Home

Awesome

Contrastive Transformation for Self-supervised Correspondence Learning

Ning Wang, Wengang Zhou, and Houqiang Li

To appear in AAAI 2021

Prerequisites

The code is tested in the following environment:

Training on the TrackingNet

Dataset

We use the TrackingNet dataset for model training.

Training command

python train_trackingnet.py 

Testing on DAVIS2017

To test on DAVIS2017 for instance segmentation mask propagation, please run:

python test.py -d /workspace/DAVIS/ -s 560

Important parameters:

Please check the test.py file for other parameters.

Testing on the VIP dataset

To test on VIP, please run the following command with your own VIP path:

python test_mask_vip.py -o results/VIP/category/ --scale_size 560 560 --pre_num 1 -d /DATA/VIP/VIP_Fine/Images/ --val_txt /DATA/VIP/VIP_Fine/lists/val_videos.txt -c weights/checkpoint_latest.pth.tar

and then:

python eval_vip.py -g DATA/VIP/VIP_Fine/Annotations/Category_ids/ -p results/VIP/category/

Citation

If you find this work useful for your research, please consider citing our work:

@inproceedings{Wang_2021_Contrastive,
    title={Contrastive Transformation for Self-supervised Correspondence Learning},
    author={Wang, Ning and Zhou, Wengang and Li, Houqiang},
    booktitle={AAAI},
    year={2021}
}

Acknowledgements