pith. sign in

arxiv: 2605.18263 · v1 · pith:3RXQ33W4new · submitted 2026-05-18 · 💻 cs.CV

RT-Splatting: Joint Reflection-Transmission Modeling with Gaussian Splatting

Ji Shi , Xianghua Ying , Bowei Xing , Ruohao Guo , Wenzhen Yue This is my paper

Pith reviewed 2026-05-20 12:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords Gaussian SplattingReflection and TransmissionSemi-transparent SurfacesNovel View SynthesisHybrid Surface-Volume RenderingReal-time Rendering
0
0 comments X

The pith

Factoring each Gaussian into separate geometric occupancy and optical opacity lets one primitive set render both sharp reflections and clear transmission.

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

The paper introduces RT-Splatting to handle semi-transparent surfaces that show both complex reflections and see-through transmission. It factors every Gaussian primitive into a geometric occupancy term and an optical opacity term. This split creates a single scene representation that a hybrid renderer can treat once as a surface for high-frequency reflections and once as a volume for unobstructed transmission. A Specular-Aware Gradient Gating step blocks misleading gradients from specular areas from corrupting the transmission optimization. The result is real-time novel-view synthesis with fewer floaters and naturally supports scene editing.

Core claim

Disentangling geometric occupancy from optical opacity inside each Gaussian produces a unified surface-volume representation that a hybrid renderer can interpret to capture high-frequency reflections while preserving clear transmission, with Specular-Aware Gradient Gating suppressing optimization conflicts that otherwise produce blurry reflections or occluded views.

What carries the argument

The factorization of geometric occupancy from optical opacity per Gaussian, which lets the same primitives serve as both surface and volume in the hybrid renderer.

If this is right

  • The same Gaussian set supports both surface reflection and volume transmission without extra primitives.
  • Specular-Aware Gradient Gating reduces floaters by blocking gradients from highly reflective regions into the transmission path.
  • Scene editing operations such as object removal or material changes become straightforward because occupancy and opacity are explicit.
  • Real-time rendering speed is retained while visual quality on mixed reflection-transmission scenes improves over prior Gaussian methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The occupancy-opacity split could be tested on scenes that combine reflection with refraction or participating media.
  • Extending the hybrid renderer to dynamic objects might require only modest changes to the occupancy term.
  • The approach may reduce the need for separate reflection and transmission layers in other splatting pipelines.

Load-bearing premise

Separating geometric occupancy from optical opacity inside each Gaussian is enough to eliminate optimization ambiguity between reflection and transmission without creating new rendering inconsistencies.

What would settle it

A test scene containing semi-transparent specular objects where the separated occupancy-opacity optimization still yields either blurry reflections, overly dark transmission, or new floaters not seen in joint-optimization baselines.

Figures

Figures reproduced from arXiv: 2605.18263 by Bowei Xing, Ji Shi, Ruohao Guo, Wenzhen Yue, Xianghua Ying.

Figure 1
Figure 1. Figure 1: Photorealistic rendering and decomposition of real-world scene with coexisting reflection and transmission. (Left) Com￾pared to prior works, our method robustly handles semi-transparent surfaces, avoiding blurry reflections or overly occluded transmission. (Right) Our high-fidelity results are achieved by decomposing the scene radiance into Reflection and Transmission layers, enabled by a unified Gaussian … view at source ↗
Figure 2
Figure 2. Figure 2: Overview of RT-Splatting. (Left) Transparent objects are represented by Gaussians with high geometric occupancy but low optical opacity, yielding strong contributions to surface aggregation while avoiding occlusion during volumetric compositing. (Right) Our hybrid rendering pipeline composites surface-based reflections from a deferred pass with volumetric transmission from a forward pass to produce the fin… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons on test-set views of real-world scenes. Our method significantly improves rendering quality over previous approaches, simultaneously yielding sharper reflections and clearer transmissions in semi-transparent regions. M, obtained from the pre-trained SAM2 model [20, 31] to provide additional supervision. During the deferred pass, we aggregate the expected optical opacity α of the fir… view at source ↗
Figure 4
Figure 4. Figure 4: Scene Editing. Left: edited appearances of car windows. Right: edited appearances of a plastic film. w/o joint opt. w/o gating Ours w/o occupancy w/o scattering Ours [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Decomposed transmission components across abla￾tion settings. occupancy and optical opacity. As shown in Tab. 3 and [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual decomposition on real-world scenes. We visualize the decomposed components of our method compared to baselines. For our method, Normal captures the surface geometry, while Depth corresponds to the volumetric accumulation. Since baseline methods do not explicitly model semi-transparent transmission, we visualize their diffuse component in the Transmission row. Our method achieves high-fidelity separa… view at source ↗
read the original abstract

