Pith. sign in

REVIEW 1 major objections 3 minor 28 references

Underwater360: Reconstructing Underwater Scenes from Panoramic Images with Omnidirectional Gaussian Splatting

T0 review · 1 major / 3 minor · reviewed 2026-06-29 · grok-4.3

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

Pith's one-line read Underwater360 casts rays directly in spherical camera space and uses pose-conditioned embeddings to separate scene radiance from water absorption and scattering.

desk verdict Underwater360 adds spherical ray casting and pose-conditioned embeddings to 3DGS for panoramic underwater scenes plus a new benchmark, but the abstract gives no equations or ablations to confirm the claimed physical decoupling actually works. read the letter →

arxiv 2605.26447 v1 pith:HDUWC7TL submitted 2026-05-26 cs.CV

Jiangbei Hu , Weichao Song , Shibo Yu , Mohan Wang , Zihan Yi , Rui Wu , Mingkang Xiang , Na Lei
show 3 more authors
Shengfa Wang Zhongxuan Luo Ying He
This is my paper
classification cs.CV
keywords underwaterscenereconstructionomnidirectionalgaussiansplattingpanoramicimagingnovelviewsynthesisparticipatingmediaappearancemodelingsphericalraycasting
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

The paper introduces a method for reconstructing 360-degree underwater scenes from panoramic images by adapting 3D Gaussian Splatting to handle spherical distortions and underwater light effects. It performs ray casting in spherical space rather than on flattened images to avoid geometric errors, and models appearance with embeddings that tie to camera pose so that depth-dependent backscatter and attenuation can be isolated from the actual scene colors. A new benchmark dataset of synthetic and real panoramic underwater scenes is provided to test the approach. If the separation works, novel views can be rendered with consistent appearance across different angles without post-processing corrections for the medium.

What carries the argument

Omnidirectional Gaussian Splatting module that performs ray casting directly in spherical camera space together with a physics-based appearance-medium modeling architecture using pose-conditioned appearance embeddings.

What would settle it

A side-by-side comparison of rendered novel views against held-out real underwater panoramic photographs where the water attenuation and backscatter coefficients are independently measured, checking whether the method's restored scene colors match the measured ground-truth radiance at corresponding depths.

Watch

Extended reading notes

Core claim

The central claim is that an Omnidirectional Gaussian Splatting module combined with physics-based appearance-medium modeling using pose-conditioned embeddings enables explicit decoupling of intrinsic scene radiance from depth-dependent backscatter and attenuation, producing higher-quality novel view synthesis and appearance restoration than standard 3D Gaussian Splatting when applied to panoramic underwater imagery.

Load-bearing premise

That casting rays in spherical space and conditioning appearance embeddings on pose will cleanly separate scene light from water effects without creating new unmodeled distortions or needing extra corrections.

Editorial extensions

If this is right

  • Novel views of underwater scenes maintain consistent appearance across wide baselines without visible medium artifacts.
  • Panoramic images can be used directly for reconstruction without first correcting for spherical projection distortion.
  • Scene appearance restoration becomes possible by removing depth-dependent water effects during rendering rather than as a separate step.
  • A released benchmark of synthetic and real panoramic underwater scenes allows direct comparison of future methods on the same data.

Reading between the lines

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

  • The same spherical ray-casting approach could be tested on other wide-FoV or non-pinhole camera models outside underwater settings.
  • If the pose-conditioned embeddings generalize, the framework might extend to dynamic scenes by adding time as an additional conditioning variable.
  • Real-time rendering speed could be measured on the released code to check suitability for underwater robotics or VR applications.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 3 minor

Summary. The paper proposes Underwater360, a physics-informed omnidirectional 3D Gaussian Splatting framework for underwater panoramic scene reconstruction. It introduces an Omnidirectional Gaussian Splatting module that performs ray casting directly in spherical camera space to reduce geometric distortions under 360° FoV, a physics-based appearance-medium modeling architecture with pose-conditioned appearance embeddings to explicitly decouple intrinsic scene radiance from depth-dependent backscatter and attenuation, and a new panoramic underwater benchmark dataset with synthetic and real-world scenes. Experiments claim superior performance in novel view synthesis and scene appearance restoration with improved rendering quality and cross-view consistency.

