Chen Mingjin ; Chen Junhao ; Ye Xiaojun ; Gao Huan-ang ; Chen Xiaoxue ; Fan Zhaoxin ; Zhao Hao

Abstract

3D human body reconstruction has been a challenge in the field of computervision. Previous methods are often time-consuming and difficult to capture thedetailed appearance of the human body. In this paper, we propose a new methodcalled \emph{Ultraman} for fast reconstruction of textured 3D human models froma single image. Compared to existing techniques, \emph{Ultraman} greatlyimproves the reconstruction speed and accuracy while preserving high-qualitytexture details. We present a set of new frameworks for human reconstructionconsisting of three parts, geometric reconstruction, texture generation andtexture mapping. Firstly, a mesh reconstruction framework is used, whichaccurately extracts 3D human shapes from a single image. At the same time, wepropose a method to generate a multi-view consistent image of the human bodybased on a single image. This is finally combined with a novel texture mappingmethod to optimize texture details and ensure color consistency duringreconstruction. Through extensive experiments and evaluations, we demonstratethe superior performance of \emph{Ultraman} on various standard datasets. Inaddition, \emph{Ultraman} outperforms state-of-the-art methods in terms ofhuman rendering quality and speed. Upon acceptance of the article, we will makethe code and data publicly available.