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis with high visual quality. However, existing methods struggle with semi-transparent specular surfaces that exhibit both complex reflections and clear transmission, often producing blurry reflections or overly occluded transmission. To address this, we present RT-Splatting, a framework that disentangles each Gaussian's geometric occupancy from its optical opacity. This factorization yields a unified surface-volume scene representation with a single set of Gaussian primitives. Our hybrid renderer interprets this representation both as a surface to capture high-frequency reflections and as a volume to preserve clear transmission. To mitigate the ambiguity in jointly optimizing reflection and transmission, we introduce Specular-Aware Gradient Gating, which suppresses misleading gradients from highly specular regions into the transmission branch, effectively reducing distracting floaters. Experiments on challenging semi-transparent scenes show that RT-Splatting achieves state-of-the-art performance, delivering high-fidelity reflections and clear transmission with real-time rendering. Moreover, our factorization naturally enables flexible scene editing. The project page is available at https://sjj118.github.io/RT-Splatting.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces RT-Splatting for joint reflection-transmission modeling in 3D Gaussian Splatting. Each Gaussian is factored into geometric occupancy and optical opacity to yield a unified surface-volume representation. A hybrid renderer interprets the primitives both as surfaces for high-frequency reflections and as volumes for clear transmission. Specular-Aware Gradient Gating suppresses misleading gradients from specular regions into the transmission branch to reduce floaters. The approach is reported to achieve state-of-the-art performance on challenging semi-transparent scenes with real-time rendering and to enable flexible scene editing.

Significance. If the quantitative results and ablations hold, the work addresses a recognized limitation of 3DGS on semi-transparent specular surfaces by providing a concrete factorization and gating mechanism that preserves real-time performance. The hybrid surface-volume formulation and editing capability represent a practical advance for novel-view synthesis in AR/VR and content creation pipelines.

major comments (2)
  1. [§3.2] §3.2 (hybrid renderer formulation): the description of how surface and volume interpretations are combined during rendering lacks explicit equations for the blending or ray integration step; without this, it is difficult to verify that the occupancy-opacity separation avoids introducing new optimization inconsistencies between reflection and transmission branches.
  2. [§4] §4 (experiments): the central SOTA claim and the effectiveness of gradient gating in reducing floaters rest on quantitative tables and ablations, yet the provided text supplies no PSNR/SSIM/LPIPS numbers, per-component reflection vs. transmission metrics, or direct comparison against the baseline 3DGS on the same semi-transparent scenes; this evidence gap is load-bearing for the performance assertions.
minor comments (2)
  1. [Introduction] The term 'floaters' is used repeatedly without a short definition or reference to its standard usage in 3DGS literature; a one-sentence clarification in the introduction would improve accessibility.
  2. [Figures] Figure captions for the qualitative results should explicitly state the input views, novel views, and which method corresponds to each column to facilitate direct visual comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive assessment of the work's significance for real-time novel view synthesis with semi-transparent specular surfaces. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (hybrid renderer formulation): the description of how surface and volume interpretations are combined during rendering lacks explicit equations for the blending or ray integration step; without this, it is difficult to verify that the occupancy-opacity separation avoids introducing new optimization inconsistencies between reflection and transmission branches.

    Authors: We agree that the hybrid renderer description in §3.2 would be strengthened by explicit equations. In the revised manuscript we will add the full blending and ray-integration formulation, showing precisely how the occupancy channel drives surface-style reflection rendering while the opacity channel drives volumetric transmission, and how the two are combined without introducing optimization inconsistencies between branches. revision: yes

  2. Referee: [§4] §4 (experiments): the central SOTA claim and the effectiveness of gradient gating in reducing floaters rest on quantitative tables and ablations, yet the provided text supplies no PSNR/SSIM/LPIPS numbers, per-component reflection vs. transmission metrics, or direct comparison against the baseline 3DGS on the same semi-transparent scenes; this evidence gap is load-bearing for the performance assertions.

