Representations for Modeling Human Bodies

Speaker: Zeng Huang () Abstract: Studying human bodies of ourselves has been a long carried out topic throughout human history. Since the Information Age, digitizing human bodies has always been a major focus in computer graphics and animation. Though high quality human scans and visual effects have already been used widely in the film industry, low-cost and accessible human digitization for every-one still remains a challenge. As people gradually adopted deep learning in this field, there has been some recent exciting work and really pushed the boundary of this task. In this talk, we will introduce research efforts in recent years taht digitizes full body clothed humans. Especially, we will go over the recent attempts of representing the body geometry using implicit functions, and its combination with animation pipelines and real-time implementations. Papers: 1. End-to-end Recovery of Human Shape and Pose, CVPR'2018 () 2. VIBE: Video Inference for Human B