Object Detection on UAV images(All the operation should be in code folder)

Group Member

• R11922029 吳泓毅
• R10922026 吳勝濬
• R10922102 林正偉

Dataset Download

Link

Environment Setting

conda create -n yolov7 python=3.9
conda activate yolov7
pip install -r requirements.txt  

Folder Structure

You need to create datasets folder by yourself

|----code
|----datasets----DIP(extract below datasets folder)

Training

python train.py --data DIP.yaml --weights yolov7_training.pt --cfg ./cfg/training/<model>.yaml --img <img size> --epoch <epoch_num>  --freeze <freeze layers> --batch-size <batch_size>

<model>.yaml: choose model strucures which are put in cfg/training folder
<img size>: training size and validation size
<epoch_num>: training epoch num, 100 for img-size 640, 50 for img-size 1280
<freeze layers>: fixed backbone weights before assigned layer nums, 50 for fixed backbone, 52 for fixed backbone and neck, if you don't want fixed any weight, just don't use this argument. <batch_size>: choose batch size depend on GPU

Inference on Testing Dataset

Weights Download

Link

python detect.py --source ../data/ --weights <weights path>  --img <img size> 

<weights folder>: path to weights
<img size>: inference size 640 or 1280 depend on training size

The results on testing images are here

Acknowledgements

Expand

• https://github.com/WongKinYiu/yolov7
• https://github.com/AlexeyAB/darknet
• https://github.com/WongKinYiu/yolor
• https://github.com/WongKinYiu/PyTorch_YOLOv4
• https://github.com/WongKinYiu/ScaledYOLOv4
• https://github.com/Megvii-BaseDetection/YOLOX
• https://github.com/ultralytics/yolov3
• https://github.com/ultralytics/yolov5
• https://github.com/DingXiaoH/RepVGG
• https://github.com/JUGGHM/OREPA_CVPR2022
• https://github.com/TexasInstruments/edgeai-yolov5/tree/yolo-pose