Learning a Unified Sample Weighting Network for Object Detection

This is the implementation of 'Learning a Unified Sample Weighting Network for Object Detection' [CVPR 2020]. The original paper can be found here.

Usage

a. Following MMDetection for installing pre-requirements.

b. clone this repo and build.

git clone https://github.com/caiqi/sample-weighting-network
cd sample-weighting-network
python setup.py develop

c. Prepare COCO dataset.

It is recommended to symlink the dataset root to sample-weighting-network/data.

sample-weighting-network
├── mmdet
├── tools
├── configs
├── data
│   ├── coco
│   │   ├── annotations
│   │   ├── train2017
│   │   ├── val2017
│   │   ├── test2017
│   ├── VOCdevkit
│   │   ├── VOC2007
│   │   ├── VOC2012

d. Train Faster R-CNN/Mask-RCNN/RetinaNet with ResNet-50/ResNet-101 backbones:

    bash scritps/[Faster/Mask/Retina]_[R50/R101].sh

Citation

If you find this code or model useful for your research, please cite our paper:

@inproceedings{cai2020learning,
  title={Learning a Unified Sample Weighting Network for Object Detection},
  author={Cai, Qi and Pan, Yingwei and Wang, Yu and Liu, Jingen and Yao, Ting and Mei, Tao},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}

Acknowledgements

Thanks the contribution of MMDetection and awesome PyTorch team.