Boosting Unsupervised Video Instance Segmentation with Automatic Quality-Guided Self-Training

TL;DR

AutoQ-VIS introduces a quality-guided self-training framework for unsupervised video instance segmentation, effectively bridging the synthetic-to-real domain gap and achieving state-of-the-art results without human annotations.

Contribution

The paper proposes AutoQ-VIS, a novel unsupervised framework that uses quality-guided self-training to improve video instance segmentation from synthetic to real videos.

Findings

Achieves 52.6 AP50 on YouTubeVIS-2019 val set.

Surpasses previous state-of-the-art VideoCutLER by 4.4%.

Requires no human annotations.

Abstract

Video Instance Segmentation (VIS) faces significant annotation challenges due to its dual requirements of pixel-level masks and temporal consistency labels. While recent unsupervised methods like VideoCutLER eliminate optical flow dependencies through synthetic data, they remain constrained by the synthetic-to-real domain gap. We present AutoQ-VIS, a novel unsupervised framework that bridges this gap through quality-guided self-training. Our approach establishes a closed-loop system between pseudo-label generation and automatic quality assessment, enabling progressive adaptation from synthetic to real videos. Experiments demonstrate state-of-the-art performance with 52.6 $\text{AP}_{50}$ on YouTubeVIS-2019 $\texttt{val}$ set, surpassing the previous state-of-the-art VideoCutLER by 4.4%, while requiring no human annotations. This demonstrates the viability of quality-aware self-training for unsupervised VIS. We will release the code at https://github.com/wcbup/AutoQ-VIS.