem of finding realistic geometric corrections to a foreground object such that it appears natural when composited into a background image. To achieve this, we propose a novel Generative Adversarial Network (GAN) architecture that utilizes Spatial Transformer Networks (STNs) as the generator, which we call Spatial Transformer GANs (ST-GANs). ST-GANs seek image realism by operating in the geometric warp parameter space. In particular, we exploit an iterative STN warping scheme and propose a sequential training strategy that achieves better results compared to naive training of a single generator. One of the key advantages of ST-GAN is its applicability to high-resolution images indirectly since the predicted warp parameters are transferable between reference frames. We demonstrate our approach in two applications: (1) visualizing how indoor furniture (e.g. from product images) might be perceived in a room, (2) hallucinating how accessories like glasses would look when matched with real portraits.",
"In recognizing objects and scenes, partial recognition of objects or their parts can be used to guide the recognition of other objects. Here, the role of individual objects in the recognition of complete figures and the influence of contextual information on the identification of ambiguous objects were investigated. Configurations of objects that were placed in either proper or improper spatial relations were used, and response times and error rates in a recognition task were measured. Two main results were obtained. First, proper spatial relations among the objects of a scene decrease response times and error rates in the recognition of individual objects. Second, the presence of objects that have a unique interpretation improves the identification of ambigous objects in the scene. Ambiguous objects were recognized faster and with fewer errors in the presence of clearly recognized objects compared with the same objects in isolation or in improper spatial relations. The implications of these findings for ...",
"This paper presents an empirical evaluation of the role of context in a contemporary, challenging object detection task - the PASCAL VOC 2008. Previous experiments with context have mostly been done on home-grown datasets, often with non-standard baselines, making it difficult to isolate the contribution of contextual information. In this work, we present our analysis on a standard dataset, using top-performing local appearance detectors as baseline. We evaluate several different sources of context and ways to utilize it. While we employ many contextual cues that have been used before, we also propose a few novel ones including the use of geographic context and a new approach for using object spatial support.",
"There is general consensus that context can be a rich source of information about an object's identity, location and scale. In fact, the structure of many real-world scenes is governed by strong configurational rules akin to those that apply to a single object. Here we introduce a simple framework for modeling the relationship between context and object properties based on the correlation between the statistics of low-level features across the entire scene and the objects that it contains. The resulting scheme serves as an effective procedure for object priming, context driven focus of attention and automatic scale-selection on real-world scenes.",
""
]
} |
1812.02350 | 2950905664 | Learning to insert an object instance into an image in a semantically coherent manner is a challenging and interesting problem. Solving it requires (a) determining a location to place an object in the scene and (b) determining its appearance at the location. Such an object insertion model can potentially facilitate numerous image editing and scene parsing applications. In this paper, we propose an end-to-end trainable neural network for the task of inserting an object instance mask of a specified class into the semantic label map of an image. Our network consists of two generative modules where one determines where the inserted object mask should be (i.e., location and scale) and the other determines what the object mask shape (and pose) should look like. The two modules are connected together via a spatial transformation network and jointly trained. We devise a learning procedure that leverage both supervised and unsupervised data and show our model can insert an object at diverse locations with various appearances. We conduct extensive experimental validations with comparisons to strong baselines to verify the effectiveness of the proposed network. | Object generation based on GANs has been studied extensively @cite_6 @cite_5 @cite_19 @cite_25 @cite_29 @cite_12 @cite_26 @cite_0 @cite_16 @cite_10 @cite_15 @cite_4 @cite_2 . Closest to this work are the methods designed to in-paint a desired patch @cite_31 or object (e.g., facial component @cite_24 , pedestrian @cite_7 ) in original images. In contrast, the object generation pipeline of this work is on the semantic layout rather than image domain. As a result, we simplify the network module on learning desirable object shapes. | {
"cite_N": [
"@cite_31",
"@cite_26",
"@cite_4",
"@cite_7",
"@cite_29",
"@cite_6",
"@cite_0",
"@cite_19",
"@cite_24",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_10",
"@cite_25",
"@cite_12"
],
"mid": [
"2342877626",
"2173520492",
"2962793481",
"2796322794",
"2592480533",
"2949212125",
"2567101557",
"2552465644",
"2611104282",
"2768959015",
"2797650215",
"2964024144",
"2553897675",
"2963800363",
"2766527293",
"2471149695"
],
"abstract": [
"We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders -- a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.",
"In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations.",
"Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.",
"State-of-the-art pedestrian detection models have achieved great success in many benchmarks. However, these models require lots of annotation information and the labeling process usually takes much time and efforts. In this paper, we propose a method to generate labeled pedestrian data and adapt them to support the training of pedestrian detectors. The proposed framework is built on the Generative Adversarial Network (GAN) with multiple discriminators, trying to synthesize realistic pedestrians and learn the background context simultaneously. To handle the pedestrians of different sizes, we adopt the Spatial Pyramid Pooling (SPP) layer in the discriminator. We conduct experiments on two benchmarks. The results show that our framework can smoothly synthesize pedestrians on background images of variations and different levels of details. To quantitatively evaluate our approach, we add the generated samples into training data of the baseline pedestrian detectors and show the synthetic images are able to improve the detectors' performance.",
"Unsupervised image-to-image translation aims at learning a joint distribution of images in different domains by using images from the marginal distributions in individual domains. Since there exists an infinite set of joint distributions that can arrive the given marginal distributions, one could infer nothing about the joint distribution from the marginal distributions without additional assumptions. To address the problem, we make a shared-latent space assumption and propose an unsupervised image-to-image translation framework based on Coupled GANs. We compare the proposed framework with competing approaches and present high quality image translation results on various challenging unsupervised image translation tasks, including street scene image translation, animal image translation, and face image translation. We also apply the proposed framework to domain adaptation and achieve state-of-the-art performance on benchmark datasets. Code and additional results are available in this https URL .",
"Collecting well-annotated image datasets to train modern machine learning algorithms is prohibitively expensive for many tasks. One appealing alternative is rendering synthetic data where ground-truth annotations are generated automatically. Unfortunately, models trained purely on rendered images often fail to generalize to real images. To address this shortcoming, prior work introduced unsupervised domain adaptation algorithms that attempt to map representations between the two domains or learn to extract features that are domain-invariant. In this work, we present a new approach that learns, in an unsupervised manner, a transformation in the pixel space from one domain to the other. Our generative adversarial network (GAN)-based method adapts source-domain images to appear as if drawn from the target domain. Our approach not only produces plausible samples, but also outperforms the state-of-the-art on a number of unsupervised domain adaptation scenarios by large margins. Finally, we demonstrate that the adaptation process generalizes to object classes unseen during training.",
"With recent progress in graphics, it has become more tractable to train models on synthetic images, potentially avoiding the need for expensive annotations. However, learning from synthetic images may not achieve the desired performance due to a gap between synthetic and real image distributions. To reduce this gap, we propose Simulated+Unsupervised (S+U) learning, where the task is to learn a model to improve the realism of a simulator's output using unlabeled real data, while preserving the annotation information from the simulator. We develop a method for S+U learning that uses an adversarial network similar to Generative Adversarial Networks (GANs), but with synthetic images as inputs instead of random vectors. We make several key modifications to the standard GAN algorithm to preserve annotations, avoid artifacts, and stabilize training: (i) a 'self-regularization' term, (ii) a local adversarial loss, and (iii) updating the discriminator using a history of refined images. We show that this enables generation of highly realistic images, which we demonstrate both qualitatively and with a user study. We quantitatively evaluate the generated images by training models for gaze estimation and hand pose estimation. We show a significant improvement over using synthetic images, and achieve state-of-the-art results on the MPIIGaze dataset without any labeled real data.",
"We investigate conditional adversarial networks as a general-purpose solution to image-to-image translation problems. These networks not only learn the mapping from input image to output image, but also learn a loss function to train this mapping. This makes it possible to apply the same generic approach to problems that traditionally would require very different loss formulations. We demonstrate that this approach is effective at synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images, among other tasks. Indeed, since the release of the pix2pix software associated with this paper, a large number of internet users (many of them artists) have posted their own experiments with our system, further demonstrating its wide applicability and ease of adoption without the need for parameter tweaking. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without hand-engineering our loss functions either.",
"In this paper, we propose an effective face completion algorithm using a deep generative model. Different from well-studied background completion, the face completion task is more challenging as it often requires to generate semantically new pixels for the missing key components (e.g., eyes and mouths) that contain large appearance variations. Unlike existing nonparametric algorithms that search for patches to synthesize, our algorithm directly generates contents for missing regions based on a neural network. The model is trained with a combination of a reconstruction loss, two adversarial losses and a semantic parsing loss, which ensures pixel faithfulness and local-global contents consistency. With extensive experimental results, we demonstrate qualitatively and quantitatively that our model is able to deal with a large area of missing pixels in arbitrary shapes and generate realistic face completion results.",
