Pith. sign in

REVIEW 2 major objections 4 minor 64 references

Predictive Photometric Uncertainty in Gaussian Splatting for Novel View Synthesis

T0 review · 2 major / 4 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A post-hoc least-squares fit of residual error to Gaussian primitives yields pixel uncertainty maps that turn 3D Gaussian Splatting into a trustworthy map for robots and safety-critical vision.

desk verdict Solid post-hoc residual-to-uncertainty layer for 3DGS that actually moves three downstream tasks without touching fidelity. read the letter →

arxiv 2603.22786 v2 pith:HB2LBCLI submitted 2026-03-24 cs.CV

Chamuditha Jayanga Galappaththige , Thomas Gottwald , Peter Stehr , Edgar Heinert , Niko Suenderhauf , Dimity Miller , Matthias Rottmann This is my paper · ORCID
classification cs.CV
keywords 3DGaussianSplattingnovelviewsynthesisuncertaintyestimationpredictiveactiveselectionscenechangedetectionanomalypost-hocmethod
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

3D Gaussian Splatting produces photorealistic novel views, but agents and safety systems also need to know where those views are unreliable. This paper shows that the same photometric residuals used to train the scene can be re-used, after training, as a linear least-squares target for a new per-primitive uncertainty channel. A Bayesian-style regularizer pulls uncertainty toward a high default wherever a primitive is poorly observed, so the method still works under sparse capture. The resulting view-dependent uncertainty maps can be rendered as cheaply as RGB, leave the original geometry and fidelity untouched, and measurably raise performance on next-best-view planning, pose-agnostic change detection, and pose-agnostic anomaly detection. The practical claim is that residual-driven uncertainty is enough to convert an already-trained Gaussian scene into a reliability-aware spatial map.

What carries the argument

Bayesian-regularized linear least-squares over reconstruction residuals: the matrix of alpha-blending weights times a vector of per-primitive spherical-harmonic uncertainty coefficients is fitted to the pixel residual vector, with an L2 prior that defaults unobserved directions to maximal uncertainty.

What would settle it

On a sparse-view Mip-NeRF360 split, measure whether the residual-fitted uncertainty maps still produce lower AUSE and higher Pearson correlation with DSSIM error on the held-out novel views than FisherRF; if they do not, or if the downstream gains in active view selection and change/anomaly detection disappear, the residual-proxy claim fails.

Watch

Extended reading notes

Core claim

Training-view reconstruction residuals (L1 + DSSIM) can be explained by a linear least-squares assignment of directional uncertainty values to each Gaussian primitive; with a Bayesian L2 pull toward maximal uncertainty in unobserved directions, the resulting alpha-blended maps align with true novel-view error better than prior post-hoc or stochastic uncertainty methods, while preserving the original rendering quality of any frozen 3DGS model.

Load-bearing premise

That the photometric residuals left on the training views, once assigned to primitives by linear least-squares and regularized toward high uncertainty where views are missing, remain a faithful proxy for true photometric error on held-out and highly novel viewpoints.

Editorial extensions

If this is right

  • Any already-trained 3DGS model can receive pixel-wise reliability maps without re-training or fidelity loss.
  • Next-best-view planners can simply pick the candidate with highest total rendered uncertainty and improve reconstruction under fixed budgets.
  • Pose-agnostic change and anomaly detectors can mask high-uncertainty pixels and reduce false positives caused by rendering artifacts.
  • The same residual-to-uncertainty pipeline applies unchanged to depth-regularized, densification-improved, and sparse-primitive 3DGS variants.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If residual assignment works for photometric error, the same least-squares construction could be applied to depth or normal residuals to separate geometric from appearance uncertainty.
  • Safety-critical pipelines that already store 3DGS maps could treat the uncertainty channel as a soft occupancy or confidence layer for planning without extra sensors.
  • The method's success under sparse capture suggests residual-based uncertainty may be more data-efficient than sampling-based ensembles for real-time robotic mapping.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper proposes a post-hoc, architecture-agnostic method for pixel-wise, view-dependent predictive photometric uncertainty in 3D Gaussian Splatting. After freezing a trained 3DGS model, it learns per-primitive SH uncertainty channels uk(d) by solving a Bayesian-regularized linear least-squares problem that fits the fixed alpha-blending matrix A to training-view residuals Lx = (1-λ)L1 + λ DSSIM (Eqs. 3–5, 7). The resulting maps are rendered exactly like RGB (Eq. 2). On Mip-NeRF360, Tanks & Temples and Deep Blending the method reports substantially better AUSE/Pearson alignment with novel-view error than FisherRF, Manifold and Var3DGS while preserving baseline fidelity and adding only ~13 % training overhead (Table 1). Ablations confirm the value of SH degree 3 and of the L2 prior toward maximal uncertainty b=1 under sparse highly-novel views (Tables 2a, Sec. 4.2). The same maps improve active view selection (Table 2b), pose-agnostic scene change detection (Table 3) and pose-agnostic anomaly detection (Table 4).

