/YOLOv4_Ghostly

Primary LanguagePython

YOLOv4

透過修改YOLOv4的網路架構來提升對於小目標的檢測效果

相關 Github
開發日誌
  • 2022-06-28 - 分支並上傳
  • 2023-08-09 - 上傳最後版本
  • 2023-08-10 - 修改README.MD
  • 2023-08-12 - 簡化資料集路徑，添加docker操作教學

安裝要求 Installation Requirements

  1. Ubuntu:
    確保先安裝 Anaconda，然後執行以下命令
git clone https://github.com/Ghostly0328/YOLOv4_Ghostly.git
conda env create -f environments.yml
conda activate yolov4
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html
  1. docker (建議):
    需要先安裝具有GPU的docker，可以參考這個網址來測試
docker pull pytorch/pytorch:1.11.0-cuda11.3-cudnn8-runtime
docker run --name yolov4 -it -v /git_path/:/yolo --workdir /yolo --rm --gpus 0 --shm-size=64g pytorch/pytorch:1.11.0-cuda11.3-cudnn8-runtime
apt update
apt install -y zip htop screen libgl1-mesa-glx
apt-get install gcc libglib2.0-0
pip install -r requirements.txt
  1. 下載權重檔和數據集並放到 ./models/weights/ 和 ./data/dataset/

需要密碼或下載點壞了可以寄email跟我拿

測試結果

總共有三個資料集可以進行測試，分別是FloW、VOC、WIDERFACE

bash models/weights/FloW/test.sh
bash models/weights/VOC/test.sh
bash models/weights/WIDER/test.sh

檢測 Detect

python detect.py -h #使用說明
python detect.py --cfg ./models/cfg/yolov3.cfg --weights ./models/weights/yolov3_320.weights --source ./data/samples/

訓練 Training

python train.py --data coco.yaml --cfg cfg/yolov4-pacsp.cfg --weights '' --name yolov4-pacsp

測試 Testing

python test.py --img 640 --conf 0.001 --data coco.yaml --cfg cfg/yolov4-pacsp.cfg --weights weights/yolov4-pacsp.pt

引用 Citation

@article{bochkovskiy2020yolov4,
  title={{YOLOv4}: Optimal Speed and Accuracy of Object Detection},
  author={Bochkovskiy, Alexey and Wang, Chien-Yao and Liao, Hong-Yuan Mark},
  journal={arXiv preprint arXiv:2004.10934},
  year={2020}
}

@inproceedings{wang2020cspnet,
  title={{CSPNet}: A New Backbone That Can Enhance Learning Capability of {CNN}},
  author={Wang, Chien-Yao and Mark Liao, Hong-Yuan and Wu, Yueh-Hua and Chen, Ping-Yang and Hsieh, Jun-Wei and Yeh, I-Hau},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  pages={390--391},
  year={2020}
}

Acknowledgements