    Authors: The referee correctly notes that the current text does not present the requested quantitative metrics. We will expand §4 with complete tables reporting PSNR, SSIM and LPIPS for both overall and per-component (reflection / transmission) results, together with direct comparisons against the 3DGS baseline on the identical semi-transparent scenes. We will also add ablations that isolate the contribution of specular-aware gradient gating to floater reduction. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents a novel factorization separating geometric occupancy from optical opacity per Gaussian, a hybrid surface-volume renderer, and Specular-Aware Gradient Gating as independent technical contributions to resolve reflection-transmission ambiguity in 3DGS. These elements are introduced explicitly rather than derived from prior equations within the paper or reduced to self-citations. Performance claims rest on experimental results on semi-transparent scenes, not on tautological predictions or fitted parameters renamed as outputs. The central construction remains self-contained with no load-bearing step that equates to its own inputs by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the unproven premise that the proposed factorization cleanly separates reflection and transmission signals during joint optimization; no explicit free parameters, axioms, or invented entities are stated in the abstract.

pith-pipeline@v0.9.0 · 5731 in / 1155 out tokens · 36854 ms · 2026-05-20T12:00:48.304283+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    we factorize the standard per-Gaussian opacity into two physically motivated, learnable attributes. The geometric occupancy σ ∈ [0,1] encodes the probability that a ray interacts with the substance of the Gaussian. The optical opacity α ∈ [0,1] then specifies the conditional probability that the ray is absorbed or scattered once such an interaction occurs. Their product α_eff = σ α defines the effective opacity

  • IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Our hybrid renderer interprets this representation both as a surface to capture high-frequency reflections and as a volume to preserve clear transmission.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 2 internal anchors

  1. [1]

    Clear-splatting: Learning residual gaussian splats for transparent object manipulation

    Aviral Agrawal, Ritaban Roy, Bardienus Pieter Duisterhof, Keerthan Bhat Hekkadka, Hongyi Chen, and Jeffrey Ich- nowski. Clear-splatting: Learning residual gaussian splats for transparent object manipulation. InRoboNerF: 1st Work- shop On Neural Fields In Robotics at ICRA 2024, 2024. 3

    work page 2024

  2. [2]

    Deferred shading of transparent surfaces with shadows and refraction

    Ali Deniz Alada ˘glı. Deferred shading of transparent surfaces with shadows and refraction. Master’s thesis, Middle East Technical University (Turkey), 2015. 2

    work page 2015

  3. [3]

    Eikonal fields for refractive novel-view synthesis

    Mojtaba Bemana, Karol Myszkowski, Jeppe Revall Frisvad, Hans-Peter Seidel, and Tobias Ritschel. Eikonal fields for refractive novel-view synthesis. InACM SIGGRAPH 2022 Conference Proceedings, New York, NY , USA, 2022. Asso- ciation for Computing Machinery. 3, 6

    work page 2022

  4. [4]

    GI-GS: global illumination decomposition on gaussian splatting for inverse rendering

    Hongze Chen, Zehong Lin, and Jun Zhang. GI-GS: global illumination decomposition on gaussian splatting for inverse rendering. InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24- 28, 2025, 2025. 2, 3

    work page 2025

  5. [5]

    Nerrf: 3d reconstruction and view synthesis for transparent and specular objects with neural refractive- reflective fields.arXiv preprint arXiv:2309.13039, 2023

    Xiaoxue Chen, Junchen Liu, Hao Zhao, Guyue Zhou, and Ya-Qin Zhang. Nerrf: 3d reconstruction and view synthesis for transparent and specular objects with neural refractive- reflective fields.arXiv preprint arXiv:2309.13039, 2023. 3

    work page arXiv 2023

  6. [6]

    Jan-Niklas Dihlmann, Arjun Majumdar, Andreas Engel- hardt, Raphael Braun, and Hendrik P.A. Lensch. Subsurface scattering for gaussian splatting. InAdvances in Neural In- formation Processing Systems, pages 121765–121789. Cur- ran Associates, Inc., 2024. 2, 3

    work page 2024

  7. [7]

    Gs-id: Illumination decomposition on gaussian splatting via adap- tive light aggregation and diffusion-guided material priors

    Kang Du, Zhihao Liang, Yulin Shen, and Zeyu Wang. Gs-id: Illumination decomposition on gaussian splatting via adap- tive light aggregation and diffusion-guided material priors. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 26220–26229, 2025. 2, 3

    work page 2025

  8. [8]

