When Variance Is Not an Error Map: Calibrated Uncertainty for Radiative Gaussian Splatting in Sparse-View CT
Not on the map. Why?
Abstract (arXiv)
Does an uncertainty map identify where a reconstruction is wrong? In sparse-view computed tomography (CT), we find a sharp gap between whole-volume evaluation and error localization inside the object. We derive clamp-aware analytic moments for factorized Gaussian-density distributions, with a variance pass through existing rendering interfaces that is 7.9× faster than a 16-sample estimator. On a 15-scene benchmark, median variance--error Spearman correlation falls from 0.846 over the whole volume to 0.108 in foreground. The pattern recurs across representations and acquisition settings. Two analyses help explain the discrepancy: region contrast dominates global covariance, while 72--96% of in-object squared error is shared across independently trained members. Spread is unchanged by a common error, although shared error alone does not determine ranking. Correcting the offset between the deployed reconstruction and predictive mean improves foreground correlation by only 0.0014. The diagnosis separates two remedies. A log-normal control improves scale transfer without restoring localization; a supervised error predictor raises foreground correlation to 0.427 on the benchmark and 0.566 on eight held-out human subjects under simulated acquisition. A verified retrospective re-execution on eight additional subjects retains a median of 0.537, with the same frozen predictors. The central lesson is to validate uncertainty in the region, and against the error target, for which it will be used.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 24 of its 33 references that are in the map. First posted on arXiv 15 Jul 2026.
Builds on
24 of 33 references.
References · 24
- 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,141
- 20163D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotationcited by 1,471
- 2021The Medical Segmentation Decathloncited by 620
- 2021Solving Inverse Problems in Medical Imaging with Score-Based Generative Modelscited by 514
- 2018Accurate Uncertainties for Deep Learning Using Calibrated Regressioncited by 417
- 2016Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challengecited by 198
- 2018Uncertainty Estimates and Multi-Hypotheses Networks for Optical Flowcited by 135
- 2022ActiveNeRF: Learning where to See with Uncertainty Estimationcited by 124
- 2022Image-to-Image Regression with Distribution-Free Uncertainty Quantification and Applications in Imagingcited by 107
- 2022NAF: Neural Attenuation Fields for Sparse-View CBCT Reconstructioncited by 106
- 2023Bayes' Rays: Uncertainty Quantification for Neural Radiance Fieldscited by 90
- 2019Evaluating and Calibrating Uncertainty Prediction in Regression Taskscited by 85
12 more
- 2021Stochastic Neural Radiance Fields: Quantifying Uncertainty in Implicit 3D Representationscited by 80
- 2024R2-Gaussian: Rectifying Radiative Gaussian Splatting for Tomographic Reconstructioncited by 65
- 2018Deep Bayesian Inversioncited by 60
- 2022Density-aware NeRF Ensembles: Quantifying Predictive Uncertainty in Neural Radiance Fieldscited by 58
- 2024Radiative Gaussian Splatting for Efficient X-ray Novel View Synthesiscited by 56
- 2022UncertaINR: Uncertainty Quantification of End-to-End Implicit Neural Representations for Computed Tomographycited by 19
- 2022Uncertainty Quantification for Deep Unrolling-Based Computational Imagingcited by 10
- 2024Variational Bayes Gaussian Splattingcited by 6
- 2026VarSplat: Uncertainty-aware 3D Gaussian Splatting for Robust RGB-D SLAMcited by 3
- 2026Rendering-Aware Bayesian 3D Gaussian Splatting with Native Uncertainty and Adaptive Complexity Controlcited by 2
- 2026Radioactive 3D Gaussian Ray Tracing for Tomographic Reconstructioncited by 1
- 2026Active View Selection with Perturbed Gaussian Ensemble for Tomographic Reconstructioncited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.