CVPR2023

Normal-guided Garment UV Prediction for Human Re-texturing

Yasamin Jafarian, Tuanfeng Y. Wang, Duygu Ceylan, Jimei Yang, Nathan Carr, Yi Zhou, Hyun Soo Park

摘要

Target texture Original video Re-textured video Re-textured images Logo insertion Original image Re-textured images Original image Figure 1 . Geomety aware texture editing from images and videos. This paper presents a novel approach to predict a geometry aware texture (UV) map of a garment from an image (first row) and video (second row). The predicted UV map preserves isometry between texture space and 3D surface by leveraging 3D surface normals. Further, we ensure temporal consistency in the predicted UV map across frames in a video, resulting in physically plausible human appearance editing.