Spaces:
Runtime error
Runtime error
A newer version of the Gradio SDK is available:
5.44.1
metadata
title: YOLOV3 GradCAM
emoji: 🐢
colorFrom: pink
colorTo: green
sdk: gradio
sdk_version: 3.40.1
app_file: app.py
pinned: false
license: mit
Gradio Object Detection App with GradCAM for YOLOv3 - ERAv1 Session 13
Table of Contents
- Introduction
- Features
- Model Performance
- Inference Samples
- How to Use
- Supported Classes
- Link to the Model
- Acknowledgements
Introduction
This Gradio app showcases an object detection model using YOLOv3 architecture. The model is trained with enhanced features like multi-resolution training and Mosaic Augmentation. Additionally, the app provides GradCAM outputs for better visualization of the model's predictions.
Features
- PytorchLightning Implementation: The codebase has been refactored to use PytorchLightning for a more modular and scalable approach.
- Multi-resolution Training: Unlike traditional models that train on a fixed resolution, this model has been trained on multiple resolutions (416, 608, 896, 1280) for better generalization.
- Mosaic Augmentation: Implemented Mosaic Augmentation to enhance the training dataset, but only applied 75% of the time to maintain variety.
- Precision Training: The model is trained using float16 precision for faster convergence and reduced memory usage.
- GradCAM Visualization: Integrated GradCAM to provide a heatmap visualization of the regions in the image that the model focuses on during prediction.
Model Performance
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Validate metric ┃ DataLoader 0 ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ val_class_accuracy_epoch │ 81.89761352539062 │
│ val_loss │ 6.100630283355713 │
│ val_no_obj_accuracy_epoch │ 97.92534637451172 │
│ val_obj_accuracy_epoch │ 71.2684097290039 │
└───────────────────────────┴───────────────────────────┘
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 619/619 [29:42<00:00, 2.88s/it]
MAP: 0.10860311985015869
Inference Samples
How to Use
- Navigate to the Gradio app interface.
- Upload a custom image or select from the provided samples.
- Click on the "Predict" button.
- View the object detection predictions along with the GradCAM heatmap.
Supported Classes
Model Architecture
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 32, 416, 416] 864
BatchNorm2d-2 [-1, 32, 416, 416] 64
LeakyReLU-3 [-1, 32, 416, 416] 0
CNNBlock-4 [-1, 32, 416, 416] 0
Conv2d-5 [-1, 64, 208, 208] 18,432
BatchNorm2d-6 [-1, 64, 208, 208] 128
LeakyReLU-7 [-1, 64, 208, 208] 0
CNNBlock-8 [-1, 64, 208, 208] 0
Conv2d-9 [-1, 32, 208, 208] 2,048
BatchNorm2d-10 [-1, 32, 208, 208] 64
LeakyReLU-11 [-1, 32, 208, 208] 0
CNNBlock-12 [-1, 32, 208, 208] 0
Conv2d-13 [-1, 64, 208, 208] 18,432
BatchNorm2d-14 [-1, 64, 208, 208] 128