"Many image-to-image translation problems are ambiguous, as a single input image may correspond to multiple possible outputs. In this work, we aim to model a of possible outputs in a conditional generative modeling setting. The ambiguity of the mapping is distilled in a low-dimensional latent vector, which can be randomly sampled at test time. A generator learns to map the given input, combined with this latent code, to the output. We explicitly encourage the connection between output and the latent code to be invertible. This helps prevent a many-to-one mapping from the latent code to the output during training, also known as the problem of mode collapse, and produces more diverse results. We explore several variants of this approach by employing different training objectives, network architectures, and methods of injecting the latent code. Our proposed method encourages bijective consistency between the latent encoding and output modes. We present a systematic comparison of our method and other variants on both perceptual realism and diversity.",
"Unsupervised image-to-image translation is an important and challenging problem in computer vision. Given an image in the source domain, the goal is to learn the conditional distribution of corresponding images in the target domain, without seeing any pairs of corresponding images. While this conditional distribution is inherently multimodal, existing approaches make an overly simplified assumption, modeling it as a deterministic one-to-one mapping. As a result, they fail to generate diverse outputs from a given source domain image. To address this limitation, we propose a Multimodal Unsupervised Image-to-image Translation (MUNIT) framework. We assume that the image representation can be decomposed into a content code that is domain-invariant, and a style code that captures domain-specific properties. To translate an image to another domain, we recombine its content code with a random style code sampled from the style space of the target domain. We analyze the proposed framework and establish several theoretical results. Extensive experiments with comparisons to the state-of-the-art approaches further demonstrates the advantage of the proposed framework. Moreover, our framework allows users to control the style of translation outputs by providing an example style image. Code and pretrained models are available at this https URL",
"Synthesizing high-quality images from text descriptions is a challenging problem in computer vision and has many practical applications. Samples generated by existing textto- image approaches can roughly reflect the meaning of the given descriptions, but they fail to contain necessary details and vivid object parts. In this paper, we propose Stacked Generative Adversarial Networks (StackGAN) to generate 256.256 photo-realistic images conditioned on text descriptions. We decompose the hard problem into more manageable sub-problems through a sketch-refinement process. The Stage-I GAN sketches the primitive shape and colors of the object based on the given text description, yielding Stage-I low-resolution images. The Stage-II GAN takes Stage-I results and text descriptions as inputs, and generates high-resolution images with photo-realistic details. It is able to rectify defects in Stage-I results and add compelling details with the refinement process. To improve the diversity of the synthesized images and stabilize the training of the conditional-GAN, we introduce a novel Conditioning Augmentation technique that encourages smoothness in the latent conditioning manifold. Extensive experiments and comparisons with state-of-the-arts on benchmark datasets demonstrate that the proposed method achieves significant improvements on generating photo-realistic images conditioned on text descriptions.",
"We study the problem of transferring a sample in one domain to an analog sample in another domain. Given two related domains, S and T, we would like to learn a generative function G that maps an input sample from S to the domain T, such that the output of a given function f, which accepts inputs in either domains, would remain unchanged. Other than the function f, the training data is unsupervised and consist of a set of samples from each domain. The Domain Transfer Network (DTN) we present employs a compound loss function that includes a multiclass GAN loss, an f-constancy component, and a regularizing component that encourages G to map samples from T to themselves. We apply our method to visual domains including digits and face images and demonstrate its ability to generate convincing novel images of previously unseen entities, while preserving their identity.",
"We present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048 A— 1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.",
"We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, e.g., CelebA images at 1024^2. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally, we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CelebA dataset.",
"We propose coupled generative adversarial network (CoGAN) for learning a joint distribution of multi-domain images. In contrast to the existing approaches, which require tuples of corresponding images in different domains in the training set, CoGAN can learn a joint distribution without any tuple of corresponding images. It can learn a joint distribution with just samples drawn from the marginal distributions. This is achieved by enforcing a weight-sharing constraint that limits the network capacity and favors a joint distribution solution over a product of marginal distributions one. We apply CoGAN to several joint distribution learning tasks, including learning a joint distribution of color and depth images, and learning a joint distribution of face images with different attributes. For each task it successfully learns the joint distribution without any tuple of corresponding images. We also demonstrate its applications to domain adaptation and image transformation."
]
} |
1812.02350 | 2950905664 | Learning to insert an object instance into an image in a semantically coherent manner is a challenging and interesting problem. Solving it requires (a) determining a location to place an object in the scene and (b) determining its appearance at the location. Such an object insertion model can potentially facilitate numerous image editing and scene parsing applications. In this paper, we propose an end-to-end trainable neural network for the task of inserting an object instance mask of a specified class into the semantic label map of an image. Our network consists of two generative modules where one determines where the inserted object mask should be (i.e., location and scale) and the other determines what the object mask shape (and pose) should look like. The two modules are connected together via a spatial transformation network and jointly trained. We devise a learning procedure that leverage both supervised and unsupervised data and show our model can insert an object at diverse locations with various appearances. We conduct extensive experimental validations with comparisons to strong baselines to verify the effectiveness of the proposed network. | Several methods have been developed to model what and where factors in different tasks. @cite_17 present a method to generate an image from the text input, which can either be conditioned on the given bounding boxes or keypoints. In @cite_8 , model the distribution of possible human poses using VAE at different locations of indoor scenes. However, both methods do not deal with the distribution of possible location conditioned on the objects in the scene. @cite_28 model the object location, scale and shape to generate a new image from a single text input. Different from the proposed approach, none of the existing methods are constructed based on end-to-end trainable networks, in which location prediction, as well as object generation, can be regularized and optimized jointly. | {
"cite_N": [
"@cite_28",
"@cite_8",
"@cite_17"
],
"mid": [
"2964216930",
"2963049618",
"2530372461"
],
"abstract": [
"We propose a novel hierarchical approach for text-to-image synthesis by inferring semantic layout. Instead of learning a direct mapping from text to image, our algorithm decomposes the generation process into multiple steps, in which it first constructs a semantic layout from the text by the layout generator and converts the layout to an image by the image generator. The proposed layout generator progressively constructs a semantic layout in a coarse-to-fine manner by generating object bounding boxes and refining each box by estimating object shapes inside the box. The image generator synthesizes an image conditioned on the inferred semantic layout, which provides a useful semantic structure of an image matching with the text description. Our model not only generates semantically more meaningful images, but also allows automatic annotation of generated images and user-controlled generation process by modifying the generated scene layout. We demonstrate the capability of the proposed model on challenging MS-COCO dataset and show that the model can substantially improve the image quality, interpretability of output and semantic alignment to input text over existing approaches.",
"In recent years, there has been a renewed interest in jointly modeling perception and action. At the core of this investigation is the idea of modeling affordances. However, when it comes to predicting affordances, even the state of the art approaches still do not use any ConvNets. Why is that? Unlike semantic or 3D tasks, there still does not exist any large-scale dataset for affordances. In this paper, we tackle the challenge of creating one of the biggest dataset for learning affordances. We use seven sitcoms to extract a diverse set of scenes and how actors interact with different objects in the scenes. Our dataset consists of more than 10K scenes and 28K ways humans can interact with these 10K images. We also propose a two-step approach to predict affordances in a new scene. In the first step, given a location in the scene we classify which of the 30 pose classes is the likely affordance pose. Given the pose class and the scene, we then use a Variational Autoencoder (VAE) to extract the scale and deformation of the pose. The VAE allows us to sample the distribution of possible poses at test time. Finally, we show the importance of large-scale data in learning a generalizable and robust model of affordances.",
"Generative Adversarial Networks (GANs) have recently demonstrated the capability to synthesize compelling real-world images, such as room interiors, album covers, manga, faces, birds, and flowers. While existing models can synthesize images based on global constraints such as a class label or caption, they do not provide control over pose or object location. We propose a new model, the Generative Adversarial What-Where Network (GAWWN), that synthesizes images given instructions describing what content to draw in which location. We show high-quality 128 × 128 image synthesis on the Caltech-UCSD Birds dataset, conditioned on both informal text descriptions and also object location. Our system exposes control over both the bounding box around the bird and its constituent parts. By modeling the conditional distributions over part locations, our system also enables conditioning on arbitrary subsets of parts (e.g. only the beak and tail), yielding an efficient interface for picking part locations."
