Multi-view Canonical Pose 3D Human Body Reconstruction Based on Volumetric TSDF

Xi Li

Abstract: In this report, we present our solution for track1, multi-view based 3D human body reconstruction, of the ECCV 2022 WCPA Challenge: From Face, Body and Fashion to 3D Virtual Avatars 1. We developed a variant network based on TetraTSDF to reconstruct detailed 3D human body models with canonical T-pose from that multi-view images of the same person with different actions and angles, which is called TetraTSDF++ in this report. This method first fuses the features of different views, then infers the volumetric truncated signed distance function (TSDF) in the preset human body shell, and finally obtains the human surface through marching cube algorithm. The best chamfer distance score of our solution is 0.9751, and our solution got the 2nd place on the leaderboard.