LeakyReLU-15 [-1, 64, 208, 208] 0
CNNBlock-16 [-1, 64, 208, 208] 0
ResidualBlock-17 [-1, 64, 208, 208] 0
Conv2d-18 [-1, 128, 104, 104] 73,728
BatchNorm2d-19 [-1, 128, 104, 104] 256
LeakyReLU-20 [-1, 128, 104, 104] 0
CNNBlock-21 [-1, 128, 104, 104] 0
Conv2d-22 [-1, 64, 104, 104] 8,192
BatchNorm2d-23 [-1, 64, 104, 104] 128
LeakyReLU-24 [-1, 64, 104, 104] 0
CNNBlock-25 [-1, 64, 104, 104] 0
Conv2d-26 [-1, 128, 104, 104] 73,728
BatchNorm2d-27 [-1, 128, 104, 104] 256
LeakyReLU-28 [-1, 128, 104, 104] 0
CNNBlock-29 [-1, 128, 104, 104] 0
Conv2d-30 [-1, 64, 104, 104] 8,192
BatchNorm2d-31 [-1, 64, 104, 104] 128
LeakyReLU-32 [-1, 64, 104, 104] 0
CNNBlock-33 [-1, 64, 104, 104] 0
Conv2d-34 [-1, 128, 104, 104] 73,728
BatchNorm2d-35 [-1, 128, 104, 104] 256
LeakyReLU-36 [-1, 128, 104, 104] 0
CNNBlock-37 [-1, 128, 104, 104] 0
ResidualBlock-38 [-1, 128, 104, 104] 0
Conv2d-39 [-1, 256, 52, 52] 294,912
BatchNorm2d-40 [-1, 256, 52, 52] 512
LeakyReLU-41 [-1, 256, 52, 52] 0
CNNBlock-42 [-1, 256, 52, 52] 0
Conv2d-43 [-1, 128, 52, 52] 32,768
BatchNorm2d-44 [-1, 128, 52, 52] 256
LeakyReLU-45 [-1, 128, 52, 52] 0
CNNBlock-46 [-1, 128, 52, 52] 0
Conv2d-47 [-1, 256, 52, 52] 294,912
BatchNorm2d-48 [-1, 256, 52, 52] 512
LeakyReLU-49 [-1, 256, 52, 52] 0
CNNBlock-50 [-1, 256, 52, 52] 0
Conv2d-51 [-1, 128, 52, 52] 32,768
BatchNorm2d-52 [-1, 128, 52, 52] 256
LeakyReLU-53 [-1, 128, 52, 52] 0
CNNBlock-54 [-1, 128, 52, 52] 0
Conv2d-55 [-1, 256, 52, 52] 294,912
BatchNorm2d-56 [-1, 256, 52, 52] 512
LeakyReLU-57 [-1, 256, 52, 52] 0
CNNBlock-58 [-1, 256, 52, 52] 0
Conv2d-59 [-1, 128, 52, 52] 32,768
BatchNorm2d-60 [-1, 128, 52, 52] 256
LeakyReLU-61 [-1, 128, 52, 52] 0
CNNBlock-62 [-1, 128, 52, 52] 0
Conv2d-63 [-1, 256, 52, 52] 294,912
BatchNorm2d-64 [-1, 256, 52, 52] 512
LeakyReLU-65 [-1, 256, 52, 52] 0
CNNBlock-66 [-1, 256, 52, 52] 0
Conv2d-67 [-1, 128, 52, 52] 32,768
BatchNorm2d-68 [-1, 128, 52, 52] 256
LeakyReLU-69 [-1, 128, 52, 52] 0
CNNBlock-70 [-1, 128, 52, 52] 0
Conv2d-71 [-1, 256, 52, 52] 294,912
BatchNorm2d-72 [-1, 256, 52, 52] 512
LeakyReLU-73 [-1, 256, 52, 52] 0
CNNBlock-74 [-1, 256, 52, 52] 0
Conv2d-75 [-1, 128, 52, 52] 32,768
BatchNorm2d-76 [-1, 128, 52, 52] 256
LeakyReLU-77 [-1, 128, 52, 52] 0
CNNBlock-78 [-1, 128, 52, 52] 0
Conv2d-79 [-1, 256, 52, 52] 294,912
BatchNorm2d-80 [-1, 256, 52, 52] 512
LeakyReLU-81 [-1, 256, 52, 52] 0
CNNBlock-82 [-1, 256, 52, 52] 0
Conv2d-83 [-1, 128, 52, 52] 32,768
BatchNorm2d-84 [-1, 128, 52, 52] 256
LeakyReLU-85 [-1, 128, 52, 52] 0
CNNBlock-86 [-1, 128, 52, 52] 0
Conv2d-87 [-1, 256, 52, 52] 294,912
BatchNorm2d-88 [-1, 256, 52, 52] 512
LeakyReLU-89 [-1, 256, 52, 52] 0
CNNBlock-90 [-1, 256, 52, 52] 0
Conv2d-91 [-1, 128, 52, 52] 32,768
BatchNorm2d-92 [-1, 128, 52, 52] 256
LeakyReLU-93 [-1, 128, 52, 52] 0
CNNBlock-94 [-1, 128, 52, 52] 0
Conv2d-95 [-1, 256, 52, 52] 294,912
