Efficient 3D Articulated Human Generation with Layered Surface Volumes

This paper introduces a novel approach to generating high-quality and diverse 3D articulated digital human assets, crucial for various applications such as virtual reality and social platforms. The authors propose a new 3D object representation for articulated digital humans, called Layered Surface Volumes (LSVs). LSVs represent a human body using multiple textured mesh layers around a conventional template, capturing fine off-surface details like hair or accessories. The LSV-GAN, trained on unstructured, single-view 2D image datasets, generates high-quality and view-consistent 3D articulated digital humans without the need for view-inconsistent 2D upsampling networks.

 

Publication date: July 11, 2023
Project Page: computationalimaging.org/publications/lsv/
Paper: https://arxiv.org/pdf/2307.05462.pdf