Hierarchical kernel decoupling for graph convolution: Enhancing skeleton-based action recognition through structured representation

Yanjun Chen

doi:10.1016/j.patcog.2025.112652

Hierarchical kernel decoupling for graph convolution: Enhancing skeleton-based action recognition through structured representation

2025-10-26

0

PRE

AI

DOI：10.1016/j.patcog.2025.112652

Original

Original request for help

Save

Abstract

En 中文

• We propose a novel HKD-GCN framework to decouple spatial graph convolution. • We propose K-hop neighborhood partitioning to construct multi-level receptive fields. • We propose a context-aware partition enhancer that adaptively reweights nodes. • The proposed HKD-GCN outperforms the state-of-the-art on popular benchmark datasets.

Hierarchical kernel decoupling for graph convolution: Enhancing skeleton-based action recognition through structured representation

Abstract

Journal

Organization