BatchNorm2d-96 [-1, 256, 52, 52] 512
LeakyReLU-97 [-1, 256, 52, 52] 0
CNNBlock-98 [-1, 256, 52, 52] 0
Conv2d-99 [-1, 128, 52, 52] 32,768
BatchNorm2d-100 [-1, 128, 52, 52] 256
LeakyReLU-101 [-1, 128, 52, 52] 0
CNNBlock-102 [-1, 128, 52, 52] 0
Conv2d-103 [-1, 256, 52, 52] 294,912
BatchNorm2d-104 [-1, 256, 52, 52] 512
LeakyReLU-105 [-1, 256, 52, 52] 0
CNNBlock-106 [-1, 256, 52, 52] 0
ResidualBlock-107 [-1, 256, 52, 52] 0
Conv2d-108 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-109 [-1, 512, 26, 26] 1,024
LeakyReLU-110 [-1, 512, 26, 26] 0
CNNBlock-111 [-1, 512, 26, 26] 0
Conv2d-112 [-1, 256, 26, 26] 131,072
BatchNorm2d-113 [-1, 256, 26, 26] 512
LeakyReLU-114 [-1, 256, 26, 26] 0
CNNBlock-115 [-1, 256, 26, 26] 0
Conv2d-116 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-117 [-1, 512, 26, 26] 1,024
LeakyReLU-118 [-1, 512, 26, 26] 0
CNNBlock-119 [-1, 512, 26, 26] 0
Conv2d-120 [-1, 256, 26, 26] 131,072
BatchNorm2d-121 [-1, 256, 26, 26] 512
LeakyReLU-122 [-1, 256, 26, 26] 0
CNNBlock-123 [-1, 256, 26, 26] 0
Conv2d-124 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-125 [-1, 512, 26, 26] 1,024
LeakyReLU-126 [-1, 512, 26, 26] 0
CNNBlock-127 [-1, 512, 26, 26] 0
Conv2d-128 [-1, 256, 26, 26] 131,072
BatchNorm2d-129 [-1, 256, 26, 26] 512
LeakyReLU-130 [-1, 256, 26, 26] 0
CNNBlock-131 [-1, 256, 26, 26] 0
Conv2d-132 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-133 [-1, 512, 26, 26] 1,024
LeakyReLU-134 [-1, 512, 26, 26] 0
CNNBlock-135 [-1, 512, 26, 26] 0
Conv2d-136 [-1, 256, 26, 26] 131,072
BatchNorm2d-137 [-1, 256, 26, 26] 512
LeakyReLU-138 [-1, 256, 26, 26] 0
CNNBlock-139 [-1, 256, 26, 26] 0
Conv2d-140 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-141 [-1, 512, 26, 26] 1,024
LeakyReLU-142 [-1, 512, 26, 26] 0
CNNBlock-143 [-1, 512, 26, 26] 0
Conv2d-144 [-1, 256, 26, 26] 131,072
BatchNorm2d-145 [-1, 256, 26, 26] 512
LeakyReLU-146 [-1, 256, 26, 26] 0
CNNBlock-147 [-1, 256, 26, 26] 0
Conv2d-148 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-149 [-1, 512, 26, 26] 1,024
LeakyReLU-150 [-1, 512, 26, 26] 0
CNNBlock-151 [-1, 512, 26, 26] 0
Conv2d-152 [-1, 256, 26, 26] 131,072
BatchNorm2d-153 [-1, 256, 26, 26] 512
LeakyReLU-154 [-1, 256, 26, 26] 0
CNNBlock-155 [-1, 256, 26, 26] 0
Conv2d-156 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-157 [-1, 512, 26, 26] 1,024
LeakyReLU-158 [-1, 512, 26, 26] 0
CNNBlock-159 [-1, 512, 26, 26] 0
Conv2d-160 [-1, 256, 26, 26] 131,072
BatchNorm2d-161 [-1, 256, 26, 26] 512
LeakyReLU-162 [-1, 256, 26, 26] 0
CNNBlock-163 [-1, 256, 26, 26] 0
Conv2d-164 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-165 [-1, 512, 26, 26] 1,024
LeakyReLU-166 [-1, 512, 26, 26] 0
CNNBlock-167 [-1, 512, 26, 26] 0
Conv2d-168 [-1, 256, 26, 26] 131,072
BatchNorm2d-169 [-1, 256, 26, 26] 512
LeakyReLU-170 [-1, 256, 26, 26] 0
