Research map / 3D vision / Gaussian splatting
PRIMU: Uncertainty Estimation for Novel Views in Gaussian Splatting from Primitive-Based Representations of Error and Coverage
Thomas Gottwald, Edgar Heinert, Peter Stehr, Chamuditha Jayanga Galappaththige, Matthias Rottmann
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use PRIMU primitive-based uncertainty feature maps with gradient boosting regression instead of prior GS UE methods | 3DGS · LF rendering error UE | AUSE: 0.753 (FisherRF) -> 0.291 | Improved |
| use PRIMU instead of BayesRays and FisherRF for depth UE | 3DGS · LF depth error UE | AUSE: 0.173 (BayesRays) -> 0.115 | Improved |
| add direction-dependent coverage and error maps (ablation) | PRIMU rendering UE · LF rendering error UE | AUSE: 0.291 -> 0.286 | Improved |
| add direction-dependent coverage and error maps (ablation) | PRIMU depth UE · LF depth error UE | AUSE: 0.115 -> 0.118 | No clear effect |
| use gradient boosting instead of linear regression for the pixel-wise regressor (ablation) | PRIMU · object-centric LF depth UE | AUSE: 0.246 -> 0.071 | Improved |
| add a meta-regression stage to FisherRF (regression-based FisherRF) (ablation) | FisherRF · LF rendering error UE, 1 view | AUSE: 0.878 -> 0.282 | Improved |
| train the regressor on more than one hold-out view (ablation) | PRIMU · LF entire-scene rendering UE | AUSE: 0.244 -> 0.222 | Improved |
| restrict training and evaluation to the central object (ablation) | PRIMU · LF rendering error UE | AUSE: 0.244 -> 0.072 | Improved |
| reduce 13 uncertainty feature maps to 9 (entire scene) or 6 (object-centric) (ablation) | PRIMU · LF and TUM UE | AUSE / Pearson: 0.244 -> 0.245 (LF entire-scene rendering AUSE, 9 maps) | No clear effect |
| use PRIMU direction-dependent coverage map (max aggregation, no opacity) for active view selection | 3DGS AVS · MipNeRF360, 20 views | PSNR: 20.392 (FisherRF) -> 20.721 | Improved |
| add FoV counter and max-aggregation coverage map without opacity (ablation) | PRIMU depth UE · LF and TUM backward regression | Pearson correlation | Improved |
| include rendering-error-derived feature maps for depth UE (ablation) | PRIMU depth UE · LF and TUM backward regression | Pearson correlation | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BayesRays — Bayes' Rays: Uncertainty Quantification for Neural Radiance Fields (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Variational Multi-scale Representation for Estimating Uncertainty in 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.87)
- PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting (CVPR 2025 · similarity 0.87)
- Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting (CVPR 2025 · similarity 0.86)
- Structure Consistent Gaussian Splatting with Matching Prior for Few-shot Novel View Synthesis (NeurIPS 2024 · similarity 0.86)
- 4D Gaussian Splatting in the Wild with Uncertainty-Aware Regularization (NeurIPS 2024 · similarity 0.86)
- Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting (ICML 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
