Spatio-temporal graph neural network based child action recognition using data-efficient methods: A systematic analysis

Sanka Mohottala

doi:10.1016/j.cviu.2025.104410

Spatio-temporal graph neural network based child action recognition using data-efficient methods: A systematic analysis

2025-06-03

0

PRE

AI

S

Sanka Mohottala *

A

Asiri Gawesha Lindamulage

D

Dharshana Kasthurirathna

P

Pradeepa Samarasinghe

C

Charith Abhayaratne

DOI：10.1016/j.cviu.2025.104410

Original

Original request for help

Save

Abstract

En 中文

• First systematic skeleton-based study of child actions in lab, wild, deployment. • GNN architectural factors adapted to enhance child action recognition performance. • Transfer learning on skeleton based actions varies with source and target properties. • ST-GNN performance on children varies with age, reflecting developmental stages. • Pose model limits stem from estimation errors, not from specific action classes.

Keywords:

skeleton-based action recognition

graph neural networks

transfer learning

child development

pose estimation errors

Spatio-temporal graph neural network based child action recognition using data-efficient methods: A systematic analysis

Abstract

Journal

Organization