Frequency Guidance Matters: Skeletal Action Recognition by Frequency-Aware Mixed Transformer

TL;DR

This paper introduces FreqMixFormer, a frequency-aware mixed transformer that enhances skeleton action recognition by capturing frequency-specific features and modeling comprehensive spatiotemporal patterns, outperforming state-of-the-art methods.

Contribution

The paper proposes a novel frequency-aware attention module and a mixed transformer architecture to better distinguish subtle skeletal motions in action recognition.

Findings

Outperforms SOTA on NTU RGB+D, NTU RGB+D 120, NW-UCLA datasets

Effectively captures frequency-specific discriminative features

Improves recognition of similar skeletal actions

Abstract

Recently, transformers have demonstrated great potential for modeling long-term dependencies from skeleton sequences and thereby gained ever-increasing attention in skeleton action recognition. However, the existing transformer-based approaches heavily rely on the naive attention mechanism for capturing the spatiotemporal features, which falls short in learning discriminative representations that exhibit similar motion patterns. To address this challenge, we introduce the Frequency-aware Mixed Transformer (FreqMixFormer), specifically designed for recognizing similar skeletal actions with subtle discriminative motions. First, we introduce a frequency-aware attention module to unweave skeleton frequency representations by embedding joint features into frequency attention maps, aiming to distinguish the discriminative movements based on their frequency coefficients. Subsequently, we develop a mixed transformer architecture to incorporate spatial features with frequency features to model the comprehensive frequency-spatial patterns. Additionally, a temporal transformer is proposed to extract the global correlations across frames. Extensive experiments show that FreqMiXFormer outperforms SOTA on 3 popular skeleton action recognition datasets, including NTU RGB+D, NTU RGB+D 120, and NW-UCLA datasets.