]
} |
1812.02162 | 2902709102 | Variations in visual factors such as viewpoint, pose, illumination and background, are usually viewed as important challenges in person re-identification (re-ID). In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking. To derive insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us have a deeper understanding of the fundamental problems in person re-ID. Our research also provides useful insights for dataset building and future practical usage, e.g., a person of a side view makes a better query. | To reduce the influence of pose variations, some work @cite_26 @cite_22 @cite_9 @cite_18 @cite_8 proposed to learn pose invariant representation for pedestrian. For example, Farenza al @cite_26 present two axes dependent on the pose of the body to learn a feature description with pose invariance. Cho al @cite_9 use the estimation of target poses to conduct multi-shot matching to reduce the influence of diverse poses. Zheng al @cite_29 design a PoseBox Fusion Network to learn a pedestrian descriptor (pose invariant embedding) based on PoseBox and jointly learning strategy. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_22",
"@cite_8",
"@cite_9",
"@cite_29"
],
"mid": [
"2760299260",
"1979260620",
"",
"2769925982",
"",
"2583528645"
],
"abstract": [
"Feature extraction and matching are two crucial components in person Re-Identification (ReID). The large pose deformations and the complex view variations exhibited by the captured person images significantly increase the difficulty of learning and matching of the features from person images. To overcome these difficulties, in this work we propose a Pose-driven Deep Convolutional (PDC) model to learn improved feature extraction and matching models from end to end. Our deep architecture explicitly leverages the human part cues to alleviate the pose variations and learn robust feature representations from both the global image and different local parts. To match the features from global human body and local body parts, a pose driven feature weighting sub-network is further designed to learn adaptive feature fusions. Extensive experimental analyses and results on three popular datasets demonstrate significant performance improvements of our model over all published state-of-the-art methods.",
"In this paper, we present an appearance-based method for person re-identification. It consists in the extraction of features that model three complementary aspects of the human appearance: the overall chromatic content, the spatial arrangement of colors into stable regions, and the presence of recurrent local motifs with high entropy. All this information is derived from different body parts, and weighted opportunely by exploiting symmetry and asymmetry perceptual principles. In this way, robustness against very low resolution, occlusions and pose, viewpoint and illumination changes is achieved. The approach applies to situations where the number of candidates varies continuously, considering single images or bunch of frames for each individual. It has been tested on several public benchmark datasets (ViPER, iLIDS, ETHZ), gaining new state-of-the-art performances.",
"",
"Person re identification is a challenging retrieval task that requires matching a person's acquired image across non overlapping camera views. In this paper we propose an effective approach that incorporates both the fine and coarse pose information of the person to learn a discriminative embedding. In contrast to the recent direction of explicitly modeling body parts or correcting for misalignment based on these, we show that a rather straightforward inclusion of acquired camera view and or the detected joint locations into a convolutional neural network helps to learn a very effective representation. To increase retrieval performance, re-ranking techniques based on computed distances have recently gained much attention. We propose a new unsupervised and automatic re-ranking framework that achieves state-of-the-art re-ranking performance. We show that in contrast to the current state-of-the-art re-ranking methods our approach does not require to compute new rank lists for each image pair (e.g., based on reciprocal neighbors) and performs well by using simple direct rank list based comparison or even by just using the already computed euclidean distances between the images. We show that both our learned representation and our re-ranking method achieve state-of-the-art performance on a number of challenging surveillance image and video datasets. The code is available online at: this https URL",
"",
"Pedestrian misalignment, which mainly arises from detector errors and pose variations, is a critical problem for a robust person re-identification (re-ID) system. With bad alignment, the background noise will significantly compromise the feature learning and matching process. To address this problem, this paper introduces the pose invariant embedding (PIE) as a pedestrian descriptor. First, in order to align pedestrians to a standard pose, the PoseBox structure is introduced, which is generated through pose estimation followed by affine transformations. Second, to reduce the impact of pose estimation errors and information loss during PoseBox construction, we design a PoseBox fusion (PBF) CNN architecture that takes the original image, the PoseBox, and the pose estimation confidence as input. The proposed PIE descriptor is thus defined as the fully connected layer of the PBF network for the retrieval task. Experiments are conducted on the Market-1501, CUHK03, and VIPeR datasets. We show that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with the state-of-the-art approaches."
]
} |
1812.02162 | 2902709102 | Variations in visual factors such as viewpoint, pose, illumination and background, are usually viewed as important challenges in person re-identification (re-ID). In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking. To derive insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us have a deeper understanding of the fundamental problems in person re-ID. Our research also provides useful insights for dataset building and future practical usage, e.g., a person of a side view makes a better query. | There are some works being related to reducing the influence of background @cite_1 @cite_13 @cite_14 @cite_6 @cite_10 . For instance, to reduce the influence of background, Chen al @cite_14 learn from foreground person and the original image, individually. In contrast, Song al @cite_6 think background can prove useful information, so they propose to segment body first, and then learn representation from body and background regions of images, respectively. | {
"cite_N": [
"@cite_14",
"@cite_10",
"@cite_1",
"@cite_6",
"@cite_13"
],
"mid": [
"2884030607",
"2798550112",
"2128648922",
"2798775284",
"2963574614"
],
"abstract": [
"In this work, we tackle the problem of person search, which is a challenging task consisted of pedestrian detection and person re-identification (re-ID). Instead of sharing representations in a single joint model, we find that separating detector and re-ID feature extraction yields better performance. In order to extract more representative features for each identity, we propose a simple yet effective re-ID method, which models foreground person and original image patches individually, and obtains enriched representations from two separate CNN streams. On the standard person search benchmark datasets, we achieve mAP of (83.0 ) and (32.6 ) respectively for CUHK-SYSU and PRW, surpassing the state of the art by a large margin (more than 5 pp).",
"Person re-identification is an important topic in intelligent surveillance and computer vision. It aims to accurately measure visual similarities between person images for determining whether two images correspond to the same person. State-of-the-art methods mainly utilize deep learning based approaches for learning visual features for describing person appearances. However, we observe that existing deep learning models are biased to capture too much relevance between background appearances of person images. We design a series of experiments with newly created datasets to validate the influence of background information. To solve the background bias problem, we propose a person-region guided pooling deep neural network based on human parsing maps to learn more discriminative person-part features, and propose to augment training data with person images with random background. Extensive experiments demonstrate the robustness and effectiveness of our proposed method.",
"Highlights? We propose a novel appearance-based method for person re-identification. ? It condenses a set of frames of the same individual into a informative signature. ? It incorporates complementary global and local statistical descriptions of the human appearance. ? Semantic segmentation of objects is exploited to define a part-based descriptor. ? The resulting descriptor define state-of-the-art results on the considered datasets. We propose a novel appearance-based method for person re-identification, that condenses a set of frames of an individual into a highly informative signature, called the Histogram Plus Epitome, HPE. It incorporates complementary global and local statistical descriptions of the human appearance, focusing on the overall chromatic content via histogram representation, and on the presence of recurrent local patches via epitomic analysis. The re-identification performance of HPE is then augmented by applying it as human part descriptor, defining a structured feature called asymmetry-based HPE (AHPE). The matching between (A)HPEs provides optimal performances against low resolution, occlusions, pose and illumination variations, defining state-of-the-art results on all the considered datasets.",
"Person Re-identification (ReID) is an important yet challenging task in computer vision. Due to the diverse background clutters, variations on viewpoints and body poses, it is far from solved. How to extract discriminative and robust features invariant to background clutters is the core problem. In this paper, we first introduce the binary segmentation masks to construct synthetic RGB-Mask pairs as inputs, then we design a mask-guided contrastive attention model (MGCAM) to learn features separately from the body and background regions. Moreover, we propose a novel region-level triplet loss to restrain the features learnt from different regions, i.e., pulling the features from the full image and body region close, whereas pushing the features from backgrounds away. We may be the first one to successfully introduce the binary mask into person ReID task and the first one to propose region-level contrastive learning. We evaluate the proposed method on three public datasets, including MARS, Market-1501 and CUHK03. Extensive experimental results show that the proposed method is effective and achieves the state-of-the-art results. Mask and code will be released upon request.",
"Existing person re-identification benchmarks and methods mainly focus on matching cropped pedestrian images between queries and candidates. However, it is different from real-world scenarios where the annotations of pedestrian bounding boxes are unavailable and the target person needs to be searched from a gallery of whole scene images. To close the gap, we propose a new deep learning framework for person search. Instead of breaking it down into two separate tasks—pedestrian detection and person re-identification, we jointly handle both aspects in a single convolutional neural network. An Online Instance Matching (OIM) loss function is proposed to train the network effectively, which is scalable to datasets with numerous identities. To validate our approach, we collect and annotate a large-scale benchmark dataset for person search. It contains 18,184 images, 8,432 identities, and 96,143 pedestrian bounding boxes. Experiments show that our framework outperforms other separate approaches, and the proposed OIM loss function converges much faster and better than the conventional Softmax loss."
