View-Dependent Uncertainty Estimation of 3D Gaussian Splatting

TL;DR

This paper tackles uncertainty estimation for 3D Gaussian Splatting by introducing a view-dependent per-gaussian uncertainty modeled with spherical harmonics. The approach mirrors the existing view-dependent color representation to maintain interpretability and compatibility with the 3DGS training pipeline, while offering a faster alternative to ensembles. Empirical results on a NeRF-based dataset show the method yields competitive uncertainty estimates with clear view-dependency, though it trails ensemble accuracy in some cases due to a simplified loss. The work enables practical applications such as active view planning and object completion by providing a lightweight, interpretable uncertainty cue alongside 3DGS reconstructions.

Abstract

3D Gaussian Splatting (3DGS) has become increasingly popular in 3D scene reconstruction for its high visual accuracy. However, uncertainty estimation of 3DGS scenes remains underexplored and is crucial to downstream tasks such as asset extraction and scene completion. Since the appearance of 3D gaussians is view-dependent, the color of a gaussian can thus be certain from an angle and uncertain from another. We thus propose to model uncertainty in 3DGS as an additional view-dependent per-gaussian feature that can be modeled with spherical harmonics. This simple yet effective modeling is easily interpretable and can be integrated into the traditional 3DGS pipeline. It is also significantly faster than ensemble methods while maintaining high accuracy, as demonstrated in our experiments.

Figures (4)

• Figure 1: Uncertainty types intuition
• Figure 2: Uncertainty visualization from different angles, some of which are close to the training views.
• Figure 3: Uncertainty visualization from an unseen view in the poster scene.
• Figure 4: A sparsification of the poster scene.