Significance. If the explicit physical decoupling holds without implicit compensation, the work would advance underwater reconstruction by addressing both spherical projection issues and participating-media effects in a unified framework, with value for applications in marine robotics and exploration. The public release of code and datasets strengthens reproducibility and enables future comparisons.

major comments (1)
  1. [Abstract and §3] Abstract and §3 (physics-based appearance-medium modeling): The central claim that pose-conditioned appearance embeddings 'explicitly decouple' intrinsic scene radiance from depth-dependent backscatter and attenuation is load-bearing. Conditioning solely on pose (rather than per-ray depth, explicit attenuation integrals, or medium coefficients along the spherical ray) risks the embeddings absorbing residual view-dependent errors. Provide ablations or per-ray visualizations demonstrating that the separation is physically consistent rather than post-hoc compensation, especially for novel trajectories or varying water properties.
minor comments (3)
  1. [§4] §4 (implementation details): Clarify the precise mathematical form of the pose-conditioned embeddings and their integration with the medium model (e.g., how backscatter and attenuation terms are parameterized and optimized).
  2. [Experiments] Experiments section: Report error bars, standard deviations across multiple runs, or statistical tests for the claimed improvements over baselines to support the superiority assertions.
  3. [Dataset] Dataset description: Provide more detailed statistics on the new benchmark (e.g., number of scenes, views per scene, water turbidity variations) to allow assessment of generalization.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our central claim regarding explicit decoupling in the appearance-medium modeling. We address the concern point-by-point below and commit to strengthening the supporting evidence in the revision.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (physics-based appearance-medium modeling): The central claim that pose-conditioned appearance embeddings 'explicitly decouple' intrinsic scene radiance from depth-dependent backscatter and attenuation is load-bearing. Conditioning solely on pose (rather than per-ray depth, explicit attenuation integrals, or medium coefficients along the spherical ray) risks the embeddings absorbing residual view-dependent errors. Provide ablations or per-ray visualizations demonstrating that the separation is physically consistent rather than post-hoc compensation, especially for novel trajectories or varying water properties.

    Authors: We appreciate the referee's scrutiny of this load-bearing claim. Our architecture separates the components by feeding the pose-conditioned embeddings only into the intrinsic radiance prediction while the backscatter and attenuation are computed via an explicit depth-dependent integral along each spherical ray, following the underwater image formation model. The pose conditioning is intended to capture residual view-specific factors (e.g., slight lighting variations) without directly modeling medium coefficients. Nevertheless, we acknowledge that additional evidence is needed to rule out implicit compensation. In the revised manuscript we will add (1) an ablation that removes pose conditioning entirely and measures degradation on novel trajectories, (2) per-ray contribution visualizations that isolate the medium term versus the radiance term, and (3) experiments on the synthetic portion of our benchmark with controlled variations in water properties. These additions will be placed in §4 and the supplementary material. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity detected in derivation chain

full rationale

The provided abstract and context describe a proposed Omnidirectional Gaussian Splatting framework with physics-based appearance-medium modeling using pose-conditioned embeddings to decouple radiance from attenuation and backscatter. No equations, fitted parameters presented as predictions, or self-citations are visible that would reduce any central claim to its own inputs by construction. The architecture is introduced as a novel contribution without load-bearing steps that equate outputs to definitions or prior self-references. The derivation chain is therefore self-contained against external benchmarks.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract-only access yields no information on free parameters, axioms, or invented entities; full paper required for ledger construction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Underwater360: Reconstructing Underwater Scenes from Panoramic Images with Omnidirectional Gaussian Splatting." pith.science (2026). https://pith.science/paper/HDUWC7TL

@misc{pith2026260526447,
  author       = {Pith},
  title        = {Pith review of: Underwater360: Reconstructing Underwater Scenes from Panoramic Images with Omnidirectional Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HDUWC7TL}},
  note         = {Machine review of arXiv:2605.26447}
}
abstract

Underwater scene reconstruction is essential for immersive exploration of aquatic environments, yet remains challenging due to complex participating-media effects such as absorption and scattering, as well as the limited field of view (FoV) of conventional cameras. Although combining panoramic imaging with 3D Gaussian Splatting (3DGS) offers a promising direction for photorealistic underwater rendering, traditional 3DGS struggles with both spherical projection distortion and underwater medium degradation. In this paper, we propose \textbf{Underwater360}, a physics-informed omnidirectional 3DGS framework for underwater panoramic scene reconstruction. First, we introduce an Omnidirectional Gaussian Splatting module that performs ray casting directly in spherical camera space instead of relying on 2D projection approximations, thereby reducing geometric distortions under 360$^\circ$ FoV. Second, we design a physics-based appearance-medium modeling architecture with pose-conditioned appearance embeddings to explicitly decouple intrinsic scene radiance from depth-dependent backscatter and attenuation, enabling physically grounded scene appearance restoration. Finally, we establish a new panoramic underwater benchmark dataset containing both synthetic and real-world scenes. Extensive experiments demonstrate that Underwater360 achieves superior performance in underwater novel view synthesis and scene appearance restoration, delivering improved rendering quality and cross-view consistency in complex underwater environments. The code and datasets are released at https://github.com/SwcK423/Underwater360

Figures

Figures reproduced from arXiv: 2605.26447 by the authors.