]
} |
1812.02162 | 2902709102 | Variations in visual factors such as viewpoint, pose, illumination and background, are usually viewed as important challenges in person re-identification (re-ID). In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking. To derive insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us have a deeper understanding of the fundamental problems in person re-ID. Our research also provides useful insights for dataset building and future practical usage, e.g., a person of a side view makes a better query. | Resolution influences the detailed information of the content, so that affects the discrimination of feature learned from different resolution images. To solve this problem, Jing al @cite_19 propose a mapping function from height-resolution gallery images and low-resolution probe images to reduce the influence of noises. Li @cite_20 al joint learns from multi-scale to get a shared subspace across different scales. Wang al @cite_17 fuses embeddings at lower layers with higher resolution images and higher layers of the network to learn resolution invariance feature. | {
"cite_N": [
"@cite_19",
"@cite_20",
"@cite_17"
],
"mid": [
"1897123318",
"",
"2798794112"
],
"abstract": [
"Person re-identification has been widely studied due to its importance in surveillance and forensics applications. In practice, gallery images are high-resolution (HR) while probe images are usually low-resolution (LR) in the identification scenarios with large variation of illumination, weather or quality of cameras. Person re-identification in this kind of scenarios, which we call super-resolution (SR) person re-identification, has not been well studied. In this paper, we propose a semi-coupled low-rank discriminant dictionary learning (SLD2L) approach for SR person re-identification. For the given training image set which consists of HR gallery and LR probe images, we aim to convert the features of LR images into discriminating HR features. Specifically, our approach learns a pair of HR and LR dictionaries and a mapping from the features of HR gallery images and LR probe images. To ensure that the converted features using the learned dictionaries and mapping have favorable discriminative capability, we design a discriminant term which requires the converted HR features of LR probe images should be close to the features of HR gallery images from the same person, but far away from the features of HR gallery images from different persons. In addition, we apply low-rank regularization in dictionary learning procedure such that the learned dictionaries can well characterize intrinsic feature space of HR and LR images. Experimental results on public datasets demonstrate the effectiveness of SLD2L.",
"",
"Not all people are equally easy to identify: color statistics might be enough for some cases while others might require careful reasoning about high- and low-level details. However, prevailing person re-identification(re-ID) methods use one-size-fits-all high-level embeddings from deep convolutional networks for all cases. This might limit their accuracy on difficult examples or makes them needlessly expensive for the easy ones. To remedy this, we present a new person re-ID model that combines effective embeddings built on multiple convolutional network layers, trained with deep-supervision. On traditional re-ID benchmarks, our method improves substantially over the previous state-of-the-art results on all five datasets that we evaluate on. We then propose two new formulations of the person re-ID problem under resource-constraints, and show how our model can be used to effectively trade off accuracy and computation in the presence of resource constraints. Code and pre-trained models are available at this https URL."
]
} |
1812.02162 | 2902709102 | Variations in visual factors such as viewpoint, pose, illumination and background, are usually viewed as important challenges in person re-identification (re-ID). In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking. To derive insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us have a deeper understanding of the fundamental problems in person re-ID. Our research also provides useful insights for dataset building and future practical usage, e.g., a person of a side view makes a better query. | The research to reduce influence of viewpoint variations @cite_21 @cite_28 @cite_2 @cite_7 @cite_15 also focuses on learning invariant feature. For example, Gray al @cite_21 design an ensemble of localized features to reduce the influence of the viewpoint changes for person representation. Karanam al @cite_7 learn a single viewpoint invariance dictionary to represent both gallery and probe images during training, then train the dictionary by enforcing explicit constraints on the associated sparse representations of the feature vectors. | {
"cite_N": [
"@cite_7",
"@cite_28",
"@cite_21",
"@cite_2",
"@cite_15"
],
"mid": [
"2228002889",
"2018507448",
"1518138188",
"1965301261",
"2780159079"
],
"abstract": [
"This paper introduces a new approach to address the person re-identification problem in cameras with non-overlapping fields of view. Unlike previous approaches that learn Mahalanobis-like distance metrics in some transformed feature space, we propose to learn a dictionary that is capable of discriminatively and sparsely encoding features representing different people. Our approach directly addresses two key challenges in person re-identification: viewpoint variations and discriminability. First, to tackle viewpoint and associated appearance changes, we learn a single dictionary to represent both gallery and probe images in the training phase. We then discriminatively train the dictionary by enforcing explicit constraints on the associated sparse representations of the feature vectors. In the testing phase, we re-identify a probe image by simply determining the gallery image that has the closest sparse representation to that of the probe image in the Euclidean sense. Extensive performance evaluations on three publicly available multi-shot re-identification datasets demonstrate the advantages of our algorithm over several state-of-the-art dictionary learning, temporal sequence matching, and spatial appearance and metric learning based techniques.",
"Human re-identification across cameras with non-overlapping fields of view is one of the most important and difficult problems in video surveillance and analysis. However, current algorithms are likely to fail in real-world scenarios for several reasons. For example, surveillance cameras are typically mounted high above the ground plane, causing serious perspective changes. Also, most algorithms approach matching across images using the same descriptors, regardless of camera viewpoint or human pose. Here, we introduce a re-identification algorithm that addresses both problems. We build a model for human appearance as a function of pose, using training data gathered from a calibrated camera. We then apply this “pose prior” in online re-identification to make matching and identification more robust to viewpoint. We further integrate person-specific features learned over the course of tracking to improve the algorithm’s performance. We evaluate the performance of the proposed algorithm and compare it to several state-of-the-art algorithms, demonstrating superior performance on standard benchmarking datasets as well as a challenging new airport surveillance scenario.",
"Viewpoint invariant pedestrian recognition is an important yet under-addressed problem in computer vision. This is likely due to the difficulty in matching two objects with unknown viewpoint and pose. This paper presents a method of performing viewpoint invariant pedestrian recognition using an efficiently and intelligently designed object representation, the ensemble of localized features (ELF). Instead of designing a specific feature by hand to solve the problem, we define a feature space using our intuition about the problem and let a machine learning algorithm find the best representation. We show how both an object class specific representation and a discriminative recognition model can be learned using the AdaBoost algorithm. This approach allows many different kinds of simple features to be combined into a single similarity function. The method is evaluated using a viewpoint invariant pedestrian recognition dataset and the results are shown to be superior to all previous benchmarks for both recognition and reacquisition of pedestrians.",
"Re-identifying people in a network of cameras requires an invariant human representation. State of the art algorithms are likely to fail in real-world scenarios due to serious perspective changes. Most of existing approaches focus on invariant and discriminative features, while ignoring the body alignment issue. In this paper we propose 3 methods for improving the performance of person re-identification. We focus on eliminating perspective distortions by using 3D scene information. Perspective changes are minimized by affine transformations of cropped images containing the target (1). Further we estimate the human pose for (2) clustering data from a video stream and (3) weighting image features. The pose is estimated using 3D scene information and motion of the target. We validated our approach on a publicly available dataset with a network of 8 cameras. The results demonstrated significant increase in the re-identification performance over the state of the art.",
"In this paper, we study the problem of Cross-View Person Identification (CVPI), which aims at identifying the same person from temporally synchronized videos taken by different wearable cameras. Our basic idea is to utilize the human motion consistency for CVPI, where human motion can be computed by optical flow. However, optical flow is view-variant - the same person's optical flow in different videos can be very different due to view angle change. In this paper, we attempt to utilize 3D human-skeleton sequences to learn a model that can extract view-invariant motion features from optical flows in different views. For this purpose, we use 3D Mocap database to build a synthetic optical flow dataset and train a Triplet Network (TN) consisting of three sub-networks: two for optical flow sequences from different views and one for the underlying 3D Mocap skeleton sequence. Finally, sub-networks for optical flows are used to extract view-invariant features for CVPI. Experimental results show that, using only the motion information, the proposed method can achieve comparable performance with the state-of-the-art methods. Further combination of the proposed method with an appearance-based method achieves new state-of-the-art performance."
