This repo is under construction.

Learning Pixel Trajectories with Multiscale Contrastive Random Walks

This is the repository for Learning Pixel Trajectories with Multiscale Contrastive Random Walks.

[Paper] [Bibtex]

For any inquiries, please contact us at jasonbian.zx@gmail.com

Use Model

Label Propagation

TODO

Estimate Optical Flow

run inference using pretrained model *.ckpt

python inference.py --input_dir ./demo --pretrained_ckpt ./pretrained/opticalflow/sintel-pretrained.ckpt

To check if you run it successfully, check output "./demo/flow_movie.mp4"` and it should be something like:

Train

Docker is under construction.

Citation

@InProceedings{bian2022learning,
  title={Learning Pixel Trajectories with Multiscale Contrastive Random Walks},
  author={Bian, Zhangxing and Jabri, Allan and Efros, Alexei A. and Owens, Andrew},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year={2022}  
}

Acknowledgements

We thank for portions of the source code from some great works such as ContrastiveRandomWalk, PWC-net, ARflow, Uflow, RAFT.

jasonbian97 / flowwalk