PARDet: Dynamic Point Set Alignment for Rotated Object Detection

Benchmark and model zoo

Model	Backbone	MS	Rotate	Lr schd	Download
PARDet	R-50-FPN	-	-	1x	BaiduNetDisk[5xn6]
PARDet_ATSS	R-50-FPN	-	-	1x	BaiduNetDisk[p4os]

Installation

create virtual environment

conda create -n pardet python=3.8 -y
conda activate pardet

install pytorch

pip3 install torch==1.7.0+cu101 torchvision==0.8.1+cu101 -f https://download.pytorch.org/whl/torch_stable.html

manually install mmdetection/mmcv

pip install mmcv-full==1.2.5 -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.7.0/index.html

cd pardet
pip install -r requirements/build.txt
pip install -v -e . # or "python setup.py develop"

Install DOTA_devkit

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

Train

CUDA_VISIBLE_DEVICES=1 python tools/train.py configs/PARDet/pardet_r50_fpn_1x.py 

Test

CUDA_VISIBLE_DEVICES=1 python tools/test.py {configs} {ckpt_path} --out {pkl_path}

etc.

CUDA_VISIBLE_DEVICES=1 python tools/test.py configs/PARDet/pardet_r50_fpn_1x.py /data/Aerial/checkpoints/paper/pardet/sift_point_num20_con_4/epoch_12.pth --out /data/Aerial/checkpoints/paper/pardet/sift_point_num20_con_4/results.pkl

Parse result

python tools/rotate/parse_results.py {configs} {pkl_path} 
						    -[nms]
                              --[type]
                              --[eval]

• configs:the model config you design
• pkl:model inference result
• nms:whether to merge result [Y/N]
• type:if you want to merge, merge rotate or horizon [HBB/OBB/ALL]
• eval: whether to eval result

etc.

python tools/rotate/parse_results.py configs/PARDet/pardet_r50_fpn_1x.py /data/Aerial/checkpoints/paper/pardet/sift_point_num20_con_4/results.pkl Y --eval=N  

Citation

@article{xu2024pardet,
  title={PARDet: Dynamic point set alignment for rotated object detection},
  author={Xu, Yihao and Shen, Jifeng and Dai, Ming and Yang, Wankou},
  journal={Pattern Recognition},
  pages={110534},
  year={2024},
  publisher={Elsevier}
}