Uncertainty-driven 3D Gaussian Splatting Active Mapping via Anisotropic Visibility Field

2026-05-28

AI Summary

This work addresses the limitations of 3D Gaussian Splatting (3DGS) in rendering unobserved regions, its lack of uncertainty quantification, and absence of active mapping capability by introducing a novel approach based on an anisotropic visibility field. The method models the visibility of each Gaussian primitive relative to training viewpoints using spherical harmonics and integrates this representation into a Bayesian rasterizer to enable principled, efficient, real-time uncertainty estimation. By further incorporating a maximum information gain strategy, the framework supports active mapping. Evaluated across diverse scenes, the proposed method significantly improves reconstruction accuracy and efficiency, achieving real-time performance at up to 200 FPS, and functions as a general-purpose post-processing module that can enhance existing 3DGS pipelines.

Abstract

We present Gaussian Splatting Anisotropic Visibility Field (GAVIS), a novel framework for uncertainty quantification and active mapping in 3DGS. Our key insight is that regions unseen from the training views yield unreliable predictions from the 3DGS. To address this, we introduce a principled and efficient method for quantifying the visibility field in 3DGS, defined as the anisotropic visibility of each particle with respect to the training views, and represented using spherical harmonics. The resulting visibility field is integrated into a Bayesian Network-based uncertainty-aware 3DGS rasterizer, enabling real-time (200 FPS) uncertainty quantification for synthesized views. Active mapping is further performed within a maximum information gain framework building on this formulation. Extensive experiments across diverse environments demonstrate that GAVIS consistently and significantly outperforms prior approaches in both accuracy and efficiency. Moreover, beyond standalone use, our method can be applied post-hoc to improve the performance of existing approaches.

Problem

• 3D Gaussian Splatting
• uncertainty quantification
• active mapping
• visibility field
• anisotropic

Innovation

• 3D Gaussian Splatting
• Anisotropic Visibility Field
• Uncertainty Quantification
• Active Mapping
• Spherical Harmonics