Significance. If the residual-to-novel-error transfer holds, the work supplies a genuinely modular reliability layer for the rapidly expanding 3DGS ecosystem. Because the method never alters geometry, appearance or the original optimization, it can be dropped onto any existing splat without fidelity loss—an advantage over stochastic or Hessian-based alternatives that either degrade rendering quality or remain parameter-centric. Demonstrated gains on three distinct downstream perception tasks (AVS, SCD, AD) move 3DGS from a pure rendering engine toward a trustworthy spatial map for autonomous agents. The formulation is simple, the empirical margins large, and the plug-and-play claim is supported by Appendix B results on multiple 3DGS variants.

major comments (2)
  1. The central modeling hypothesis (Sec. 3.2) that multi-view linear assignment of training residuals Lx to primitives yields a reliable proxy for predictive photometric error on held-out and highly novel views is only partially stress-tested. Table 1 and the four-view sparse experiment (Sec. 4.2, Fig. 5) show strong average gains, yet no failure-mode analysis is given for residuals driven by view-dependent lighting, specularities or overfitting that do not transfer under the same α-blending. A short controlled study (e.g., synthetic lighting change or deliberate overfit) would make the claim more robust.
  2. Tables 1–4 report point estimates without error bars, multi-seed statistics or scene-level variance. Given that AUSE/Pearson and the downstream F1/mIoU/AUROC lifts are the quantitative backbone of the SOTA claims, at least seed-averaged results (or bootstrap intervals) on the three main datasets are needed to confirm that the large margins are stable.
minor comments (4)
  1. Appendix C and Figs. 8–9 correctly note that Manifold/Var3DGS produce slightly different renderings; a brief quantitative statement of how much this affects AUSE comparability would strengthen the baseline discussion.
  2. The choice of residual mix λ=0.2 is inherited from 3DGS without ablation; a one-line sensitivity check would be useful.
  3. Notation for the expanded SH matrix A (Appendix A) is clear, but the main text could briefly remind the reader that direction d(xj,Gk) remains fixed once geometry is frozen, preserving linearity.
  4. Fig. 1 caption and the qualitative panels would benefit from a common color-scale bar so that uncertainty magnitudes can be compared across methods.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: post-hoc least-squares fit of uncertainty channels to training residuals is ordinary supervised calibration, evaluated on independent novel-view error and separate downstream protocols.

full rationale

The derivation chain is self-contained and non-circular. Uncertainty channels uk (SH coefficients) are obtained by solving the linear least-squares problem arg min_u ||y - A u||_2^2 (Eqs. 3-5), where y collects training-view photometric residuals Lx = (1-λ)L1 + λ DSSIM and A is the fixed α-blending matrix from the frozen 3DGS representation; a Bayesian-inspired L2 prior (Eqs. 6-7) merely pulls unobserved directions toward maximal uncertainty b=1. This is then rasterized to novel views and scored by AUSE/Pearson against true held-out error (Table 1) plus used as a post-hoc attenuator in independent AVS/SCD/AD pipelines (Tables 2b-4). None of these steps reduce by construction to the training residuals: novel-view error is an external target, the residual-to-error transfer is an empirical hypothesis (Sec. 3.2) that is tested rather than assumed true, and the prior is a regularizer not a fitted constant renamed as prediction. Mild self-citation of PRIMU (overlapping authors) and reuse of the authors' own SCD/AD baselines are disclosed and non-load-bearing; they supply neither uniqueness theorems nor the numerical UE metrics. The formulation therefore constitutes ordinary post-hoc residual modeling, not a circular derivation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 1 invented entities

