RobustNeRF: Ignoring Distractors with Robust Losses

TL;DR

RobustNeRF introduces a robust training method for neural radiance fields that effectively removes distractors like moving objects and lighting variations, enhancing scene synthesis quality without complex pre-processing.

Contribution

It presents a simple, hyper-parameter-light robust estimation approach for NeRF training that handles distractors as outliers without prior knowledge of their nature.

Findings

Successfully removes distractors from synthetic and real scenes.

Improves scene rendering quality over baseline methods.

Easy to integrate into existing NeRF frameworks.

Abstract

Neural radiance fields (NeRF) excel at synthesizing new views given multi-view, calibrated images of a static scene. When scenes include distractors, which are not persistent during image capture (moving objects, lighting variations, shadows), artifacts appear as view-dependent effects or 'floaters'. To cope with distractors, we advocate a form of robust estimation for NeRF training, modeling distractors in training data as outliers of an optimization problem. Our method successfully removes outliers from a scene and improves upon our baselines, on synthetic and real-world scenes. Our technique is simple to incorporate in modern NeRF frameworks, with few hyper-parameters. It does not assume a priori knowledge of the types of distractors, and is instead focused on the optimization problem rather than pre-processing or modeling transient objects. More results on our page https://robustnerf.github.io.