CNNBlock-171 [-1, 256, 26, 26] 0
Conv2d-172 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-173 [-1, 512, 26, 26] 1,024
LeakyReLU-174 [-1, 512, 26, 26] 0
CNNBlock-175 [-1, 512, 26, 26] 0
ResidualBlock-176 [-1, 512, 26, 26] 0
Conv2d-177 [-1, 1024, 13, 13] 4,718,592
BatchNorm2d-178 [-1, 1024, 13, 13] 2,048
LeakyReLU-179 [-1, 1024, 13, 13] 0
CNNBlock-180 [-1, 1024, 13, 13] 0
Conv2d-181 [-1, 512, 13, 13] 524,288
BatchNorm2d-182 [-1, 512, 13, 13] 1,024
LeakyReLU-183 [-1, 512, 13, 13] 0
CNNBlock-184 [-1, 512, 13, 13] 0
Conv2d-185 [-1, 1024, 13, 13] 4,718,592
BatchNorm2d-186 [-1, 1024, 13, 13] 2,048
LeakyReLU-187 [-1, 1024, 13, 13] 0
CNNBlock-188 [-1, 1024, 13, 13] 0
Conv2d-189 [-1, 512, 13, 13] 524,288
BatchNorm2d-190 [-1, 512, 13, 13] 1,024
LeakyReLU-191 [-1, 512, 13, 13] 0
CNNBlock-192 [-1, 512, 13, 13] 0
Conv2d-193 [-1, 1024, 13, 13] 4,718,592
BatchNorm2d-194 [-1, 1024, 13, 13] 2,048
LeakyReLU-195 [-1, 1024, 13, 13] 0
CNNBlock-196 [-1, 1024, 13, 13] 0
Conv2d-197 [-1, 512, 13, 13] 524,288
BatchNorm2d-198 [-1, 512, 13, 13] 1,024
LeakyReLU-199 [-1, 512, 13, 13] 0
CNNBlock-200 [-1, 512, 13, 13] 0
Conv2d-201 [-1, 1024, 13, 13] 4,718,592
BatchNorm2d-202 [-1, 1024, 13, 13] 2,048
LeakyReLU-203 [-1, 1024, 13, 13] 0
CNNBlock-204 [-1, 1024, 13, 13] 0
Conv2d-205 [-1, 512, 13, 13] 524,288
BatchNorm2d-206 [-1, 512, 13, 13] 1,024
LeakyReLU-207 [-1, 512, 13, 13] 0
CNNBlock-208 [-1, 512, 13, 13] 0
Conv2d-209 [-1, 1024, 13, 13] 4,718,592
BatchNorm2d-210 [-1, 1024, 13, 13] 2,048
LeakyReLU-211 [-1, 1024, 13, 13] 0
CNNBlock-212 [-1, 1024, 13, 13] 0
ResidualBlock-213 [-1, 1024, 13, 13] 0
Conv2d-214 [-1, 1024, 13, 13] 1,048,576
BatchNorm2d-215 [-1, 1024, 13, 13] 2,048
LeakyReLU-216 [-1, 1024, 13, 13] 0
CNNBlock-217 [-1, 1024, 13, 13] 0
Conv2d-218 [-1, 2048, 13, 13] 18,874,368
BatchNorm2d-219 [-1, 2048, 13, 13] 4,096
LeakyReLU-220 [-1, 2048, 13, 13] 0
CNNBlock-221 [-1, 2048, 13, 13] 0
Conv2d-222 [-1, 1024, 13, 13] 2,097,152
BatchNorm2d-223 [-1, 1024, 13, 13] 2,048
LeakyReLU-224 [-1, 1024, 13, 13] 0
CNNBlock-225 [-1, 1024, 13, 13] 0
Conv2d-226 [-1, 2048, 13, 13] 18,874,368
BatchNorm2d-227 [-1, 2048, 13, 13] 4,096
LeakyReLU-228 [-1, 2048, 13, 13] 0
CNNBlock-229 [-1, 2048, 13, 13] 0
ResidualBlock-230 [-1, 2048, 13, 13] 0
Conv2d-231 [-1, 1024, 13, 13] 2,097,152
BatchNorm2d-232 [-1, 1024, 13, 13] 2,048
LeakyReLU-233 [-1, 1024, 13, 13] 0
CNNBlock-234 [-1, 1024, 13, 13] 0
Conv2d-235 [-1, 2048, 13, 13] 18,874,368
BatchNorm2d-236 [-1, 2048, 13, 13] 4,096
LeakyReLU-237 [-1, 2048, 13, 13] 0
CNNBlock-238 [-1, 2048, 13, 13] 0
Conv2d-239 [-1, 75, 13, 13] 153,675
CNNBlock-240 [-1, 75, 13, 13] 0
ScalePrediction-241 [-1, 3, 13, 13, 25] 0
Conv2d-242 [-1, 256, 13, 13] 262,144
BatchNorm2d-243 [-1, 256, 13, 13] 512
LeakyReLU-244 [-1, 256, 13, 13] 0
CNNBlock-245 [-1, 256, 13, 13] 0