The central claim rests on standard 3DGS rendering, the modeling choice that residuals proxy predictive uncertainty, linearity of uncertainty under fixed alpha-blending, a few hand-set scalars (b, λ_reg, SH degree, photometric λ), and the invented per-primitive uncertainty channel. No new physical entities; the free parameters are optimization/prior knobs ablated in Sec. 4.2 and Tab. 2a.

free parameters (5)
  • λ_reg (Bayesian L2 regularization weight)
    Controls pull of directional uncertainty toward maximal prior; ablated from 0 to 20.48 in sparse-view experiments; optima differ for L1 vs DSSIM.
  • b (maximal isotropic uncertainty level)
    Set to 1 as the upper bound of normalized photometric residual; centers the Gaussian prior for unobserved directions (Sec. 3.4, Sec. 4.2).
  • SH degree for uk(d)
    Chosen as 3 after ablation (Tab. 2a); capacity of view-dependent uncertainty.
  • photometric residual mix λ
    Fixed at 0.2 following 3DGS for Lx = (1-λ)L1 + λ LDSSIM; defines the regression target y.
  • uncertainty training iteration budgets
    e.g. 400 iters sparse UE, 50×Nviews before each AVS selection (I=50 chosen after Tab. 7); affect fit quality and ranking stability.
assumptions (4)
  • ad hoc to paper Training-view photometric residuals, after multi-view linear assignment to primitives, generalize as predictive uncertainty on novel views.
    Stated hypothesis in Sec. 3.2; load-bearing for all UE and downstream claims; not derived from first principles.
  • domain assumption Uncertainty may be alpha-blended with the same fixed opacities and transmittance as color (Eq. 2 mirrors Eq. 1).
    Enables the linear map A and post-hoc plug-in; standard rasterization assumption but not proven optimal for uncertainty.
  • standard math L2 regularization of SH uncertainty toward constant b is Bayesian linear regression under a Gaussian prior centered at maximal uncertainty.
    Invokes Bishop-style interpretation (Sec. 3.4, ref. [6]); mathematically standard once the prior mean b is chosen.
  • domain assumption Frozen 3DGS geometry and appearance remain a valid scene map; only uncertainty channels are free.
    Post-hoc premise throughout Sec. 3; fidelity preservation follows by construction.
invented entities (1)
  • Per-primitive view-dependent uncertainty channel uk(d) (SH coefficients)
    purpose: Stores residual-explained reliability per Gaussian so uncertainty can be rasterized like color to any pose.
    New representation attribute introduced by the method; independent_evidence is empirical (correlation with novel-view error and task gains), not an external physical prediction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predictive Photometric Uncertainty in Gaussian Splatting for Novel View Synthesis." pith.science (2026). https://pith.science/paper/HB2LBCLI

@misc{pith2026260322786,
  author       = {Pith},
  title        = {Pith review of: Predictive Photometric Uncertainty in Gaussian Splatting for Novel View Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HB2LBCLI}},
  note         = {Machine review of arXiv:2603.22786}
}
read the original abstract

Recent advances in 3D Gaussian Splatting have enabled impressive photorealistic novel view synthesis. However, to transition from a pure rendering engine to a reliable spatial map for autonomous agents and safety-critical applications, knowing where the representation is uncertain is as important as the rendering fidelity itself. We bridge this critical gap by introducing a lightweight, plug-and-play framework for pixel-wise, view-dependent predictive uncertainty estimation. Our post-hoc method formulates uncertainty as a Bayesian-regularized linear least-squares optimization over reconstruction residuals. This architecture-agnostic approach extracts a per-primitive uncertainty channel without modifying the underlying scene representation or degrading baseline visual fidelity. Crucially, we demonstrate that providing this actionable reliability signal successfully translates 3D Gaussian splatting into a trustworthy spatial map, further improving state-of-the-art performance across three critical downstream perception tasks: active view selection, pose-agnostic scene change detection, and pose-agnostic anomaly detection.

Figures

Figures reproduced from arXiv: 2603.22786 by the authors.

