ReLU Fields: The Little Non-linearity That Could

TL;DR

ReLU Fields introduce a minimal modification to grid-based representations by applying a ReLU non-linearity, enabling high-quality modeling of complex scenes with faster training and inference, rivaling traditional MLPs.

Contribution

The paper demonstrates that adding a simple ReLU non-linearity to grid-based models significantly improves their fidelity, matching MLPs while maintaining efficiency.

Findings

ReLU Fields achieve high-quality scene modeling.

The approach is competitive with state-of-the-art methods.

Fast training and inference are possible with minimal changes.

Abstract

In many recent works, multi-layer perceptions (MLPs) have been shown to be suitable for modeling complex spatially-varying functions including images and 3D scenes. Although the MLPs are able to represent complex scenes with unprecedented quality and memory footprint, this expressive power of the MLPs, however, comes at the cost of long training and inference times. On the other hand, bilinear/trilinear interpolation on regular grid based representations can give fast training and inference times, but cannot match the quality of MLPs without requiring significant additional memory. Hence, in this work, we investigate what is the smallest change to grid-based representations that allows for retaining the high fidelity result of MLPs while enabling fast reconstruction and rendering times. We introduce a surprisingly simple change that achieves this task -- simply allowing a fixed non-linearity (ReLU) on interpolated grid values. When combined with coarse to-fine optimization, we show that such an approach becomes competitive with the state-of-the-art. We report results on radiance fields, and occupancy fields, and compare against multiple existing alternatives. Code and data for the paper are available at https://geometry.cs.ucl.ac.uk/projects/2022/relu_fields.