YOLO: Official Implementation of YOLOv9, YOLOv7

We wanted to inform you that the training code for this project is still in progress, and there are two known issues:

CPU memory leak during training

Slower convergence speed

We strongly recommend refraining from training the model until version 1.0 is released. However, inference and validation with pre-trained weights on COCO are available and can be used safely.

Welcome to the official implementation of YOLOv7 and YOLOv9. This repository will contains the complete codebase, pre-trained models, and detailed instructions for training and deploying YOLOv9.

TL;DR

This is the official YOLO model implementation with an MIT License.
For quick deployment: you can directly install by pip+git:

pip install git+https://github.com/WongKinYiu/YOLO.git
yolo task.data.source=0 # source could be a single file, video, image folder, webcam ID

Introduction

Installation

To get started using YOLOv9's developer mode, we recommand you clone this repository and install the required dependencies:

git clone git@github.com:WongKinYiu/YOLO.git
cd YOLO
pip install -r requirements.txt

Features

Task

These are simple examples. For more customization details, please refer to Notebooks and lower-level modifications HOWTO.

Training

To train YOLO on your machine/dataset:

Modify the configuration file yolo/config/dataset/**.yaml to point to your dataset.
Run the training script:

python yolo/lazy.py task=train dataset=** use_wandb=True
python yolo/lazy.py task=train task.data.batch_size=8 model=v9-c weight=False # or more args

Transfer Learning

To perform transfer learning with YOLOv9:

python yolo/lazy.py task=train task.data.batch_size=8 model=v9-c dataset={dataset_config} device={cpu, mps, cuda}

Inference

To use a model for object detection, use:

python yolo/lazy.py # if cloned from GitHub
python yolo/lazy.py task=inference \ # default is inference
                    name=AnyNameYouWant \ # AnyNameYouWant
                    device=cpu \ # hardware cuda, cpu, mps
                    model=v9-s \ # model version: v9-c, m, s
                    task.nms.min_confidence=0.1 \ # nms config
                    task.fast_inference=onnx \ # onnx, trt, deploy
                    task.data.source=data/toy/images/train \ # file, dir, webcam
                    +quite=True \ # Quite Output
yolo task.data.source={Any Source} # if pip installed
yolo task=inference task.data.source={Any}

Validation

To validate model performance, or generate a json file in COCO format:

python yolo/lazy.py task=validation
python yolo/lazy.py task=validation dataset=toy

Contributing

Contributions to the YOLO project are welcome! See CONTRIBUTING for guidelines on how to contribute.

TODO Diagrams

flowchart TB
    subgraph Features
      Taskv7-->Segmentation["#35 Segmentation"]
      Taskv7-->Classification["#34 Classification"]
      Taskv9-->Segmentation
      Taskv9-->Classification
      Trainv7
    end
    subgraph Model
      MODELv7-->v7-X
      MODELv7-->v7-E6
      MODELv7-->v7-E6E
      MODELv9-->v9-T
      MODELv9-->v9-S
      MODELv9-->v9-E
    end
    subgraph Bugs
      Fix-->Fix1["#12 mAP > 1"]
      Fix-->Fix2["v9 Gradient Bump"]
      Reply-->Reply1["#39"]
      Reply-->Reply2["#36"]
    end

Star History

Citations

@misc{wang2022yolov7,
      title={YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors},
      author={Chien-Yao Wang and Alexey Bochkovskiy and Hong-Yuan Mark Liao},
      year={2022},
      eprint={2207.02696},
      archivePrefix={arXiv},
      primaryClass={id='cs.CV' full_name='Computer Vision and Pattern Recognition' is_active=True alt_name=None in_archive='cs' is_general=False description='Covers image processing, computer vision, pattern recognition, and scene understanding. Roughly includes material in ACM Subject Classes I.2.10, I.4, and I.5.'}
}
@misc{wang2024yolov9,
      title={YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information},
      author={Chien-Yao Wang and I-Hau Yeh and Hong-Yuan Mark Liao},
      year={2024},
      eprint={2402.13616},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}