]
} |
1812.02162 | 2902709102 | Variations in visual factors such as viewpoint, pose, illumination and background, are usually viewed as important challenges in person re-identification (re-ID). In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking. To derive insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us have a deeper understanding of the fundamental problems in person re-ID. Our research also provides useful insights for dataset building and future practical usage, e.g., a person of a side view makes a better query. | Modern computer vision research often needs expensive data acquisition and accurate manual labeling. To address this problem, some work uses 3D synthetic data to assist computer vision problems such as semantic segmentation @cite_12 , object tracking @cite_27 , traffic vision research @cite_24 For person re-ID, SOMAset @cite_3 is a synthetic dataset containing 50 characters and 11 types of outfit. Barbosa al build the SOMAset dataset to train a network SOMAnet and find SOMAnet can identify people even if they change apparel between cameras. Bak al @cite_25 also create a synthetic dataset SyRI including 100 characters to solve the limitation of lighting conditions and use domain adaptation technology to assist the model training of new datasets. In this paper, we aim at analyzing how visual factors influence the identification system by using the controllability of synthetic data. | {
"cite_N": [
"@cite_3",
"@cite_24",
"@cite_27",
"@cite_25",
"@cite_12"
],
"mid": [
"2583147637",
"2889054948",
"2949907962",
"2951220054",
"2949255235"
],
"abstract": [
"Abstract Re-identification is generally carried out by encoding the appearance of a subject in terms of outfit, suggesting scenarios where people do not change their attire. In this paper we overcome this restriction, by proposing a framework based on a deep convolutional neural network, SOMAnet, that additionally models other discriminative aspects, namely, structural attributes of the human figure (e.g. height, obesity, gender). Our method is unique in many respects. First, SOMAnet is based on the Inception architecture, departing from the usual siamese framework. This spares expensive data preparation (pairing images across cameras) and allows the understanding of what the network learned. Second, and most notably, the training data consists of a synthetic 100K instance dataset, SOMAset, created by photorealistic human body generation software. SOMAset will be released with a open source license to enable further developments in re-identification. Synthetic data represents a cost-effective way of acquiring semi-realistic imagery (full realism is usually not required in re-identification since surveillance cameras capture low-resolution silhouettes), while at the same time providing complete control of the samples in terms of ground truth. Thus it is relatively easy to customize the data w.r.t. the surveillance scenario at-hand, e.g. ethnicity. SOMAnet, trained on SOMAset and fine-tuned on recent re-identification benchmarks, matches subjects even with different apparel.",
"Dataset plays an essential role in the training and testing of traffic vision algorithms. However, the collection and annotation of images from the real world is time-consuming, labor-intensive, and error-prone. Therefore, more and more researchers have begun to explore the virtual dataset, to overcome the disadvantages of real datasets. In this paper, we propose a systematic method to construct large-scale artificial scenes and collect a new virtual dataset (named “ParallelEye”) for the traffic vision research. The Unity3D rendering software is used to simulate environmental changes in the artificial scenes and generate ground-truth labels automatically, including semantic instance segmentation, object bounding boxes, and so on. In addition, we utilize ParallelEye in combination with real datasets to conduct experiments. The experimental results show the inclusion of virtual data helps to enhance the per-class accuracy in object detection and semantic segmentation. Meanwhile, it is also illustrated that the virtual data with controllable imaging conditions can be used to design evaluation experiments flexibly.",
"Modern computer vision algorithms typically require expensive data acquisition and accurate manual labeling. In this work, we instead leverage the recent progress in computer graphics to generate fully labeled, dynamic, and photo-realistic proxy virtual worlds. We propose an efficient real-to-virtual world cloning method, and validate our approach by building and publicly releasing a new video dataset, called Virtual KITTI (see this http URL), automatically labeled with accurate ground truth for object detection, tracking, scene and instance segmentation, depth, and optical flow. We provide quantitative experimental evidence suggesting that (i) modern deep learning algorithms pre-trained on real data behave similarly in real and virtual worlds, and (ii) pre-training on virtual data improves performance. As the gap between real and virtual worlds is small, virtual worlds enable measuring the impact of various weather and imaging conditions on recognition performance, all other things being equal. We show these factors may affect drastically otherwise high-performing deep models for tracking.",
"Drastic variations in illumination across surveillance cameras make the person re-identification problem extremely challenging. Current large scale re-identification datasets have a significant number of training subjects, but lack diversity in lighting conditions. As a result, a trained model requires fine-tuning to become effective under an unseen illumination condition. To alleviate this problem, we introduce a new synthetic dataset that contains hundreds of illumination conditions. Specifically, we use 100 virtual humans illuminated with multiple HDR environment maps which accurately model realistic indoor and outdoor lighting. To achieve better accuracy in unseen illumination conditions we propose a novel domain adaptation technique that takes advantage of our synthetic data and performs fine-tuning in a completely unsupervised way. Our approach yields significantly higher accuracy than semi-supervised and unsupervised state-of-the-art methods, and is very competitive with supervised techniques.",
"Visual Domain Adaptation is a problem of immense importance in computer vision. Previous approaches showcase the inability of even deep neural networks to learn informative representations across domain shift. This problem is more severe for tasks where acquiring hand labeled data is extremely hard and tedious. In this work, we focus on adapting the representations learned by segmentation networks across synthetic and real domains. Contrary to previous approaches that use a simple adversarial objective or superpixel information to aid the process, we propose an approach based on Generative Adversarial Networks (GANs) that brings the embeddings closer in the learned feature space. To showcase the generality and scalability of our approach, we show that we can achieve state of the art results on two challenging scenarios of synthetic to real domain adaptation. Additional exploratory experiments show that our approach: (1) generalizes to unseen domains and (2) results in improved alignment of source and target distributions."
