Baking Neural Radiance Fields for Real-Time View Synthesis

TL;DR

This paper introduces SNeRG, a novel sparse voxel grid representation that allows for real-time rendering of neural radiance fields by precomputing and storing scene information, enabling photorealistic view synthesis on standard hardware.

Contribution

The authors propose a new method to 'bake' NeRFs into a sparse voxel grid, significantly improving rendering speed while maintaining detail and view-dependent effects.

Findings

Real-time rendering at over 30 fps on a laptop GPU

Scene size reduced to less than 90 MB

Retains fine geometric and appearance details

Abstract

Neural volumetric representations such as Neural Radiance Fields (NeRF) have emerged as a compelling technique for learning to represent 3D scenes from images with the goal of rendering photorealistic images of the scene from unobserved viewpoints. However, NeRF's computational requirements are prohibitive for real-time applications: rendering views from a trained NeRF requires querying a multilayer perceptron (MLP) hundreds of times per ray. We present a method to train a NeRF, then precompute and store (i.e. "bake") it as a novel representation called a Sparse Neural Radiance Grid (SNeRG) that enables real-time rendering on commodity hardware. To achieve this, we introduce 1) a reformulation of NeRF's architecture, and 2) a sparse voxel grid representation with learned feature vectors. The resulting scene representation retains NeRF's ability to render fine geometric details and view-dependent appearance, is compact (averaging less than 90 MB per scene), and can be rendered in real-time (higher than 30 frames per second on a laptop GPU). Actual screen captures are shown in our video.