3D Clothing Transfer in Virtual Fitting Based on UV Mapping

PubDate: February 2019

Teams: Beijing Language and Culture University

Writers: Ting Liu; Hong Xu; XiWen Zhang

PDF: 3D Clothing Transfer in Virtual Fitting Based on UV Mapping

Abstract

3D virtual fitting based on human body contours can greatly improve the try-on effect of some tight-fitting clothes. This paper presents a fitting approach requiring merely a single RGB image and transferring clothing texture onto the human body contour, under the instruction of a UV texture guide map. DensePose is used to segment a human body into 24 patches and predict UV coordinates of each patch. 24 clothing UV patches and their textures are extracted from SURREAL dataset. Clothing patches are filtered by the patch index and the alpha value of each pixel in the patch, avoiding unnecessary replacement such as users’ heads and exposed skin. They are subsequently transformed into human body corresponding patches by UV mapping. Human bodies with various figures, poses and perspectives can be fitted with clothes. Some users and clothes are tested, and satisfactory results are attained.