Figure 1
Figure 1. Underwater reconstruction from perspective views is particularly challenging due to limited field of view, severe scattering, and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Challenges of conventional underwater capture with [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of Underwater360. Our framework combines omnidirectional Gaussian Splatting with underwater image formation modeling for panoramic underwater scene reconstruction. Starting from multi-view panoramas, we estimate camera parameters and initial￾ize 3D Gaussians. A pose-conditioned appearance module adjusts the Gaussians base color using view-dependent embeddings, producing corrected intrinsic appearance for re… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Illustration of Omnidirectional imaging model under [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison on both synthetic and real underwater scenes. The boxed regions are enlarged for detailed comparison. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison of UIFM decomposition on the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: More qualitative comparisons on both synthetic and real underwater scenes. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: More qualitative comparison of UIFM decomposition results. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references

  1. [1]

    Agisoft metashape standard, 2026

    Agisoft LLC. Agisoft metashape standard, 2026. 7

    2026

  2. [2]

    A revised underwater image formation model

    Derya Akkaynak and Tali Treibitz. A revised underwater image formation model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6723–6732, 2018. 2, 4

    2018

  3. [3]

    Sea-thru: A method for removing water from underwater images

    Derya Akkaynak and Tali Treibitz. Sea-thru: A method for removing water from underwater images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1682–1691, 2019. 2

    2019

  4. [4]

    Derya Akkaynak, Tali Treibitz, Tom Shlesinger, Yossi Loya, Raz Tamir, and David Iluz. What is the space of attenuation coefficients in underwater computer vision? InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 568–577, 2017. 1, 2

    2017

  5. [5]

    360-gs: Layout-guided panoramic gaus- sian splatting for indoor roaming

    Jiayang Bai, Letian Huang, Jie Guo, Wen Gong, Yuanqi Li, and Yanwen Guo. 360-gs: Layout-guided panoramic gaus- sian splatting for indoor roaming. InInternational Confer- ence on 3D Vision (3DV), pages 1042–1053, 2025. 2, 3, 7

    2025

  6. [6]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Zip-nerf: Anti-aliased grid-based neural radiance fields. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 19640–19648, 2023. 7

    2023

  7. [7]

    Bal- anced spherical grid for egocentric view synthesis

    Changwoon Choi, Sang Min Kim, and Young Min Kim. Bal- anced spherical grid for egocentric view synthesis. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12663–12673, 2023. 2, 3, 7

    2023

  8. [8]

    2d gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. InACM SIGGRAPH 2024 Conference Papers, 2024. 7

    2024

Show all 28 references
  1. [9]

    Real-time omnidirectional roaming in large scale in- door scenes

    Huajian Huang, Yingshu Chen, Tianjia Zhang, and Sai-Kit Yeung. Real-time omnidirectional roaming in large scale in- door scenes. InSIGGRAPH Asia 2022 Technical Communi- cations, New York, NY , USA, 2022. Association for Com- puting Machinery. 3

    2022

  2. [10]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions On Graphics (TOG), 42(4), 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkuehler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions On Graphics (TOG), 42(4), 2023. 1, 3, 7

    2023

  3. [11]

    Odgs: 3d scene reconstruction from om- nidirectional images with 3d gaussian splattings

    Suyoung Lee, Jaeyoung Chung, Jaeyoo Huh, and Ky- oung Mu Lee. Odgs: 3d scene reconstruction from om- nidirectional images with 3d gaussian splattings. InAd- vances in Neural Information Processing Systems (NIPS), pages 57050–57075. Curran Associates, Inc., 2024. 2, 3, 7

    2024

  4. [12]

    Seathru- nerf: Neural radiance fields in scattering media

    Deborah Levy, Amit Peleg, Naama Pearl, Dan Rosenbaum, Derya Akkaynak, Simon Korman, and Tali Treibitz. Seathru- nerf: Neural radiance fields in scattering media. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 56–65, 2023. 1, 2, 7