Figure 1
Figure 1. Per-pixel UE for NVS. Our post-hoc method generates a view-dependent un￾certainty map that closely mirrors regions of error within the RGB render. efficiency of rasterization. While recent advancements have drastically improved 3DGS visual fidelity [25,61], geometric consistency [9,24], and efficiency [16,38], equipping these models with robust, system-level uncertainty estimation (UE) remains a critical, underexplo… view at source ↗
Figure 2
Figure 2. Visual example of reconstruction residual on training views. (Left) ground truth, (Center) rendering, (Right) residual map. The residual maps capture errors caused by both imperfect geometry (e.g. under-reconstructed regions in vegetation) and the limited capacity of view-dependent appearance (e.g. reflections on the table) [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of our approach: We learn a primitive-level uncertainty channel uk on a well-trained 3DGS, leveraging a (Bayesian-inspired) regularized linear least-squares formulation of training residuals. The learned view-dependent uncertainty can be ras￾terized to any given (novel) viewpoint Pj . Our method is post-hoc and therefore can be seamlessly integrated with any 3DGS variant without loss of fidelity. camera wit… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of predicted uncertainty maps. Our UE more closely reflects the true error when compared to the baselines. 4 Experiments Datasets: We follow the exact evaluation protocol established in 3DGS [23] and assess our UE performance on three standard be…
Figure 5
Figure 5. Figure 5: a. Due to the sparse coverage of the scene, most test views observe the scene from highly novel viewing directions. We evaluate on Mip-NeRF360 [4], training the base 3DGS model for 4,000 it￾erations and our uncertainty channel for 400 iterations to account for the spar…
Figure 6
Figure 6. Figure 6: Qualitative examples of uncertainty maps with Bayesian-inspired regularization for different weights λreg (see Eq. (7)) in the setting of highly novel views. The given AUSE values are with respect to DSSIM. port PSNR, SSIM, and LPIPS on the holdout views after reconstr…
Figure 7
Figure 7. Figure 7: Qualitative examples of uncertainty-guided scene change detection. Our pre￾dicted uncertainty maps effectively capture rendering artifacts, allowing the system to suppress false positives caused by these artifacts rather than actual scene changes. a 3D change represent…
Figure 8
Figure 8. Figure 8: Qualitative comparison of predicted uncertainty maps to DSSIM error map. Including separate renderings and error maps for the stochastic 3DGS methods Mani￾fold [37] and Var3DGS [33]. Top: bicycle from Mip-NeRF360 [4]. Bottom: counter from Mip-NeRF360 [4] [PITH_FULL_IM…
Figure 9
Figure 9. Figure 9: Qualitative comparison of predicted uncertainty maps to DSSIM error map. Including separate renderings and error maps for the stochastic 3DGS methods Man￾ifold [37] and Var3DGS [33]. Top: room from Mip-NeRF360 [4]. Bottom: truck from Tanks&Temples [28] [PITH_FULL_IMAG…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 1 linked inside Pith

  1. [1]

    IEEE Transactions on Neural Networks and Learning Systems36(6), 11657–11663 (2025) 2, 3

    Aira, L.S., Valsesia, D., Magli, E.: Modeling uncertainty for gaussian splatting. IEEE Transactions on Neural Networks and Learning Systems36(6), 11657–11663 (2025) 2, 3

    2025

  2. [2]

    Autonomous Robots42(7), 1301–1322 (2018) 4

    Alcantarilla, P.F., Stent, S., Ros, G., Arroyo, R., Gherardi, R.: Street-view change detection with deconvolutional networks. Autonomous Robots42(7), 1301–1322 (2018) 4

    2018

  3. [3]

    Alpherts, T., Ghebreab, S., van Noord, N.: Emplace: Self-supervised urban scene changedetection.In:ProceedingsoftheAAAIConferenceonArtificialIntelligence. vol. 39 (2), pp. 1737–1745 (2025) 5

    2025

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Mip- nerf 360: Unbounded anti-aliased neural radiance fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5470–5479 (2022) 9, 10, 11, 2, 5, 6

    2022

  5. [5]

    2019 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) pp

    Bergmann, P., Fauser, M., Sattlegger, D., Steger, C.: Mvtec ad — a comprehensive real-world dataset for unsupervised anomaly detection. 2019 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) pp. 9584–9592 (2019) 5 8 Galappaththige et al

