DVIS: Decoupled Video Instance Segmentation Framework

Tao Zhang, XingYe Tian, Yu Wu, ShunPing Ji, Xuebo Wang, Yuan Zhang, Pengfei Wan

News

• DVIS-DAQ achieves 57.1 AP on the OVIS dataset and also sets a new SOTA performance on YTVIS19/21 and VIPSeg. The code will be released in DVIS-DAQ. The paper is available at DVIS-DAQ: Improving Video Segmentation via Dynamic Anchor Queries and the project page can be found in project page.
• The improved version of DVIS, DVIS++, is now available. Please refer to DVIS++ for more information. DVIS++ achieves 41.2 AP, 56.7 AP, and 52.0 AP, as well as 48.6 mIOU and 44.2 VPQ in OVIS, YTVIS19, YTVIS21, VSPW, and VIPSeg, respectively. Additionally, OV-DVIS++ supports open-vocabulary universal video segmentation.
• DVIS achieved 1st place in the VPS Track of the PVUW challenge at CVPR 2023. 2023.5.25
• DVIS has been accepted by ICCV 2023. 2023.7.15
• DVIS achieved 1st place in the VIS Track of the 5th LSVOS challenge at ICCV 2023. 2023.8.15

Features

• DVIS is a universal video segmentation framework that supports VIS, VPS and VSS.
• DVIS can run in both online and offline modes.
• DVIS achieved SOTA performance on YTVIS, OVIS, VIPSeg and VSPW datasets.
• DVIS can complete training and inference on GPUs with only 11G memory.

Demos

Installation

See Installation Instructions.

Getting Started

See Preparing Datasets for DVIS.

See Getting Started with DVIS.

Model Zoo

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

Citing DVIS

@article{DVIS,
  title={DVIS: Decoupled Video Instance Segmentation Framework},
  author={Zhang, Tao and Tian, Xingye and Wu, Yu and Ji, Shunping and Wang, Xuebo and Zhang, Yuan and Wan, Pengfei},
  journal={arXiv preprint arXiv:2306.03413},
  year={2023}
}

@article{zhang2023vis1st,
  title={1st Place Solution for the 5th LSVOS Challenge: Video Instance Segmentation},
  author={Zhang, Tao and Tian, Xingye and Zhou, Yikang and Wu, Yu and Ji, Shunping and Yan, Cilin and Wang, Xuebo and Tao, Xin and Zhang, Yuan and Wan, Pengfei},
  journal={arXiv preprint arXiv:2308.14392},
  year={2023}
}

@article{zhang2023vps1st,
  title={1st Place Solution for PVUW Challenge 2023: Video Panoptic Segmentation},
  author={Zhang, Tao and Tian, Xingye and Wei, Haoran and Wu, Yu and Ji, Shunping and Wang, Xuebo and Zhang, Yuan and Wan, Pengfei},
  journal={arXiv preprint arXiv:2306.04091},
  year={2023}
}

Acknowledgement

This repo is largely based on Mask2Former, MinVIS and VITA. Thanks for their excellent works.