Graph convolutional network with STC attention and adaptive normalization for skeleton-based action recognition

Graph Convolutional Network (GCN) have been widely used in the field of skeleton-based action recognition and have achieved exciting results. Introducing attention mechanism in the process of extracting skeleton features has always been a hot spot in GCN-related research. In this article, we design a new graph convolutional network, which combines the advanced decoupling graph convolutional network (DC-GCN) with spatial, temporal, channel (STC) series attention module and adaptive normalization (AN). The STC attention module helps the network tend to extract important information from skeleton features. In addition, in order to improve the adaptability of the normalization method to GCN, we design the AN module instead of the BN module, which can train the weights of different normalization methods, so that each normalization layer in the network adopts the most suitable normalization operation. The experimental results show that the accuracy of our method is competitive with the state-of-the-art action recognition methods.