Vertex Features for Neural Global Illumination

TL;DR

This paper introduces neural vertex features that store learnable representations directly at mesh vertices, significantly reducing memory use while maintaining rendering quality in neural rendering tasks.

Contribution

It proposes a novel vertex-based neural representation that improves memory efficiency and aligns with surface geometry for neural rendering applications.

Findings

Memory consumption reduced to one-fifth of grid-based methods

Maintains comparable rendering quality

Lower inference overhead

Abstract

Recent research on learnable neural representations has been widely adopted in the field of 3D scene reconstruction and neural rendering applications. However, traditional feature grid representations often suffer from substantial memory footprint, posing a significant bottleneck for modern parallel computing hardware. In this paper, we present neural vertex features, a generalized formulation of learnable representation for neural rendering tasks involving explicit mesh surfaces. Instead of uniformly distributing neural features throughout 3D space, our method stores learnable features directly at mesh vertices, leveraging the underlying geometry as a compact and structured representation for neural processing. This not only optimizes memory efficiency, but also improves feature representation by aligning compactly with the surface using task-specific geometric priors. We validate our neural representation across diverse neural rendering tasks, with a specific emphasis on neural radiosity. Experimental results demonstrate that our method reduces memory consumption to only one-fifth (or even less) of grid-based representations, while maintaining comparable rendering quality and lowering inference overhead.