Taming Modality Entanglement in Continual Audio-Visual Segmentation

TL;DR

This paper introduces a novel continual audio-visual segmentation task and proposes a Collision-based Multi-modal Rehearsal framework to address modality entanglement issues, significantly improving performance in fine-grained multi-modal continual learning.

Contribution

The paper presents a new CAVS task and a CMR framework with strategies to mitigate semantic drift and co-occurrence confusion in multi-modal continual learning.

Findings

Outperforms single-modal continual learning methods

Effectively addresses modality entanglement challenges

Demonstrates significant improvements in audio-visual segmentation

Abstract

Recently, significant progress has been made in multi-modal continual learning, aiming to learn new tasks sequentially in multi-modal settings while preserving performance on previously learned ones. However, existing methods mainly focus on coarse-grained tasks, with limitations in addressing modality entanglement in fine-grained continual learning settings. To bridge this gap, we introduce a novel Continual Audio-Visual Segmentation (CAVS) task, aiming to continuously segment new classes guided by audio. Through comprehensive analysis, two critical challenges are identified: 1) multi-modal semantic drift, where a sounding objects is labeled as background in sequential tasks; 2) co-occurrence confusion, where frequent co-occurring classes tend to be confused. In this work, a Collision-based Multi-modal Rehearsal (CMR) framework is designed to address these challenges. Specifically, for multi-modal semantic drift, a Multi-modal Sample Selection (MSS) strategy is proposed to select samples with high modal consistency for rehearsal. Meanwhile, for co-occurence confusion, a Collision-based Sample Rehearsal (CSR) mechanism is designed, allowing for the increase of rehearsal sample frequency of those confusable classes during training process. Moreover, we construct three audio-visual incremental scenarios to verify effectiveness of our method. Comprehensive experiments demonstrate that our method significantly outperforms single-modal continual learning methods.