    2019

  6. [6]

    Bishop, C.M., Nasrabadi, N.M.: Pattern recognition and machine learning, vol. 4. Springer (2006) 8

    2006

  7. [7]

    In: 2021 IEEE Intelligent Vehicles Symposium (IV)

    Chen, S., Yang, K., Stiefelhagen, R.: Dr-tanet: Dynamic receptive temporal atten- tion network for street scene change detection. In: 2021 IEEE Intelligent Vehicles Symposium (IV). pp. 502–509. IEEE (2021) 4

    2021

  8. [8]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Cho, K., Kim, D.Y., Kim, E.: Zero-shot scene change detection. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39 (3), pp. 2509–2517 (2025) 5

    2025

Show all 64 references
  1. [9]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Chung, J., Oh, J., Lee, K.M.: Depth-regularized optimization for 3d gaussian splat- ting in few-shot images. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 811–820 (2024) 2

    2024

  2. [10]

    In: 2018 25th IEEE international conference on image processing (ICIP)

    Daudt, R.C., Le Saux, B., Boulch, A.: Fully convolutional siamese networks for change detection. In: 2018 25th IEEE international conference on image processing (ICIP). pp. 4063–4067. IEEE (2018) 4

    2018

  3. [11]

    IEEE Transactions on Visualization and Computer Graphics (2024) 1

    Fei, B., Xu, J., Zhang, R., Zhou, Q., Yang, W., He, Y.: 3d gaussian splatting as new era: A survey. IEEE Transactions on Visualization and Computer Graphics (2024) 1

    2024

  4. [12]

    arXiv preprint arXiv:2511.12370 (2025) 5, 13, 14

    Galappaththige, C.J., Lai, J., Windrim, L., Dansereau, D., Sünderhauf, N., Miller, D.: Changes in real time: Online scene change detection with multi-view fusion. arXiv preprint arXiv:2511.12370 (2025) 5, 13, 14

    2025

  5. [13]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Galappaththige, C.J., Lai, J., Windrim, L., Dansereau, D., Sunderhauf, N., Miller, D.: Multi-view pose-agnostic change localization with zero labels. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 11600–11610 (2025) 2, 5, 12, 13, 14

    2025

  6. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Goli, L., Reading, C., Sellán, S., Jacobson, A., Tagliasacchi, A.: Bayes’ rays: Uncer- tainty quantification for neural radiance fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20061–20070 (2024) 3

    2024

  7. [15]

    arXiv preprint arXiv:2508.02443 (2025) 4

    Gottwald, T., Heinert, E., Stehr, P., Galappaththige, C.J., Rottmann, M.: Primu: Uncertainty estimation for novel views in gaussian splatting from primitive-based representations of error and coverage. arXiv preprint arXiv:2508.02443 (2025) 4

    2025

  8. [16]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Hanson, A., Tu, A., Lin, G., Singla, V., Zwicker, M., Goldstein, T.: Speedy-splat: Fast 3d gaussian splatting with sparse pixels and sparse primitives. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 21537–21546 (2025) 2, 4

    2025

  9. [17]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)

    Hanson, A., Tu, A., Singla, V., Jayawardhana, M., Zwicker, M., Goldstein, T.: PUP 3D-GS: Principled uncertainty pruning for 3d gaussian splatting. In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR). pp. 5949–5958 (2025) 4

    2025

  10. [18]

    ACM Transactions on Graphics (Proc

    Hedman, P., Philip, J., Price, T., Frahm, J.M., Drettakis, G., Brostow, G.: Deep blending for free-viewpoint image-based rendering. ACM Transactions on Graphics (Proc. SIGGRAPH Asia)37(6), 257:1–257:15 (2018) 9, 10

    2018

  11. [19]

    In: Proceedings of the European Conference on Computer Vision (ECCV)

    Ilg, E., Cicek, O., Galesso, S., Klein, A., Makansi, O., Hutter, F., Brox, T.: Uncer- tainty estimates and multi-hypotheses networks for optical flow. In: Proceedings of the European Conference on Computer Vision (ECCV). pp. 652–667 (2018) 9

    2018

  12. [20]

