Rendering-Aware Bayesian 3D Gaussian Splatting with Native Uncertainty and Adaptive Complexity Control

2026-07-06

AI Summary

Standard 3D Gaussian splatting lacks native uncertainty estimation and adaptive complexity control, making it difficult to identify weakly supported geometry or actively select informative novel viewpoints under sparse-view conditions. This work proposes the first Bayesian 3D Gaussian splatting framework, modeling Gaussian geometry via a Normal-Inverse-Wishart posterior and employing a Dirichlet process prior to enable adaptive component selection. The method incorporates rendering-aware surrogate summarization for efficient inference while explicitly delineating the boundaries between closed-form and approximate inference. It supports native uncertainty quantification and Bayesian active view selection, achieving a PSNR gain of 0.453 dB and an LPIPS reduction of 0.0146 in the 16→32 view task. The 95% coverage error is reduced by approximately 17× compared to surrogate models, with training costs only one-third of those required by deep ensembles.

Abstract

3D Gaussian splatting (3DGS) is a strong representation for real-time novel-view synthesis, but its standard training pipeline relies on point estimates and hand-tuned heuristics, providing no native uncertainty or principled complexity control. This is most limiting under sparse views or fixed acquisition budgets, where a model must identify weakly supported geometry and select informative views. We introduce a rendering-aware Bayesian 3DGS framework that tracks Gaussian geometry with a Normal-Inverse-Wishart posterior over means and covariances using renderer-derived surrogate summaries. An optional Dirichlet-process extension adds a probabilistic component-usage signal, and the training schedule makes the closed-form versus approximate inference boundary explicit. Re-rendering posterior geometry samples yields native predictive uncertainty for interval calibration and active view selection. In a fixed-budget 16-to-32 active-view task, native NIW acquisition improves PSNR by +0.453 dB and LPIPS by -0.0146 over a scoring-only 3-member standard-ensemble baseline, winning 29/39 scene-seed pairs and 10/13 scene means; it also improves over PPU-style (+0.355 dB) and NIW-proxy (+0.401 dB) acquisition. NIW native intervals reduce 95% coverage error by about 17x relative to a shared proxy (0.046 vs. 0.796) and are about 10x closer to nominal coverage than a 3-member deep ensemble (0.047 vs. 0.454) at roughly one-third the training cost. As a reconstruction compatibility check, paired NIW-vs-standard analysis over 39 scene-seed runs yields +0.030 dB PSNR with 1.6% additional training time. These results position Bayesian 3DGS as a practical probabilistic scene representation for decision-facing tasks such as active view selection.

Problem

• 3D Gaussian splatting
• uncertainty quantification
• complexity control
• active view selection
• sparse views

Innovation

• Bayesian 3D Gaussian Splatting
• Normal-Inverse-Wishart posterior
• native uncertainty
• adaptive complexity control
• active view selection