BMVOS

This is the official implementation of Pixel-Level Bijective Matching for Video Object Segmentation, to appear in WACV 2022.

@article{cho2021pixel,
  title={Pixel-Level Bijective Matching for Video Object Segmentation},
  author={Cho, Suhwan and Lee, Heansung and Kim, Minjung and Jang, Sungjun and Lee, Sangyoun},
  journal={arXiv preprint arXiv:2110.01644},
  year={2021}
}

Benchmark Results

Architecture

Download

[pre-computed results]

[pre-trained model on DAVIS]

[pre-trained model on YouTube-VOS]

Usage

1. Define the paths in 'local_config.py'.

2. Select the pre-trained model and testing dataset by modifying 'main_runfile.py'.

3. Run BMVOS!

python main_runfile.py

Note

Code and models are only available for non-commercial research purposes.

If you have any questions, please feel free to contact me :)

E-mail: [email protected]