    In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    Jiang, B., Huang, R., Zhao, Q., Zhang, Y.: Gaussian difference: Find any change instance in 3d scenes. In: ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 1–5. IEEE (2025) 5

    2025

  13. [21]

    In: European Conference on Computer Vision

    Jiang, W., Lei, B., Daniilidis, K.: Fisherrf: Active view selection and mapping with radiance fields using fisher information. In: European Conference on Computer Vision. pp. 422–440. Springer (2024) 2, 4, 9, 10, 11, 12, 3 3DGS-U 9

    2024

  14. [22]

    In: 2025 IEEE International Conference on Robotics and Automation (ICRA)

    Kannan, S.S., Min, B.C.: Zeroscd: Zero-shot street scene change detection. In: 2025 IEEE International Conference on Robotics and Automation (ICRA). pp. 4665–4671 (2025) 5

    2025

  15. [23]

    ACM Transactions on Graphics (TOG)42(4), 1–14 (2023) 1, 3, 5, 7, 8, 9, 10, 12, 2

    Kerbl, B., Kopanas, G., Leimkuehler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics (TOG)42(4), 1–14 (2023) 1, 3, 5, 7, 8, 9, 10, 12, 2

    2023

  16. [24]

    ACM Transactions on Graphics (TOG)43(4), 1–15 (2024) 2

    Kerbl, B., Meuleman, A., Kopanas, G., Wimmer, M., Lanvin, A., Drettakis, G.: A hierarchical 3d gaussian representation for real-time rendering of very large datasets. ACM Transactions on Graphics (TOG)43(4), 1–15 (2024) 2

    2024

  17. [25]

    Advances in Neural Information Processing Systems37, 80965–80986 (2024) 2

    Kheradmand, S., Rebain, D., Sharma, G., Sun, W., Tseng, Y.C., Isack, H., Kar, A., Tagliasacchi, A., Yi, K.M.: 3d gaussian splatting as markov chain monte carlo. Advances in Neural Information Processing Systems37, 80965–80986 (2024) 2

    2024

  18. [26]

    In: Proceed- ings of the Computer Vision and Pattern Recognition Conference

    Kim, J.W., Kim, U.H.: Towards generalizable scene change detection. In: Proceed- ings of the Computer Vision and Pattern Recognition Conference. pp. 24463–24473 (2025) 5

    2025

  19. [27]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023) 5

    2023

  20. [28]

    ACM Transactions on Graphics36(4) (2017) 9, 10, 6

    Knapitsch, A., Park, J., Zhou, Q.Y., Koltun, V.: Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics36(4) (2017) 9, 10, 6

    2017

  21. [29]

    In: Guthe, M., Grosch, T

    Kopanas,G.,Drettakis,G.:ImprovingNeRFQualitybyProgressiveCameraPlace- ment for Free-Viewpoint Navigation. In: Guthe, M., Grosch, T. (eds.) Vision, Mod- eling, and Visualization. The Eurographics Association (2023) 4

    2023

  22. [30]

    Kowsky, W.: A quadrature formula over the sphere with application to high reso- lution spherical harmonic analysis. Bull. Gdod60, 1–14 (1986) 8

    1986

  23. [31]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Kruse, M., Rudolph, M., Woiwode, D., Rosenhahn, B.: Splatpose & detect: Pose- agnostic 3d anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3950–3960 (2024) 2, 5, 13, 14

    2024

  24. [32]

    IEEE Transactions on Image Processing 30, 55–67 (2020) 4

    Lei, Y., Peng, D., Zhang, P., Ke, Q., Li, H.: Hierarchical paired channel fusion network for street scene change detection. IEEE Transactions on Image Processing 30, 55–67 (2020) 4

    2020

  25. [33]

    Li, R., Cheung, Y.m.: Variational multi-scale representation for estimating uncer- taintyin3dgaussiansplatting.AdvancesinNeuralInformationProcessingSystems 37, 87934–87958 (2024) 2, 3, 9, 10, 11, 4, 5, 6

    2024

  26. [34]

    In: 2025 IEEE Inter- national Conference on Robotics and Automation (ICRA)

    Lin, C.J., Garg, S., Chin, T.J., Dayoub, F.: Robust scene change detection using visual foundation models and cross-attention mechanisms. In: 2025 IEEE Inter- national Conference on Robotics and Automation (ICRA). pp. 8337–8343. IEEE (2025) 4

