Dynamic and static mutual fitting for action recognition

Wenxuan Liu, Xuemei Jia, Xian Zhong, Kui Jiang, Xiaohan Yu, Mang Ye

Abstract: Highlights•A new temporal argument perspective for action recognition in videos.•A new pseudo occlusion strategy, emphasizing the real-world ill-posed problem.•A new framework captured the background auxiliary information and action regions.•A significance improvement both in argument and framework dimensions by our proposal.