Multi-Resolution Audio-Visual Feature Fusion for Temporal Action Localization

TL;DR

This paper proposes MRAV-FF, a novel hierarchical gated cross-attention method for multi-resolution audio-visual feature fusion, significantly improving temporal action localization accuracy by integrating audio cues across different temporal scales.

Contribution

It introduces a hierarchical gated cross-attention mechanism for effective multi-resolution audio-visual feature fusion in TAL, enhancing existing FPN architectures with audio integration.

Findings

Improved boundary regression accuracy

Enhanced classification confidence with audio data

Versatile integration with existing TAL architectures

Abstract

Temporal Action Localization (TAL) aims to identify actions' start, end, and class labels in untrimmed videos. While recent advancements using transformer networks and Feature Pyramid Networks (FPN) have enhanced visual feature recognition in TAL tasks, less progress has been made in the integration of audio features into such frameworks. This paper introduces the Multi-Resolution Audio-Visual Feature Fusion (MRAV-FF), an innovative method to merge audio-visual data across different temporal resolutions. Central to our approach is a hierarchical gated cross-attention mechanism, which discerningly weighs the importance of audio information at diverse temporal scales. Such a technique not only refines the precision of regression boundaries but also bolsters classification confidence. Importantly, MRAV-FF is versatile, making it compatible with existing FPN TAL architectures and offering a significant enhancement in performance when audio data is available.