    2025

  27. [35]

    In: European Conference on Computer Vision Workshops

    Liu, Y., Hu, Y.S., Chen, Y., Zelek, J.: Splatpose+: Real-time image-based pose- agnostic 3d anomaly detection. In: European Conference on Computer Vision Workshops. pp. 378–391. Springer (2024) 5, 13, 14

    2024

  28. [36]

    IEEE Robotics and Automation Letters (2025) 5

    Lu, Z., Ye, J., Leonard, J.: 3dgs-cd: 3d gaussian splatting-based change detection for physical object rearrangement. IEEE Robotics and Automation Letters (2025) 5

    2025

  29. [37]

    In: SIGGRAPH Asia 2024 Conference Papers

    Lyu, L., Tewari, A., Habermann, M., Saito, S., Zollhöfer, M., Leimkühler, T., Theobalt, C.: Manifold sampling for differentiable uncertainty in radiance fields. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–11 (2024) 3, 4, 9, 10, 11, 12, 5, 6 10 Galappaththige et al

    2024

  30. [38]

    In: SIGGRAPH Asia 2024 Conference Papers

    Mallick, S.S., Goel, R., Kerbl, B., Steinberger, M., Carrasco, F.V., De La Torre, F.: Taming 3dgs: High-quality radiance fields with limited resources. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–11 (2024) 2

    2024

  31. [39]

    In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition

    Matsuki, H., Murai, R., Kelly, P.H., Davison, A.J.: Gaussian splatting slam. In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition. pp. 18039–18048 (2024) 1

    2024

  32. [40]

    Commu- nications of the ACM65(1), 99–106 (2021) 1, 3, 5

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021) 1, 3, 5

    2021

  33. [41]

    Advances in Neural Information Processing Systems37, 99145–99160 (2024) 3

    Nakayama, K., Uy, M.A., You, Y., Li, K., Guibas, L.J.: Provnerf: Modeling per point provenance in nerfs as a stochastic field. Advances in Neural Information Processing Systems37, 99145–99160 (2024) 3

    2024

  34. [42]

    arXiv preprint arXiv:2304.07193 (2023) 5, 12, 13

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023) 5, 12, 13

    2023 arXiv

  35. [43]

    In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T

    Pan, X., Lai, Z., Song, S., Huang, G.: Activenerf: Learning where to see with uncertainty estimation. In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision – ECCV 2022. pp. 230–246. Springer Nature Switzerland, Cham (2022) 2, 4

    2022

  36. [44]

    In: Proceedings of the European conference on computer vision (ECCV)

    Rota Bulò, S., Porzi, L., Kontschieder, P.: Revising densification in gaussian splat- ting. In: Proceedings of the European conference on computer vision (ECCV). p. 347–362 (2024) 4

    2024

  37. [45]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp

    Roth, K., Pemula, L., Zepeda, J., Scholkopf, B., Brox, T., Gehler, P.: Towards total recall in industrial anomaly detection. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp. 14298–14308 (2021) 5

    2022

  38. [46]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Sachdeva, R., Zisserman, A.: The change you want to see. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 3993–4002 (2023) 4

    2023

  39. [47]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision Workshops

    Sachdeva, R., Zisserman, A.: The change you want to see (now in 3d). In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision Workshops. pp. 2060–2069 (2023) 4

  40. [48]

    In: Proceedings of the British Machine Vi- sion Conference (BMVC)

    Sakurada, K., Okatani, T.: Change detection from a street image pair using cnn features and superpixel segmentation. In: Proceedings of the British Machine Vi- sion Conference (BMVC). pp. 61.1–61.12 (2015) 4

    2015

  41. [49]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 4104–4113 (2016) 12

    2016

  42. [50]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Seo, S., Chang, Y., Kwak, N.: Flipnerf: Flipped reflection rays for few-shot novel view synthesis. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 22883–22893 (2023) 3

    2023

  43. [51]

    In: European Conference on Computer Vision

    Shen, J., Agudo, A., Moreno-Noguer, F., Ruiz, A.: Conditional-flow nerf: Accurate 3d modelling with reliable uncertainty quantification. In: European Conference on Computer Vision. pp. 540–557. Springer (2022) 3

