Posterior Variance Is a Constraint Map, Not an Error Map: Closed-Form Uncertainty for Radiative Gaussian Splatting in Sparse-View CT

Abstract:Radiative Gaussian splatting reconstructs sparse-view CT fast and accurately, and recent work attaches per-Gaussian posteriors to yield per-voxel uncertainty maps. We ask what such a map actually measures: posterior variance is a data-constraint map, not an error map -- its alarms are trustworthy, its all-clears are not. Exploiting the strict linearity of X-ray rendering in the per-Gaussian densities, we derive a clamp-aware closed form that the unchanged rasterizer evaluates exactly in one forward pass, in volume and projection space: the infinite-sample limit of the sampling estimator of concurrent work, at ~8x lower cost. On the official 15-scene benchmark this uncertainty ranks true error on 14 of 15 scenes. Restricted to the object interior -- the tissue a clinician reads -- the ranking collapses (median Spearman 0.11, 0/15 pass), identically for a deep ensemble and for a strictly positive log-normal posterior: three constructions, two estimator families, no survivors. The mechanism is structural: about 90% of in-object error is bias that reproduces across retrainings, invisible to model disagreement; 73-81% of the full-volume correlation is carried by object/surround contrast; and an exactly solvable control puts the observed in-object ranking 4-5x below what a perfectly calibrated posterior with the same sigma-spread would score. The error scale, by contrast, is an engineering problem, and we solve it: reparameterizing the posterior contracts the cross-scene temperature spread from 19.3x to 2.6x, one scene-agnostic temperature transfers to unseen scenes (10/15 leave-one-scene-out), and the repaired scale tracks photon count at the Poisson-predicted -1/2 power. We distill evaluation practice that would have caught the illusion -- masked calibration, seed-wise bias decomposition, an exact-posterior reference -- and release all protocols, seeds and per-run evidence.

Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV)

arXiv:2607.13682 [cs.CV]

(or arXiv:2607.13682v2 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2607.13682 Focus to learn more arXiv-issued DOI via DataCite

evidence/cgca/abdomen_finals.json

evidence/cgca/chest_finals.json

evidence/closed_form/abdomen25_projection.json

evidence/closed_form/chest25_projection.json

evidence/closed_form/head25_projection.json

evidence/closed_form/var_d_abdomen25_s0_volume.json

evidence/closed_form/var_d_abdomen25_s0_volume_finalgrid.json

evidence/closed_form/var_d_abdomen50_s0_volume.json

evidence/closed_form/var_d_chest25_s0_volume.json

evidence/closed_form/var_d_chest25_s0_volume_finalgrid.json

evidence/closed_form/var_d_head25_s0_volume.json

evidence/closed_form/var_d_head25_s0_volume_finalgrid.json

evidence/closed_form/var_d_head50_s0_volume.json

evidence/dose/counts_provenance.json

evidence/dose/summary_dose.json

evidence/dose/summary_initcv.json

evidence/kl_ablation/var_kl0.json

evidence/kl_ablation/var_kl1e_2.json

evidence/kl_ablation/var_kl1e_3.json

evidence/kl_ablation/var_kl1e_4.json

evidence/lognormal/summary_strict.json

evidence/lognormal/summary_strict_noanneal.json

evidence/masked/d1_masked_ensemble_perscene_t25.json

evidence/masked/d1_masked_perrun.json

evidence/masked/d1_masked_perscene_t25.json

evidence/sixstep/residual_backprojection_chest.json

evidence/sixstep/seed_decomposition_diag_a.json

evidence/sixstep/structure_probes_diag_d.json

evidence/tau_transfer/d2_tau_transfer.json

evidence/tau_transfer/d2_tau_transfer_ensemble.json

evidence/tau_transfer/d2_tau_transfer_lognormal.json

evidence/tau_transfer/lognormal_fg_pitks_d3protocol.json

evidence/theory/e1.npy

evidence/theory/e2.npy

evidence/theory/object_mask.npy

evidence/theory/prop_empirical.json

evidence/theory/ratio.npy

evidence/theory/sigma.npy

evidence/theory/sp_bg.npy

evidence/theory/sp_full.npy

evidence/theory/sp_obj.npy

evidence/theory/toy_results.json

evidence/threeway/abdomen_ensemble.json

evidence/threeway/abdomen_perturbation.json

evidence/threeway/abdomen_variational.json

evidence/threeway/chest_ensemble.json

evidence/threeway/chest_perturbation.json

evidence/threeway/chest_variational.json

evidence/threeway/devset_recompute_finalgrid.json

evidence/threeway/head_ensemble.json

evidence/threeway/head_perturbation.json

evidence/threeway/head_variational.json

tools/gen_appendix_tables.py

tools/make_figE.py

tools/make_figF.py

tools/make_figG.py

tools/p4_table_to_latex.py

(53 additional files not shown) You must enabled JavaScript to view entire file list.

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.