WarpRF: Multi-View Consistency for Training-Free Uncertainty Quantification and Applications in Radiance Fields

TL;DR

WarpRF is a training-free, versatile framework that quantifies uncertainty in radiance fields by leveraging multi-view consistency, enhancing applications like view selection and mapping without additional training.

Contribution

It introduces WarpRF, a novel, training-free method for uncertainty quantification in radiance fields based on multi-view consistency, applicable across various models.

Findings

Outperforms existing uncertainty methods in radiance fields

Enables effective active view selection and mapping

Does not require training or additional data

Abstract

We introduce WarpRF, a training-free general-purpose framework for quantifying the uncertainty of radiance fields. Built upon the assumption that photometric and geometric consistency should hold among images rendered by an accurate model, WarpRF quantifies its underlying uncertainty from an unseen point of view by leveraging backward warping across viewpoints, projecting reliable renderings to the unseen viewpoint and measuring the consistency with images rendered there. WarpRF is simple and inexpensive, does not require any training, and can be applied to any radiance field implementation for free. WarpRF excels at both uncertainty quantification and downstream tasks, e.g., active view selection and active mapping, outperforming any existing method tailored to specific frameworks.