    2023

  5. [13]

    Watersplatting: Fast underwater 3d scene reconstruction using gaussian splatting

    Huapeng Li, Wenxuan Song, Tianao Xu, Alexandre El- sig, and Jonas Kulhanek. Watersplatting: Fast underwater 3d scene reconstruction using gaussian splatting. InInter- national Conference on 3D Vision (3DV), pages 969–978,

  6. [14]

    J. Li, F. Hahlbohm, T. Scholz, M. Eisemann, J.P. Tauscher, and M. Magnor. Spags: Fast and accurate 3d gaussian splat- ting for spherical panoramas.Computer Graphics Forum (CGF), 44(4):e70171, 2025. 2, 3, 5

    2025

  7. [15]

    Omnigs: Fast radiance field reconstruction using omnidirec- tional gaussian splatting

    Longwei Li, Huajian Huang, Sai-Kit Yeung, and Hui Cheng. Omnigs: Fast radiance field reconstruction using omnidirec- tional gaussian splatting. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 2260–2268, 2025. 2, 3, 7

    2025

  8. [16]

    Aquatic-gs: A hybrid 3d representation for under- water scenes, 2025

    Shaohua Liu, Junzhe Lu, Zuoya Gu, Jiajun Li, and Yue Deng. Aquatic-gs: A hybrid 3d representation for under- water scenes, 2025. 3

    2025

  9. [17]

    Ricardo Martin-Brualla, Noha Radwan, Mehdi S. M. Sajjadi, Jonathan T. Barron, Alexey Dosovitskiy, and Daniel Duck- worth. Nerf in the wild: Neural radiance fields for uncon- strained photo collections. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Rec...

    2021

  10. [18]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 1

    2021

  11. [19]

    Openmvg: Open multiple view geometry

    Pierre Moulon, Pascal Monasse, Romuald Perrot, and Re- naud Marlet. Openmvg: Open multiple view geometry. In Reproducible Research in Pattern Recognition, pages 60–74. Springer International Publishing, 2017. 3

    2017

  12. [20]

    Gaussian splashing: Direct volumetric rendering under- water, 2025

    Nir Mualem, Roy Amoyal, Oren Freifeld, and Derya Akkay- nak. Gaussian splashing: Direct volumetric rendering under- water, 2025. 1, 7

    2025

  13. [21]

    Instant neural graphics primitives with a mul- tiresolution hash encoding.ACM Transactions On Graphics (TOG), 41(4), 2022

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a mul- tiresolution hash encoding.ACM Transactions On Graphics (TOG), 41(4), 2022. 7

    2022

  14. [22]

    Sch ¨onberger and Jan-Michael Frahm

    Johannes L. Sch ¨onberger and Jan-Michael Frahm. Structure- from-motion revisited. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4104–4113, 2016. 1

    2016

  15. [23]

    Uw-gs: Distractor-aware 3d gaussian splatting for enhanced underwater scene reconstruction

    Haoran Wang, Nantheera Anantrasirichai, Fan Zhang, and David Bull. Uw-gs: Distractor-aware 3d gaussian splatting for enhanced underwater scene reconstruction. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3280–3289, 2025. 1, 3, 7

    2025

  16. [24]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing (TIP), 13(4):600–612, 2004. 7

    2004

  17. [25]

    Leonard, and Yogesh Girdhar

    Daniel Yang, John J. Leonard, and Yogesh Girdhar. Sea- splat: Representing underwater scenes with 3d gaussian splatting and a physically grounded image formation model. InIEEE International Conference on Robotics and Automa- tion (ICRA), pages 7632–7638, 2025. 1, 3, 7

    2025

  18. [26]

    Mip-splatting: Alias-free 3d gaussian splat- ting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splat- ting. InProceedings of the IEEE/CVF Conference on Com- 9 puter Vision and Pattern Recognition (CVPR), pages 19447– 19456, 2024. 6

    2024

  19. [27]

    3d-uir: 3d gaussian for underwater 3d scene reconstruction via physics based appearance-medium decoupling, 2025

    Jieyu Yuan, Yujun Li, Yuanlin Zhang, Chunle Guo, Xiongxin Tang, Ruixing Wang, and Chongyi Li. 3d-uir: 3d gaussian for underwater 3d scene reconstruction via physics based appearance-medium decoupling, 2025. 1, 3, 7

    2025

  20. [28]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 586–595, 2018. 7 10 buildi...

    2018

Pith tools

Reviewed June 29, 2026 · model on record in the stance chip above.