Enhancing Action Recognition with Channel Attention Modules in GCN

Skeleton-based GCN methods have found widespread application in the field of action recognition.. ST-GCN++ represents a significant advancement over the ST-GCN method, resulting in a remarkable enhancement in performance. However, when dealing with complex action data, there are still potential issues related to information loss and redundancy. To address these challenges, we incorporate the channel attention module into ST-GCN++. Specifically, we introduced Se Blocks into each branch of the MSTCN module. After experimental validation on the NTU60 dataset, our model has achieved state-of-the-art (SOTA) performance levels in the current research domain.