    2022

  44. [52]

    In: 2021 International Con- ference on 3D Vision (3DV)

    Shen,J.,Ruiz,A.,Agudo,A.,Moreno-Noguer,F.:Stochasticneuralradiancefields: Quantifying uncertainty in implicit 3d representations. In: 2021 International Con- ference on 3D Vision (3DV). pp. 972–981. IEEE (2021) 2

    2021

  45. [53]

    The International journal of robotics research37(4-5), 405– 420 (2018) 1 3DGS-U 11

    Sünderhauf, N., Brock, O., Scheirer, W., Hadsell, R., Fox, D., Leitner, J., Upcroft, B., Abbeel, P., Burgard, W., Milford, M., et al.: The limits and potentials of deep learning for robotics. The International journal of robotics research37(4-5), 405– 420 (2018) 1 3DGS-U 11

    2018

  46. [54]

    In: 2023 IEEE International Conference on Robotics and Automation (ICRA)

    Sünderhauf, N., Abou-Chakra, J., Miller, D.: Density-aware nerf ensembles: Quan- tifying predictive uncertainty in neural radiance fields. In: 2023 IEEE International Conference on Robotics and Automation (ICRA). pp. 9370–9376 (2023) 3, 4

    2023

  47. [55]

    SIAM (2005) 1

    Tarantola, A.: Inverse problem theory and methods for model parameter estima- tion. SIAM (2005) 1

    2005

  48. [56]

    In: Proceedings of the European conference on computer vision (ECCV) workshops

    Varghese, A., Gubbi, J., Ramaswamy, A., Balamuralidhar, P.: Changenet: A deep learning architecture for visual change detection. In: Proceedings of the European conference on computer vision (ECCV) workshops. pp. 0–0 (2018) 4

    2018

  49. [57]

    IEEE transactions on image processing 13(4), 600–612 (2004) 12

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing 13(4), 600–612 (2004) 12

    2004

  50. [58]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)

    Wilson, J., Almeida, M., Mahajan, S., Labrie, M., Ghaffari, M., Ghasemalizadeh, O., Sun, M., Kuo, C.H., Sen, A.: POp-GS: Next Best View in 3D-Gaussian Splat- ting with P-Optimality. In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR). pp. 3646–3655...

    2025

  51. [59]

    Wilson, J., Almeida, M., Sun, M., Mahajan, S., Ghaffari, M., Ewen, P., Ghase- malizadeh, O., Kuo, C.H., Sen, A.: Modeling uncertainty in 3d gaussian splatting through continuous semantic splatting (2024) 3

    2024

  52. [60]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., Wang, X.: 4d gaussian splatting for real-time dynamic scene rendering. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 20310– 20320 (2024) 5

    2024

  53. [61]

    In: Proceedings of the 32nd ACM International Conference on Multimedia

    Ye, Z., Li, W., Liu, S., Qiao, P., Dou, Y.: Absgs: Recovering fine details in 3d gaussian splatting. In: Proceedings of the 32nd ACM International Conference on Multimedia. pp. 1053–1061 (2024) 2

    2024

  54. [62]

    Advances in Neural Information Processing Systems36, 44558–44571 (2023) 5, 14

    Zhou, Q., Li, W., Jiang, L., Wang, G., Zhou, G., Zhang, S., Zhao, H.: Pad: A dataset and benchmark for pose-agnostic anomaly detection. Advances in Neural Information Processing Systems36, 44558–44571 (2023) 5, 14

    2023

  55. [63]

    In: 2025 International Conference on 3D Vision (3DV)

    Zhu, L., Li, Y., Sandström, E., Huang, S., Schindler, K., Armeni, I.: Loopsplat: Loop closure by registering 3d gaussian splats. In: 2025 International Conference on 3D Vision (3DV). pp. 156–167. IEEE (2025) 1

    2025

  56. [64]

    In: Pro- ceedings Visualization, 2001

    Zwicker, M., Pfister, H., Van Baar, J., Gross, M.: Ewa volume splatting. In: Pro- ceedings Visualization, 2001. VIS’01. pp. 29–538. IEEE (2001) 6

    2001

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.