Multi-Scale Adaptive Skeleton Transformer for action recognition

Xiaotian Wang, Kai Chen, Zhifu Zhao, Guangming Shi, Xuemei Xie, Xiang Jiang, Yifan Yang

Abstract: Highlights•MSAST: a novel Transformer model for skeleton-based action recognition.•ASPEM decouples position encoding to capture sample-specific latent dependencies.•MSEM generates multi-scale tokens for multi-scale feature extraction.•ARLM learns unique location information for various samples.•State-of-the-art results on NTU-60 dataset.