]
} |
1812.02154 | 2902435184 | There has been an unprecedented surge in the number of service providers offering a wide range of machine learning prediction APIs for tasks such as image classification, language translation, etc. thereby monetizing the underlying data and trained models. Typically, a data owner (API provider) develops a model, often over proprietary data, and leverages the infrastructure services of a cloud vendor for hosting and serving API requests. Clearly, this model assumes complete trust between the API Provider and cloud vendor. On the other hand, a malicious buggy cloud vendor may copy the APIs and offer an identical service, under-report model usage metrics, or unfairly discriminate between different API providers by offering them a nominal share of the revenue. In this work, we present the design of a blockchain based decentralized trustless API marketplace that enables all the stakeholders in the API ecosystem to audit the behavior of the parties without having to trust a single centralized entity. In particular, our system divides an AI model into multiple pieces and deploys them among multiple cloud vendors who then collaboratively execute the APIs. Our design ensures that cloud vendors cannot collude with each other to steal the combined model, while individual cloud vendors and clients cannot repudiate their input or model executions. | Kaggle is an example of centralized machine learning marketplace that connects data owners with machine learning developers and rewards developers that submit a model with the highest accuracy. The Danku protocol @cite_23 by Algorithmia offers a similar decentralized service, which uses blockchain. Using a smart contract, the data owners publish their data and an evaluation function on the blockchain. The developers utilize this data and submit trained models that are automatically tested using the evaluation function by miners on the blockchain network to announce a winner. | {
"cite_N": [
"@cite_23"
],
"mid": [
"2788246026"
],
"abstract": [
"Using blockchain technology, it is possible to create contracts that offer a reward in exchange for a trained machine learning model for a particular data set. This would allow users to train machine learning models for a reward in a trustless manner. The smart contract will use the blockchain to automatically validate the solution, so there would be no debate about whether the solution was correct or not. Users who submit the solutions won't have counterparty risk that they won't get paid for their work. Contracts can be created easily by anyone with a dataset, even programmatically by software agents. This creates a market where parties who are good at solving machine learning problems can directly monetize their skillset, and where any organization or software agent that has a problem to solve with AI can solicit solutions from all over the world. This will incentivize the creation of better machine learning models, and make AI more accessible to companies and software agents."
]
} |
1812.02132 | 2902833081 | One major factor impeding more widespread adoption of deep neural networks (DNNs) is their issues with robustness, which is essential for safety critical applications such as autonomous driving. This has motivated much recent work on adversarial attacks for DNNs, which mostly focus on pixel-level perturbations void of semantic meaning. In contrast, we present a general framework for adversarial black-box attacks on trained agents, which covers semantic perturbations to the environment of the agent performing the task as well as pixel-level attacks. To do this, we re-frame the adversarial attack problem as learning a distribution of parameters that always fool the agent. In the semantic case, our proposed adversary (denoted as BBGAN) is trained to sample parameters that describe the environment with which the black-box agent interacts, such that the agent performs its dedicated task poorly in this environment. We apply BBGAN on three different tasks (primarily targeting aspects of autonomous navigation): object detection, self-driving, and autonomous UAV racing. On these tasks, BBGAN can generate failure cases that consistently fool an agent. We also demonstrate the usefulness of our framework as an analysis tool by visualizing systemic failure cases and uncovering semantic insights about the agents themselves. | Attacks on Classifiers by Pixel Perturbation The work of Szegedy @cite_37 was the first to introduce a formulation of attacking neural networks as an optimization. The method minimizes the perturbation of the image pixels, while still fooling a trained classifier to predict a wrong class label. Several works followed the same approach but with different formulations, such as Fast Gradient Sign Method (FGSM) @cite_33 and Projected Gradient Descent @cite_53 @cite_42 . The notable work of Carlini and Wagner @cite_32 performs a comprehensive study of different losses and ways to fool the network with minimal pixel perturbation. As an alternative to pixel attacks, recent work by Hosseini and Poovendran @cite_46 investigates color shifting. Several works have shown how these attacks can be transferred to the real-world by printing perturbed images on paper @cite_1 or building 3D objects based on fooling images @cite_20 . However, all these methods are limited to pixel perturbations and only fool classifiers, while we consider more general cases of attacks, changes in camera viewpoint to fool a detector or change in weather conditions to fool a self-driving car. | {
"cite_N": [
"@cite_37",
"@cite_33",
"@cite_53",
"@cite_42",
"@cite_1",
"@cite_32",
"@cite_46",
"@cite_20"
],
"mid": [
"1673923490",
"2963207607",
"2963389226",
"2243397390",
"2460937040",
"2963857521",
"2795995837",
"2736899637"
],
"abstract": [
"Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.",
"Abstract: Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.",
"Adversarial examples are malicious inputs designed to fool machine learning models. They often transfer from one model to another, allowing attackers to mount black box attacks without knowledge of the target model's parameters. Adversarial training is the process of explicitly training a model on adversarial examples, in order to make it more robust to attack or to reduce its test error on clean inputs. So far, adversarial training has primarily been applied to small problems. In this research, we apply adversarial training to ImageNet. Our contributions include: (1) recommendations for how to succesfully scale adversarial training to large models and datasets, (2) the observation that adversarial training confers robustness to single-step attack methods, (3) the finding that multi-step attack methods are somewhat less transferable than single-step attack methods, so single-step attacks are the best for mounting black-box attacks, and (4) resolution of a label leaking'' effect that causes adversarially trained models to perform better on adversarial examples than on clean examples, because the adversarial example construction process uses the true label and the model can learn to exploit regularities in the construction process.",
"State-of-the-art deep neural networks have achieved impressive results on many image classification tasks. However, these same architectures have been shown to be unstable to small, well sought, perturbations of the images. Despite the importance of this phenomenon, no effective methods have been proposed to accurately compute the robustness of state-of-the-art deep classifiers to such perturbations on large-scale datasets. In this paper, we fill this gap and propose the DeepFool algorithm to efficiently compute perturbations that fool deep networks, and thus reliably quantify the robustness of these classifiers. Extensive experimental results show that our approach outperforms recent methods in the task of computing adversarial perturbations and making classifiers more robust.1",
"Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model. Up to now, all previous work have assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as an input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera.",
"Neural networks provide state-of-the-art results for most machine learning tasks. Unfortunately, neural networks are vulnerable to adversarial examples: given an input x and any target classification t, it is possible to find a new input x' that is similar to x but classified as t. This makes it difficult to apply neural networks in security-critical areas. Defensive distillation is a recently proposed approach that can take an arbitrary neural network, and increase its robustness, reducing the success rate of current attacks' ability to find adversarial examples from 95 to 0.5 .In this paper, we demonstrate that defensive distillation does not significantly increase the robustness of neural networks by introducing three new attack algorithms that are successful on both distilled and undistilled neural networks with 100 probability. Our attacks are tailored to three distance metrics used previously in the literature, and when compared to previous adversarial example generation algorithms, our attacks are often much more effective (and never worse). Furthermore, we propose using high-confidence adversarial examples in a simple transferability test we show can also be used to break defensive distillation. We hope our attacks will be used as a benchmark in future defense attempts to create neural networks that resist adversarial examples.",
"Deep neural networks are known to be vulnerable to adversarial examples, i.e., images that are maliciously perturbed to fool the model. Generating adversarial examples has been mostly limited to finding small perturbations that maximize the model prediction error. Such images, however, contain artificial perturbations that make them somewhat distinguishable from natural images. This property is used by several defense methods to counter adversarial examples by applying denoising filters or training the model to be robust to small perturbations. In this paper, we introduce a new class of adversarial examples, namely \"Semantic Adversarial Examples,\" as images that are arbitrarily perturbed to fool the model, but in such a way that the modified image semantically represents the same object as the original image. We formulate the problem of generating such images as a constrained optimization problem and develop an adversarial transformation based on the shape bias property of human cognitive system. In our method, we generate adversarial images by first converting the RGB image into the HSV (Hue, Saturation and Value) color space and then randomly shifting the Hue and Saturation components, while keeping the Value component the same. Our experimental results on CIFAR10 dataset show that the accuracy of VGG16 network on adversarial color-shifted images is 5.7 .",
"Standard methods for generating adversarial examples for neural networks do not consistently fool neural network classifiers in the physical world due to a combination of viewpoint shifts, camera noise, and other natural transformations, limiting their relevance to real-world systems. We demonstrate the existence of robust 3D adversarial objects, and we present the first algorithm for synthesizing examples that are adversarial over a chosen distribution of transformations. We synthesize two-dimensional adversarial images that are robust to noise, distortion, and affine transformation. We apply our algorithm to complex three-dimensional objects, using 3D-printing to manufacture the first physical adversarial objects. Our results demonstrate the existence of 3D adversarial objects in the physical world."
