Local and global self-attention enhanced graph convolutional network for skeleton-based action recognition

Zhize Wu, Yue Ding, Long Wan, Teng Li, Fudong Nian

Abstract: Highlights•We build a local-global graph convolutional network to extract spatial feature reps.•We develop a local-global temporal convolutional network for joint global temporal modeling.•We introduce a dynamic frame weighting module to model human action dynamics efficiently.•Extensive experiments confirm the effectiveness and efficiency of our method.