Object Discovery via Contrastive Learning for Weakly Supervised Object Detection

Jinhwan Seo, Wonho Bae, Danica J. Sutherland, Junhyug Noh, and Daijin Kim

[Paper], [Project page]

The official implementation of ECCV2022 paper: "Object Discovery via Contrastive Learning for Weakly Supervised Object Detection"

Environment setup:

• Python 3.7
• CUDA 11.0
• PyTorch 1.7.1

git clone https://github.com/jinhseo/OD-WSCL/
cd OD-WSCL

conda create --name OD-WSCL python=3.7
conda activate OD-WSCL

pip install ninja yacs cython matplotlib tqdm opencv-python tensorboardX pycocotools
conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=11.0 -c pytorch

git clone --branch 22.04-dev https://github.com/NVIDIA/apex.git
cd apex
pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
cd ../
python setup.py build develop

Dataset:

• PASCAL VOC (2007, 2012)
• MS-COCO (2014, 2017)

mkdir -p datasets/{coco/voc}
    datasets/
    ├── voc/
    │   ├── VOC2007
    │   │   ├── Annotations/
    │   │   ├── JPEGImages/
    │   │   ├── ...
    │   ├── VOC2012/
    │   │   ├── ...
    ├── coco/
    │   ├── annotations/
    │   ├── train2014/
    │   ├── val2014/
    │   ├── train2017/
    │   ├── ...
    ├── ...

Proposal:

Download .pkl file from Dropbox

mkdir proposal
    proposal/
    ├── SS/
    │   ├── voc
    │   │   ├── SS-voc07_trainval.pkl/
    │   │   ├── SS-voc07_test.pkl/
    │   │   ├── ...
    ├── MCG/
    │   ├── voc
    │   │   ├── ...
    │   ├── coco
    │   │   ├── MCG-coco_2014_train_boxes.pkl/
    │   │   ├── ...
    ├── ...

Train:

python -m torch.distributed.launch --nproc_per_node={NO_GPU} tools/train_net.py  
                                   --config-file "configs/{config_file}.yaml"
                                   OUTPUT_DIR {output_dir}
                                   nms {nms threshold}
                                   lmda {lambda value}
                                   iou {iou threshold}
                                   temp {temperature}

Example:

python -m torch.distributed.launch --nproc_per_node=1 tools/train_net.py 
                                   --config-file "configs/voc07_contra_db_b8_lr0.01_mcg.yaml" 
                                   OUTPUT_DIR OD-WSCL/output 
                                   nms 0.1 
                                   lmda 0.03 
                                   iou 0.5
                                   temp 0.2

Note: We trained our model on a single large-memory GPU (e.g., A100 40GB) to maintain large mini-batch size for the best performance.
The hyperparameter settings may vary with multiple small GPUs, and results will be provided later.

Eval:

python -m torch.distributed.launch --nproc_per_node={NO_GPU} tools/test_net.py
                                   --config-file "configs/{config_file}.yaml" 
                                   TEST.IMS_PER_BATCH 8 
                                   OUTPUT_DIR {output_dir} 
                                   MODEL.WEIGHT {model_weight}.pth

Example:

python -m torch.distributed.launch --nproc_per_node=1 tools/test_net.py 
                                   --config-file "configs/voc07_contra_db_b8_lr0.01_mcg.yaml" 
                                   TEST.IMS_PER_BATCH 8 
                                   OUTPUT_DIR OD-WSCL/output 
                                   MODEL.WEIGHT OD-WSCL/output/model_final.pth

Citation:

If you find helpful our work in your research, please consider cite this:

@inproceedings{seo2022object,
  title={Object discovery via contrastive learning for weakly supervised object detection},
  author={Seo, Jinhwan and Bae, Wonho and Sutherland, Danica J and Noh, Junhyug and Kim, Daijin},
  booktitle={European Conference on Computer Vision},
  pages={312--329},
  year={2022},
  organization={Springer}
}

We borrowed the main code from wetectron, please consider cite it as well.
Thank you for sharing your great work!

@inproceedings{ren-cvpr020,
  title = {Instance-aware, Context-focused, and Memory-efficient Weakly Supervised Object Detection},
  author = {Zhongzheng Ren and Zhiding Yu and Xiaodong Yang and Ming-Yu Liu and Yong Jae Lee and Alexander G. Schwing and Jan Kautz},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2020}
}

Acknowledgement:

This work was supported by Institute of Information & communications Technology Planning & Evaluation(IITP) grant funded by the Korea government(MSIT) (No.2017-0-00897, Development of Object Detection and Recognition for Intelligent Vehicles)