]
} |
1812.02132 | 2902833081 | One major factor impeding more widespread adoption of deep neural networks (DNNs) is their issues with robustness, which is essential for safety critical applications such as autonomous driving. This has motivated much recent work on adversarial attacks for DNNs, which mostly focus on pixel-level perturbations void of semantic meaning. In contrast, we present a general framework for adversarial black-box attacks on trained agents, which covers semantic perturbations to the environment of the agent performing the task as well as pixel-level attacks. To do this, we re-frame the adversarial attack problem as learning a distribution of parameters that always fool the agent. In the semantic case, our proposed adversary (denoted as BBGAN) is trained to sample parameters that describe the environment with which the black-box agent interacts, such that the agent performs its dedicated task poorly in this environment. We apply BBGAN on three different tasks (primarily targeting aspects of autonomous navigation): object detection, self-driving, and autonomous UAV racing. On these tasks, BBGAN can generate failure cases that consistently fool an agent. We also demonstrate the usefulness of our framework as an analysis tool by visualizing systemic failure cases and uncovering semantic insights about the agents themselves. | Most of these attacks are white-box attacks, in which the algorithm has access to network gradients. Dezfooli al @cite_23 introduce a universal attack that works well on networks that are not used in training the adversary. The work of Chen al @cite_50 develops a black box attack by only assuming access to the output of the classifier. Since back-propagation cannot be performed, the gradient is approximates by finite differences instead. Bhagoji al @cite_11 perform a black-box attack with a small number of network queries by using PCA and randomized dimension selection in the gradient estimation step. While we assume black box constraints in our work, we are interested in the distribution of the semantic parameters that fool the agent, more so than individual fooling examples. This limits the usefulness of approximating the gradient to reach a single local solution. | {
"cite_N": [
"@cite_50",
"@cite_23",
"@cite_11"
],
"mid": [
"2746600820",
"2953047670",
""
],
"abstract": [
"Deep neural networks (DNNs) are one of the most prominent technologies of our time, as they achieve state-of-the-art performance in many machine learning tasks, including but not limited to image classification, text mining, and speech processing. However, recent research on DNNs has indicated ever-increasing concern on the robustness to adversarial examples, especially for security-critical tasks such as traffic sign identification for autonomous driving. Studies have unveiled the vulnerability of a well-trained DNN by demonstrating the ability of generating barely noticeable (to both human and machines) adversarial images that lead to misclassification. Furthermore, researchers have shown that these adversarial images are highly transferable by simply training and attacking a substitute model built upon the target model, known as a black-box attack to DNNs. Similar to the setting of training substitute models, in this paper we propose an effective black-box attack that also only has access to the input (images) and the output (confidence scores) of a targeted DNN. However, different from leveraging attack transferability from substitute models, we propose zeroth order optimization (ZOO) based attacks to directly estimate the gradients of the targeted DNN for generating adversarial examples. We use zeroth order stochastic coordinate descent along with dimension reduction, hierarchical attack and importance sampling techniques to efficiently attack black-box models. By exploiting zeroth order optimization, improved attacks to the targeted DNN can be accomplished, sparing the need for training substitute models and avoiding the loss in attack transferability. Experimental results on MNIST, CIFAR10 and ImageNet show that the proposed ZOO attack is as effective as the state-of-the-art white-box attack (e.g., Carlini and Wagner's attack) and significantly outperforms existing black-box attacks via substitute models.",
"Given a state-of-the-art deep neural network classifier, we show the existence of a universal (image-agnostic) and very small perturbation vector that causes natural images to be misclassified with high probability. We propose a systematic algorithm for computing universal perturbations, and show that state-of-the-art deep neural networks are highly vulnerable to such perturbations, albeit being quasi-imperceptible to the human eye. We further empirically analyze these universal perturbations and show, in particular, that they generalize very well across neural networks. The surprising existence of universal perturbations reveals important geometric correlations among the high-dimensional decision boundary of classifiers. It further outlines potential security breaches with the existence of single directions in the input space that adversaries can possibly exploit to break a classifier on most natural images.",
""
]
} |
1812.02132 | 2902833081 | One major factor impeding more widespread adoption of deep neural networks (DNNs) is their issues with robustness, which is essential for safety critical applications such as autonomous driving. This has motivated much recent work on adversarial attacks for DNNs, which mostly focus on pixel-level perturbations void of semantic meaning. In contrast, we present a general framework for adversarial black-box attacks on trained agents, which covers semantic perturbations to the environment of the agent performing the task as well as pixel-level attacks. To do this, we re-frame the adversarial attack problem as learning a distribution of parameters that always fool the agent. In the semantic case, our proposed adversary (denoted as BBGAN) is trained to sample parameters that describe the environment with which the black-box agent interacts, such that the agent performs its dedicated task poorly in this environment. We apply BBGAN on three different tasks (primarily targeting aspects of autonomous navigation): object detection, self-driving, and autonomous UAV racing. On these tasks, BBGAN can generate failure cases that consistently fool an agent. We also demonstrate the usefulness of our framework as an analysis tool by visualizing systemic failure cases and uncovering semantic insights about the agents themselves. | Attacks beyond Classifiers Sharif al @cite_15 demonstrate attacks on other perception networks and attack a face recognition system by painting frames of sunglasses. Calini and Vagner use adversarial attacks to confuse a speech recognition system @cite_17 . Xie al @cite_26 perform adversarial attacks on object detection and segmentation by perturbing image pixels and show the transferability of the perturbation to different networks.Chen al @cite_2 propose expectation over transformation to construct adversarial attacks on faster R-CNN @cite_30 that is robust to perspective and occlusion. However, they only demonstrate how it works on stop-signs. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_2",
"@cite_15",
"@cite_17"
],
"mid": [
"639708223",
"2604505099",
"2797328537",
"2535873859",
"2783113218"
],
"abstract": [
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [1] and Fast R-CNN [2] have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features—using the recently popular terminology of neural networks with ’attention’ mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model [3] , our detection system has a frame rate of 5 fps ( including all steps ) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.",
"It has been well demonstrated that adversarial examples, i.e., natural images with visually imperceptible perturbations added, cause deep networks to fail on image classification. In this paper, we extend adversarial examples to semantic segmentation and object detection which are much more difficult. Our observation is that both segmentation and detection are based on classifying multiple targets on an image (e.g., the target is a pixel or a receptive field in segmentation, and an object proposal in detection). This inspires us to optimize a loss function over a set of targets for generating adversarial perturbations. Based on this, we propose a novel algorithm named Dense Adversary Generation (DAG), which applies to the state-of-the-art networks for segmentation and detection. We find that the adversarial perturbations can be transferred across networks with different training data, based on different architectures, and even for different recognition tasks. In particular, the transfer ability across networks with the same architecture is more significant than in other cases. Besides, we show that summing up heterogeneous perturbations often leads to better transfer performance, which provides an effective method of black-box adversarial attack.",
"Given the ability to directly manipulate image pixels in the digital input space, an adversary can easily generate imperceptible perturbations to fool a Deep Neural Network (DNN) image classifier, as demonstrated in prior work. In this work, we tackle the more challenging problem of crafting physical adversarial perturbations to fool image-based object detectors like Faster R-CNN. Attacking an object detector is more difficult than attacking an image classifier, as it needs to mislead the classification results in multiple bounding boxes with different scales. Extending the digital attack to the physical world adds another layer of difficulty, because it requires the perturbation to be robust enough to survive real-world distortions due to different viewing distances and angles, lighting conditions, and camera limitations. We show that the Expectation over Transformation technique, which was originally proposed to enhance the robustness of adversarial perturbations in image classification, can be successfully adapted to the object detection setting. Our approach can generate adversarially perturbed stop signs that are consistently mis-detected by Faster R-CNN as other objects, posing a potential threat to autonomous vehicles and other safety-critical computer vision systems.",
"Machine learning is enabling a myriad innovations, including new algorithms for cancer diagnosis and self-driving cars. The broad use of machine learning makes it important to understand the extent to which machine-learning algorithms are subject to attack, particularly when used in applications where physical security or safety is at risk. In this paper, we focus on facial biometric systems, which are widely used in surveillance and access control. We define and investigate a novel class of attacks: attacks that are physically realizable and inconspicuous, and allow an attacker to evade recognition or impersonate another individual. We develop a systematic method to automatically generate such attacks, which are realized through printing a pair of eyeglass frames. When worn by the attacker whose image is supplied to a state-of-the-art face-recognition algorithm, the eyeglasses allow her to evade being recognized or to impersonate another individual. Our investigation focuses on white-box face-recognition systems, but we also demonstrate how similar techniques can be used in black-box scenarios, as well as to avoid face detection.",
"We construct targeted audio adversarial examples on automatic speech recognition. Given any audio waveform, we can produce another that is over 99.9 similar, but transcribes as any phrase we choose (at a rate of up to 50 characters per second). We apply our iterative optimization-based attack to Mozilla's implementation DeepSpeech end-to-end, and show it has a 100 success rate. The feasibility of this attack introduce a new domain to study adversarial examples."
