面向边缘智能光学感知的航空紧固件旋转检测

符长虹, 陈锟辉, 鲁昆瀚, 郑光泽, 赵吉林. 面向边缘智能光学感知的航空紧固件旋转检测[J]. 应用光学. 2022, 43(3):472-480.

Install OLAFDet

git clone https://github.com/vision4robotics/OLAFDet.git
cd OLAFDet
pip install -r requirements.txt
cd utils/nms_rotated
python setup.py develop  #or "pip install -v -e ."

cd ../..
cd DOTA_devkit
sudo apt-get install swig
swig -c++ -python polyiou.i
python setup.py build_ext --inplace

Preprocessing

Download weight: weights.zip 提取码：aufh

Download dataset: fastener.zip 提取码：vu8k

Make sure the labels format is [poly classname diffcult], e.g.,

  x1      y1       x2        y2       x3       y3       x4       y4       classname     diffcult

1686.0   1517.0   1695.0   1511.0   1711.0   1535.0   1700.0   1541.0   large-vehicle      1

*(Note: You can set diffcult=0)

If you need to split the dataset.

python DOTA_devkit/ImgSplit_multi_process.py

train

python train.py --weights weights/yolov5s.pt --cfg models/olafdet_dota_v1.yaml  --data data/olafdet_dota_v1.yaml --device 0

python train.py --weights weights/dota_v1.pt --cfg models/olafdet.yaml  --data data/olafdet_fastener.yaml --device 0

detect

python detect.py --weights weights/dota_v1.pt --source dataset/dota

python detect.py --weights weights/fastener.pt --source dataset/fastener

Acknowledgments

We sincerely thank the contribution of following repos: yolov5_obb, yolov5 and CSL_RetinaNet_Tensorflow

Contact

If you have any questions, please contact Kunhui Chen at chenkunhui@tongji.edu.cn