Awesome
<div align="center">DVIS-DAQ: Improving Video Segmentation via Dynamic Anchor Queries
Yikang Zhou, Tao Zhang, Shunping Ji, Shuicheng Yan, Xiangtai Li
Project Page: https://zhang-tao-whu.github.io/projects/DVIS_DAQ
<img src="https://github.com/zhang-tao-whu/paper_images/blob/master/dvis_daq/dvis-daq-teaser.png" width="800"/> </div>News
- This work is accepted by ECCV-2024.
TODO
- Release the code and pre-trained weights.
Features
- We propose Dynamic Anchor Queries and Emergence and Disappearance Simulation to significantly enhance the ability to handle the emergence and disappearance of objects.
- The Dynamic Anchor Queries and Emergence and Disappearance Simulation proposed by us can be easily integrated into existing query-based video segmentation methods.
- By integrating with DVIS, we constructed DVIS-DAQ and achieved SOTA performance on YTVIS 19/21/22 and VIPSeg, with performance reaching as high as 57.1 AP on OVIS.
Demos
We provide visualization results for multiple challenging cases on the project page.
<a name="CitingDAQ"></a>Citing DVIS-DAQ
@article{dvisdaq,
title={DVIS-DAQ: Improving Video Segmentation via Dynamic Anchor Queries},
author={Yikang Zhou and Tao Zhang and Shunping Ji and Shuicheng Yan and Xiangtai Li},
journal={ECCV},
year={2024},
}