]
} |
1812.02019 | 2902016061 | Accurate traffic forecast is a challenging problem due to the large-scale problem size, as well as the complex and dynamic nature of spatio-temporal dependency of traffic flow. Most existing graph-based CNNs attempt to capture the static relations while largely neglecting the dynamics underlying sequential data. In this paper, we present dynamic spatio-temporal graph-based CNNs (DST-GCNNs) by learning expressive features to represent spatio-temporal structures and predict future traffic from historical traffic flow. In particular, DST-GCNN is a two stream network. In the flow prediction stream, we present a novel graph-based spatio-temporal convolutional layer to extract features from a graph representation of traffic flow. Then several such layers are stacked together to predict future traffic over time. Meanwhile, the proximity relations between nodes in the graph are often time variant as the traffic condition changes over time. To capture the graph dynamics, we use the graph prediction stream to predict the dynamic graph structures, and the predicted structures are fed into the flow prediction stream. Experiments on real traffic datasets demonstrate that the proposed model achieves competitive performances compared with the other state-of-the-art methods. | With the development of deep learning, various network architectures have been proposed for traffic prediction. CNN-based methods @cite_26 @cite_15 @cite_16 and RNN-based methods @cite_5 @cite_23 separately model the spatial and temporal dependency of traffic. However, spatial and temporal dependency are not simultaneously considered. To close this gap, hybrid models where temporal models such as LSTM and GRU are combined with spatial models like 1D-CNN @cite_12 @cite_6 and graphs @cite_2 are proposed and achieve impressive performances. Nevertheless, these recurrent models are restricted to process sequential data successively one-after-one, which limits the parallelization of underlying computations. In contrast, the proposed model utilizes convolutions to capture both spatial and temporal data dependencies, which can reach much more efficiency than the compared recurrent models. Similar to our work, @cite_21 also models spatial and temporal dynamics using graph-CNNs. However, their method fails to consider the dynamics of graph structure which is an important information to traffic prediction. Our proposed model includes an extra stream to model the dynamic graph structure, thus the traffic prediction can benefit from the dynamic graph prediction. | {
"cite_N": [
"@cite_26",
"@cite_21",
"@cite_6",
"@cite_23",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_12"
],
"mid": [
"",
"2756203131",
"2790882348",
"",
"",
"2949888546",
"2579495707",
"2798640882",
"2560675361"
],
"abstract": [
"",
"Timely accurate traffic forecast is crucial for urban traffic control and guidance. Due to the high nonlinearity and complexity of traffic flow, traditional methods cannot satisfy the requirements of mid-and-long term prediction tasks and often neglect spatial and temporal dependencies. In this paper, we propose a novel deep learning framework, Spatio-Temporal Graph Convolutional Networks (STGCN), to tackle the time series prediction problem in traffic domain. Instead of applying regular convolutional and recurrent units, we formulate the problem on graphs and build the model with complete convolutional structures, which enable much faster training speed with fewer parameters. Experiments show that our model STGCN effectively captures comprehensive spatio-temporal correlations through modeling multi-scale traffic networks and consistently outperforms state-of-the-art baselines on various real-world traffic datasets.",
"Traffic flow forecasting has been regarded as a key problem of intelligent transport systems. In this work, we propose a hybrid multimodal deep learning method for short-term traffic flow forecasting, which can jointly and adaptively learn the spatial-temporal correlation features and long temporal interdependence of multi-modality traffic data by an attention auxiliary multimodal deep learning architecture. According to the highly nonlinear characteristics of multi-modality traffic data, the base module of our method consists of one-dimensional Convolutional Neural Networks (1D CNN) and Gated Recurrent Units (GRU) with the attention mechanism. The former is to capture the local trend features and the latter is to capture the long temporal dependencies. Then, we design a hybrid multimodal deep learning framework (HMDLF) for fusing share representation features of different modality traffic data by multiple CNN-GRU-Attention modules. The experimental results indicate that the proposed multimodal deep learning model is capable of dealing with complex nonlinear urban traffic flow forecasting with satisfying accuracy and effectiveness.",
"",
"",
"Deep Neural Networks (DNNs) are powerful models that have achieved excellent performance on difficult learning tasks. Although DNNs work well whenever large labeled training sets are available, they cannot be used to map sequences to sequences. In this paper, we present a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure. Our method uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to a vector of a fixed dimensionality, and then another deep LSTM to decode the target sequence from the vector. Our main result is that on an English to French translation task from the WMT'14 dataset, the translations produced by the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did not have difficulty on long sentences. For comparison, a phrase-based SMT system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM to rerank the 1000 hypotheses produced by the aforementioned SMT system, its BLEU score increases to 36.5, which is close to the previous best result on this task. The LSTM also learned sensible phrase and sentence representations that are sensitive to word order and are relatively invariant to the active and the passive voice. Finally, we found that reversing the order of the words in all source sentences (but not target sentences) improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.",
"This paper proposes a convolutional neural network (CNN)-based method that learns traffic as images and predicts large-scale, network-wide traffic speed with a high accuracy. Spatiotemporal traffic dynamics are converted to images describing the time and space relations of traffic flow via a two-dimensional time-space matrix. A CNN is applied to the image following two consecutive steps: abstract traffic feature extraction and network-wide traffic speed prediction. The effectiveness of the proposed method is evaluated by taking two real-world transportation networks, the second ring road and north-east transportation network in Beijing, as examples, and comparing the method with four prevailing algorithms, namely, ordinary least squares, k-nearest neighbors, artificial neural network, and random forest, and three deep learning architectures, namely, stacked autoencoder, recurrent neural network, and long-short-term memory network. The results show that the proposed method outperforms other algorithms by an average accuracy improvement of 42.91 within an acceptable execution time. The CNN can train the model in a reasonable time and, thus, is suitable for large-scale transportation networks.",
"Real-time traffic flow prediction can not only provide travelers with reliable traffic information so that it can save people's time, but also assist the traffic management agency to manage traffic system. It can greatly improve the efficiency of the transportation system. Traditional traffic flow prediction approaches usually need a large amount of data but still give poor performances. With the development of deep learning, researchers begin to pay attention to artificial neural networks (ANNs) such as RNN and LSTM. However, these ANNs are very time-consuming. In our research, we improve the Deep Residual Network and build a dynamic model which previous researchers hardly use. We firstly integrate the input and output of the @math layer to the input of the @math layer and prove that each layer will fit a simpler function so that the error rate will be much smaller. Then, we use the concept of online learning in our model to update pre-trained model during prediction. Our result shows that our model has higher accuracy than some state-of-the-art models. In addition, our dynamic model can perform better in practical applications.",
"Deep learning approaches have reached a celebrity status in artificial intelligence field, its success have mostly relied on Convolutional Networks (CNN) and Recurrent Networks. By exploiting fundamental spatial properties of images and videos, the CNN always achieves dominant performance on visual tasks. And the Recurrent Networks (RNN) especially long short-term memory methods (LSTM) can successfully characterize the temporal correlation, thus exhibits superior capability for time series tasks. Traffic flow data have plentiful characteristics on both time and space domain. However, applications of CNN and LSTM approaches on traffic flow are limited. In this paper, we propose a novel deep architecture combined CNN and LSTM to forecast future traffic flow (CLTFP). An 1-dimension CNN is exploited to capture spatial features of traffic flow, and two LSTMs are utilized to mine the short-term variability and periodicities of traffic flow. Given those meaningful features, the feature-level fusion is performed to achieve short-term forecasting. The proposed CLTFP is compared with other popular forecasting methods on an open datasets. Experimental results indicate that the CLTFP has considerable advantages in traffic flow forecasting. in additional, the proposed CLTFP is analyzed from the view of Granger Causality, and several interesting properties of CLTFP are discovered and discussed ."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.