Upsample-246 [-1, 256, 26, 26] 0
Conv2d-247 [-1, 256, 26, 26] 196,608
BatchNorm2d-248 [-1, 256, 26, 26] 512
LeakyReLU-249 [-1, 256, 26, 26] 0
CNNBlock-250 [-1, 256, 26, 26] 0
Conv2d-251 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-252 [-1, 512, 26, 26] 1,024
LeakyReLU-253 [-1, 512, 26, 26] 0
CNNBlock-254 [-1, 512, 26, 26] 0
Conv2d-255 [-1, 256, 26, 26] 131,072
BatchNorm2d-256 [-1, 256, 26, 26] 512
LeakyReLU-257 [-1, 256, 26, 26] 0
CNNBlock-258 [-1, 256, 26, 26] 0
Conv2d-259 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-260 [-1, 512, 26, 26] 1,024
LeakyReLU-261 [-1, 512, 26, 26] 0
CNNBlock-262 [-1, 512, 26, 26] 0
ResidualBlock-263 [-1, 512, 26, 26] 0
Conv2d-264 [-1, 256, 26, 26] 131,072
BatchNorm2d-265 [-1, 256, 26, 26] 512
LeakyReLU-266 [-1, 256, 26, 26] 0
CNNBlock-267 [-1, 256, 26, 26] 0
Conv2d-268 [-1, 512, 26, 26] 1,179,648
BatchNorm2d-269 [-1, 512, 26, 26] 1,024
LeakyReLU-270 [-1, 512, 26, 26] 0
CNNBlock-271 [-1, 512, 26, 26] 0
Conv2d-272 [-1, 75, 26, 26] 38,475
CNNBlock-273 [-1, 75, 26, 26] 0
ScalePrediction-274 [-1, 3, 26, 26, 25] 0
Conv2d-275 [-1, 128, 26, 26] 32,768
BatchNorm2d-276 [-1, 128, 26, 26] 256
LeakyReLU-277 [-1, 128, 26, 26] 0
CNNBlock-278 [-1, 128, 26, 26] 0
Upsample-279 [-1, 128, 52, 52] 0
Conv2d-280 [-1, 128, 52, 52] 49,152
BatchNorm2d-281 [-1, 128, 52, 52] 256
LeakyReLU-282 [-1, 128, 52, 52] 0
CNNBlock-283 [-1, 128, 52, 52] 0
Conv2d-284 [-1, 256, 52, 52] 294,912
BatchNorm2d-285 [-1, 256, 52, 52] 512
LeakyReLU-286 [-1, 256, 52, 52] 0
CNNBlock-287 [-1, 256, 52, 52] 0
Conv2d-288 [-1, 128, 52, 52] 32,768
BatchNorm2d-289 [-1, 128, 52, 52] 256
LeakyReLU-290 [-1, 128, 52, 52] 0
CNNBlock-291 [-1, 128, 52, 52] 0
Conv2d-292 [-1, 256, 52, 52] 294,912
BatchNorm2d-293 [-1, 256, 52, 52] 512
LeakyReLU-294 [-1, 256, 52, 52] 0
CNNBlock-295 [-1, 256, 52, 52] 0
ResidualBlock-296 [-1, 256, 52, 52] 0
Conv2d-297 [-1, 128, 52, 52] 32,768
BatchNorm2d-298 [-1, 128, 52, 52] 256
LeakyReLU-299 [-1, 128, 52, 52] 0
CNNBlock-300 [-1, 128, 52, 52] 0
Conv2d-301 [-1, 256, 52, 52] 294,912
BatchNorm2d-302 [-1, 256, 52, 52] 512
LeakyReLU-303 [-1, 256, 52, 52] 0
CNNBlock-304 [-1, 256, 52, 52] 0
Conv2d-305 [-1, 75, 52, 52] 19,275
CNNBlock-306 [-1, 75, 52, 52] 0
ScalePrediction-307 [-1, 3, 52, 52, 25] 0
================================================================
Total params: 107,980,481
Trainable params: 107,980,481
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.98
Forward/backward pass size (MB): 1253.79
Params size (MB): 411.91
Estimated Total Size (MB): 1667.68
----------------------------------------------------------------
Examples
App includes some examples images for testing
Github
Training code may be found here
References
https://arxiv.org/abs/1804.02767
https://www.youtube.com/watch?v=Grir6TZbc1M
https://github.com/jacobgil/pytorch-grad-cam