Dynamics of human body skeletons convey significant information for humanaction recognition. Conventional approaches for modeling skeletons usually relyon hand-crafted parts or traversal rules, thus resulting in limited expressivepower and difficulties of generalization. In this work, we propose a novelmodel of dynamic skeletons called Spatial-Temporal Graph Convolutional Networks(ST-GCN), which moves beyond the limitations of previous methods byautomatically learning both the spatial and temporal patterns from data. Thisformulation not only leads to greater expressive power but also strongergeneralization capability. On two large datasets, Kinetics and NTU-RGBD, itachieves substantial improvements over mainstream methods.