NerfAcc: A General NeRF Acceleration Toolbox

TL;DR

NerfAcc addresses the slow convergence and rendering bottlenecks of vanilla NeRFs by offering a general acceleration toolbox built on Instant-NGP-inspired pruning and rendering strategies. It decomposes rendering into ray marching and differentiable rendering, introduces an occupancy-grid and scene-contraction to support bounded, dynamic, and unbounded scenes, and provides a user-friendly PyTorch API for plug-and-play integration with diverse NeRF variants. The approach yields substantial speedups across multiple settings (e.g., vanilla NeRF in ~1 hour, Instant-NGP in ~4.5 minutes, D-NeRF in ~1 hour, unbounded scenes in ~20 minutes) with competitive or improved quality on standard datasets. This framework broadens the applicability of fast NeRF rendering and training, enabling researchers to accelerate a wide range of radiance-field models with minimal code changes.

Abstract

We propose NerfAcc, a toolbox for efficient volumetric rendering of radiance fields. We build on the techniques proposed in Instant-NGP, and extend these techniques to not only support bounded static scenes, but also for dynamic scenes and unbounded scenes. NerfAcc comes with a user-friendly Python API, and is ready for plug-and-play acceleration of most NeRFs. Various examples are provided to show how to use this toolbox. Code can be found here: https://github.com/KAIR-BAIR/nerfacc. Note this write-up matches with NerfAcc v0.3.5. For the latest features in NerfAcc, please check out our more recent write-up at arXiv:2305.04966