TCD-GCN-Light: A Lightweight Temporal-Channel Decoupling Graph Convolutional Network for human early action prediction based on channel fusion

Open Webpage

Xianshan Li, Yuan Dong, Guoliang Jia, Fengda Zhao, Pengwei Zhang

Published: 2025, Last Modified: 04 Nov 2025Expert Syst. Appl. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Adopting a lightweight design to reduce model training time and parameter count.•Utilizing multi-stream fusion to enhance early human action prediction performance.•Using a training method with random observation ratio based on normal distribution.•Employing a single model to predict action samples across all observation ratios.•Significantly improve the prediction accuracy of data with low observation ratios.