Abstract: A novel method for skeleton-based action recognition by fusing multi-level spatial features and multi-level temporal features is proposed in this article. Recently, Graph Convolutional ...