The article introduces ORTexME, a new method for 3D human shape and pose estimation from monocular videos. ORTexME improves the accuracy of estimations by using temporal information from the video to better regularize occluded body parts. It uses a novel average texture learning approach to determine reliable regions for ray sampling and to infer a mask based on the average texture. The method also uses a human body mesh to guide updates in the opacity field and reduce blur and noise. The method significantly improves results on the challenging multi-person 3DPW dataset, reducing error by 1.8 P-MPJPE.
Publication date: 21 Sep 2023
Project Page: https://arxiv.org/abs/2309.12183v1
Paper: https://arxiv.org/pdf/2309.12183
