A module selection-based approach for efficient skeleton human action recognition

doi:10.1016/j.displa.2025.103233

A module selection-based approach for efficient skeleton human action recognition

2025-09-30

0

PRE

AI

DOI：10.1016/j.displa.2025.103233

Original

Original request for help

Save

Abstract

En 中文

• Adaptive module selection strategy for skeleton-based action recognition is proposed. • Conventional methods employ fixed network whereas our approach is adaptive. • Introduces sparsity loss to balance accuracy and efficiency. • Reduces inference time by up to 3×while maintaining the performance. • Compatible with multiple SOTA models.

Keywords:

Graph convolutional networks

Skeleton human action recognition

Human–computer interaction

Efficient network

Dynamic network

A module selection-based approach for efficient skeleton human action recognition

Abstract

Journal

Organization