Audio-Visual Instance Segmentation

TL;DR

This paper introduces the novel task of audio-visual instance segmentation, a new benchmark dataset AVISeg, and a baseline model, advancing the understanding of sounding object localization, segmentation, and tracking in videos.

Contribution

It presents the first comprehensive framework for audio-visual instance segmentation, including a large-scale dataset and a strong baseline model for the task.

Findings

Our model outperforms existing methods on AVISeg.

Multi-modal large models show limited performance on instance-level tasks.

AVISeg dataset enables new research directions in audio-visual understanding.

Abstract

In this paper, we propose a new multi-modal task, termed audio-visual instance segmentation (AVIS), which aims to simultaneously identify, segment and track individual sounding object instances in audible videos. To facilitate this research, we introduce a high-quality benchmark named AVISeg, containing over 90K instance masks from 26 semantic categories in 926 long videos. Additionally, we propose a strong baseline model for this task. Our model first localizes sound source within each frame, and condenses object-specific contexts into concise tokens. Then it builds long-range audio-visual dependencies between these tokens using window-based attention, and tracks sounding objects among the entire video sequences. Extensive experiments reveal that our method performs best on AVISeg, surpassing the existing methods from related tasks. We further conduct the evaluation on several multi-modal large models. Unfortunately, they exhibits subpar performance on instance-level sound source localization and temporal perception. We expect that AVIS will inspire the community towards a more comprehensive multi-modal understanding. Dataset and code is available at https://github.com/ruohaoguo/avis.