AI summaryⓘ
The authors introduce GAVIS, a new method to measure and handle uncertainty in 3D Gaussian Splatting (3DGS), which helps make 3D scene reconstructions more reliable. They found that parts of a scene not seen during training give unreliable results, so they created a way to track how visible each point is from the training views using spherical harmonics. This visibility information is then used with a Bayesian approach to quickly estimate uncertainty in real time. They also use this to actively decide the best new viewpoints to improve the map. Their tests show better accuracy and speed compared to previous methods, and their approach can enhance other existing techniques after they are applied.
Gaussian Splatting3D Scene ReconstructionUncertainty QuantificationVisibility FieldSpherical HarmonicsBayesian NetworkActive MappingMaximum Information GainReal-Time RenderingUncertainty-Aware Rasterization
Authors
Shangjie Xue, Jesse Dill, Dhruv Ahuja, Frank Dellaert, Panagiotis Tsiotras, Danfei Xu
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.