DONeRF: Towards Real-Time Rendering of Compact Neural Radiance Fields using Depth Oracle Networks

TL;DR

DONeRF introduces a novel neural rendering approach that significantly reduces inference costs by predicting surface locations with a depth oracle network, enabling real-time rendering of neural radiance fields without extra caching structures.

Contribution

The paper proposes DONeRF, a compact dual network system with a depth oracle network for efficient surface sampling, achieving up to 48x faster inference without additional memory overhead.

Findings

Reduces inference costs by up to 48x compared to NeRF.

Enables real-time rendering at 20 FPS on a single GPU.

Uses a classification network for surface location prediction rather than direct depth estimation.

Abstract

The recent research explosion around implicit neural representations, such as NeRF, shows that there is immense potential for implicitly storing high-quality scene and lighting information in compact neural networks. However, one major limitation preventing the use of NeRF in real-time rendering applications is the prohibitive computational cost of excessive network evaluations along each view ray, requiring dozens of petaFLOPS. In this work, we bring compact neural representations closer to practical rendering of synthetic content in real-time applications, such as games and virtual reality. We show that the number of samples required for each view ray can be significantly reduced when samples are placed around surfaces in the scene without compromising image quality. To this end, we propose a depth oracle network that predicts ray sample locations for each view ray with a single network evaluation. We show that using a classification network around logarithmically discretized and spherically warped depth values is essential to encode surface locations rather than directly estimating depth. The combination of these techniques leads to DONeRF, our compact dual network design with a depth oracle network as its first step and a locally sampled shading network for ray accumulation. With DONeRF, we reduce the inference costs by up to 48x compared to NeRF when conditioning on available ground truth depth information. Compared to concurrent acceleration methods for raymarching-based neural representations, DONeRF does not require additional memory for explicit caching or acceleration structures, and can render interactively (20 frames per second) on a single GPU.