MA-AVT: Modality Alignment for Parameter-Efficient Audio-Visual Transformers

TL;DR

This paper introduces MA-AVT, a parameter-efficient audio-visual transformer that employs deep modality alignment, hierarchical feature alignment, and foreground feature suppression to improve multimodal learning performance.

Contribution

It proposes a novel modality alignment method with joint token learning, blockwise contrastive learning, and foreground mining for enhanced audio-visual transformer performance.

Findings

Achieves state-of-the-art results on AVE, VGGSound, and CREMA-D datasets.

Demonstrates the effectiveness of deep hierarchical feature alignment.

Shows significant performance improvements over existing methods.

Abstract

Recent advances in pre-trained vision transformers have shown promise in parameter-efficient audio-visual learning without audio pre-training. However, few studies have investigated effective methods for aligning multimodal features in parameter-efficient audio-visual transformers. In this paper, we propose MA-AVT, a new parameter-efficient audio-visual transformer employing deep modality alignment for corresponding multimodal semantic features. Specifically, we introduce joint unimodal and multimodal token learning for aligning the two modalities with a frozen modality-shared transformer. This allows the model to learn separate representations for each modality, while also attending to the cross-modal relationships between them. In addition, unlike prior work that only aligns coarse features from the output of unimodal encoders, we introduce blockwise contrastive learning to align coarse-to-fine-grain hierarchical features throughout the encoding phase. Furthermore, to suppress the background features in each modality from foreground matched audio-visual features, we introduce a robust discriminative foreground mining scheme. Through extensive experiments on benchmark AVE, VGGSound, and CREMA-D datasets, we achieve considerable performance improvements over SOTA methods.