Spatio-temporal Relation Modeling for Few-shot Action Recognition

TL;DR

This paper introduces STRM, a novel few-shot action recognition framework that enhances feature discriminability and temporal understanding through spatio-temporal enrichment modules, achieving state-of-the-art results on multiple benchmarks.

Contribution

The paper proposes a new spatio-temporal enrichment module and a query-class similarity classifier for improved few-shot action recognition.

Findings

Achieves state-of-the-art accuracy on Kinetics, SSv2, HMDB51, and UCF101.

Provides a 3.5% accuracy gain on SSv2 over previous best methods.

Demonstrates the effectiveness of local patch and global frame enrichment in action recognition.

Abstract

We propose a novel few-shot action recognition framework, STRM, which enhances class-specific feature discriminability while simultaneously learning higher-order temporal representations. The focus of our approach is a novel spatio-temporal enrichment module that aggregates spatial and temporal contexts with dedicated local patch-level and global frame-level feature enrichment sub-modules. Local patch-level enrichment captures the appearance-based characteristics of actions. On the other hand, global frame-level enrichment explicitly encodes the broad temporal context, thereby capturing the relevant object features over time. The resulting spatio-temporally enriched representations are then utilized to learn the relational matching between query and support action sub-sequences. We further introduce a query-class similarity classifier on the patch-level enriched features to enhance class-specific feature discriminability by reinforcing the feature learning at different stages in the proposed framework. Experiments are performed on four few-shot action recognition benchmarks: Kinetics, SSv2, HMDB51 and UCF101. Our extensive ablation study reveals the benefits of the proposed contributions. Furthermore, our approach sets a new state-of-the-art on all four benchmarks. On the challenging SSv2 benchmark, our approach achieves an absolute gain of $3.5\%$ in classification accuracy, as compared to the best existing method in the literature. Our code and models are available at https://github.com/Anirudh257/strm.