Co-mining: Self-Supervised Learning for Sparsely Annotated Object Detection

This repository is an official implementation of the AAAI 2021 paper Co-mining: Self-Supervised Learning for Sparsely Annotated Object Detection.

Introduction

TL; DR. Co-ming is a self-supervised learning framework for sparsely annotated object detection.

Get Started

Install cvpods following the instructions

# Install cvpods
git clone https://github.com/Megvii-BaseDetection/cvpods.git
cd cvpods 
## build cvpods (requires GPU)
python3 setup.py build develop
## preprare data path
mkdir datasets
ln -s /path/to/your/coco/dataset datasets/coco

Download the sparse-annotations from here of four cases and put them into /coco/annotations/. Note that the annotation of “missing_50p” is from the authors of BRL paper.
For fast evaluation, please download trained model from here.
Run the project

git clone https://github.com/megvii-research/Co-mining.git

# for example(e.g. miss50p)
cd co-mining/retinanet.res101.comining.score.06.miss50p/

# train
pods_train --num-gpus 8

# test
pods_test --num-gpus 8
# test with provided weights
pods_test --num-gpus 8 MODEL.WEIGHTS /path/to/your/model.pth

Results

Model	Multi-scale training	AP (minival)	Link
Comining_RetinaNet_Res50_Full	No	36.8	download
Comining_RetinaNet_Res50_Easy	No	35.4	download
Comining_RetinaNet_Res50_Hard	No	31.8	download
Comining_RetinaNet_Res50_Extreme	No	23.0	download
Comining_RetinaNet_Res101_Miss50p	No	33.9	download

Citing Co-mining

If you find Co-mining useful to your research, please consider citing:

@article{wang2021comining,
  title={Co-mining: Self-Supervised Learning for Sparsely Annotated Object Detection},
  author={Wang, Tiancai and Yang, Tong and Cao, Jiale and Zhang, Xiangyu},
  journal={Proceedings of the AAAI conference on artificial intelligence},
  year={2021}
}