UV Volumes for Real-time Rendering of Editable Free-view Human Performance

CVPR 2023

Yue Chen^1, Xuan Wang^2,3, Xingyu Chen¹, Qi Zhang³, Xiaoyu Li³, Yu Guo^1†, Jue Wang³, Fei Wang¹

^*Equal Contribution ^†Corresponding Author ¹Xi'an Jiaotong University ²Ant Group ³Tencent AI Lab

Paper

Code

Supplementary

Abstract

Neural volume rendering enables photo-realistic renderings of a human performer in free-view, a critical task in immersive VR/AR applications. But the practice is severely limited by high computational costs in the rendering process. To solve this problem, we propose the UV Volumes, a new approach that can render an editable free-view video of a human performer in real-time. It separates the high-frequency (i.e., non-smooth) human appearance from the 3D volume, and encodes them into 2D neural texture stacks (NTS). The smooth UV volumes allow much smaller and shallower neural networks to obtain densities and texture coordinates in 3D while capturing detailed appearance in 2D NTS. For editability, the mapping between the parameterized human model and the smooth texture coordinates allows us a better generalization on novel poses and shapes. Furthermore, the use of NTS enables interesting applications, e.g., retexturing. Extensive experiments on CMU Panoptic, ZJU Mocap, and H36M datasets show that our model can render 960 x 540 images in 30FPS on average with comparable photo-realism to state-of-the-art methods.

Pipeline

Our model has two main branches: 1) Based on a human pose 𝜃, a volume generator constructs UV volumes involving the feature of UV information. Then a feature map can be rendered via differentiable raymarching and decoded to texture coordinates (UV) pixel-by-pixel. 2) A texture generator produces a pose-dependent Neural Texture Stack(NTS) 𝓔 that encodes the highly-detailed appearance information. The UV coordinates and the texture embedding interpolated from NTS are passed into an MLP to predict the color at the desired ray direction 𝐝.