Video Human Action Recognition with Channel Attention on ST-GCN

Abstract Action recognition based on human skeleton information is a hot research topic in the field of computer vision, and ST-GCN graph convolutional network is widely used to extract spatial and temporal features of human skeleton to represent the human skeleton structure. However, in the process of extracting features, the weights on each channel of the feature are the same, so it is difficult to effectively discriminate the useful features from the useless ones. In this paper, we propose Channel Attention module, which learns the importance of each feature channel to perform human action recognition more effectively. Experimental results on Kinetics and NTU-RGB+D datasets show that Channel Attention module can achieve better accuracy.