MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training

De-An Huang, Zhiding Yu, Anima Anandkumar

[arXiv] [Project] [BibTeX]

Features

Video instance segmentation by only training an image instance segmentation model.
Support major video instance segmentation datasets: YouTubeVIS 2019/2021, Occluded VIS (OVIS).

Qualitative Results on Occluded VIS

Installation

See installation instructions.

Getting Started

See Preparing Datasets for MinVIS.

See Getting Started with MinVIS.

Model Zoo

Trained models are available for download in the MinVIS Model Zoo.

License

The majority of MinVIS is made available under the Nvidia Source Code License-NC. The trained models in the MinVIS Model Zoo are made available under the CC BY-NC-SA 4.0 License.

Portions of the project are available under separate license terms: Mask2Former is licensed under a MIT License. Swin-Transformer-Semantic-Segmentation is licensed under the MIT License, Deformable-DETR is licensed under the Apache-2.0 License.

Citing MinVIS

@article{huang2022minvis,
  title={MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training},
  author={De-An Huang and Zhiding Yu and Anima Anandkumar},
  journal={arXiv preprint arXiv:2208.02245},
  year={2022}
}

Acknowledgement

This repo is largely based on Mask2Former (https://github.com/facebookresearch/Mask2Former).

AkbarTrilaksana / MinVIS