Human Pose as Compositional Tokens 

                      Zigang Geng     Chunyu Wang     Yixuan Wei     Ze Liu     Houqiang Li     Han Hu     

University of Science and Technology of China     Tsinghua University     Microsoft

Method

In this paper, we present a structured representation PCT, to explore the joint dependency. It represents a pose by M discrete tokens with each characterizing a sub-structure with several interdependent joints. It achieves significantly better results when occlusion occurs.

Qualitative Results