    Planar reflection-aware neural radiance fields

    Chen Gao, Yipeng Wang, Changil Kim, Jia-Bin Huang, and Johannes Kopf. Planar reflection-aware neural radiance fields. InSIGGRAPH Asia 2024 Conference Papers, New York, NY , USA, 2024. Association for Computing Machin- ery. 3, 5

    work page 2024

  9. [9]

    Transparent object reconstruction via im- plicit differentiable refraction rendering

    Fangzhou Gao, Lianghao Zhang, Li Wang, Jiamin Cheng, and Jiawan Zhang. Transparent object reconstruction via im- plicit differentiable refraction rendering. InSIGGRAPH Asia 2023 Conference Papers, New York, NY , USA, 2023. Asso- ciation for Computing Machinery. 3

    work page 2023

  10. [10]

    Relightable 3d gaussians: Re- alistic point cloud relighting with brdf decomposition and ray tracing

    Jian Gao, Chun Gu, Youtian Lin, Zhihao Li, Hao Zhu, Xun Cao, Li Zhang, and Yao Yao. Relightable 3d gaussians: Re- alistic point cloud relighting with brdf decomposition and ray tracing. InEuropean Conference on Computer Vision, pages 73–89. Springer, 2024. 2

    work page 2024

  11. [11]

    Ref-neus: Ambiguity-reduced neural implicit surface learning for multi-view reconstruction with reflection

    Wenhang Ge, Tao Hu, Haoyu Zhao, Shu Liu, and Ying-Cong Chen. Ref-neus: Ambiguity-reduced neural implicit surface learning for multi-view reconstruction with reflection. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4251–4260, 2023. 2

    work page 2023

  12. [12]

    Prtgs: Precomputed radiance transfer of gaussian splats for real-time high-quality relighting

    Yijia Guo, Yuanxi Bai, Liwen Hu, Ziyi Guo, Mianzhi Liu, Yu Cai, Tiejun Huang, and Lei Ma. Prtgs: Precomputed radiance transfer of gaussian splats for real-time high-quality relighting. InProceedings of the 32nd ACM International Conference on Multimedia, page 5112–5120, New York, NY , USA, 2024. Association for Computing Machinery. 2

    work page 2024

  13. [13]

    Nerfren: Neural radiance fields with reflec- tions

    Yuan-Chen Guo, Di Kang, Linchao Bao, Yu He, and Song- Hai Zhang. Nerfren: Neural radiance fields with reflec- tions. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 18409– 18418, 2022. 5

    work page 2022

  14. [14]

    2d gaussian splatting for geometrically accu- rate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accu- rate radiance fields. InACM SIGGRAPH 2024 Conference Papers, New York, NY , USA, 2024. Association for Com- puting Machinery. 2, 3, 6, 7, 1, 4

    work page 2024

  15. [15]

    Transparentgs: Fast inverse rendering of transpar- ent objects with gaussians.ACM Trans

    Letian Huang, Dongwei Ye, Jialin Dan, Chengzhi Tao, Hui- wen Liu, Kun Zhou, Bo Ren, Yuanqi Li, Yanwen Guo, and Jie Guo. Transparentgs: Fast inverse rendering of transpar- ent objects with gaussians.ACM Trans. Graph., 44(4), 2025. 2, 3, 6

    work page 2025

  16. [16]

    Dex-nerf: Using a neural radiance field to grasp trans- parent objects

    Jeffrey Ichnowski, Yahav Avigal, Justin Kerr, and Ken Gold- berg. Dex-nerf: Using a neural radiance field to grasp trans- parent objects. InProceedings of the 5th Conference on Robot Learning, pages 526–536. PMLR, 2022. 3

    work page 2022

  17. [17]

    Gaussian- shader: 3d gaussian splatting with shading functions for re- flective surfaces

    Yingwenqi Jiang, Jiadong Tu, Yuan Liu, Xifeng Gao, Xi- aoxiao Long, Wenping Wang, and Yuexin Ma. Gaussian- shader: 3d gaussian splatting with shading functions for re- flective surfaces. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 5322–5332, 2024. 2, 7, 3, 4

    work page 2024

  18. [18]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

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

    work page 2023

  19. [19]

    Ref2- nerf: Reflection and refraction aware neural radiance field

    Wooseok Kim, Taiki Fukiage, and Takeshi Oishi. Ref2- nerf: Reflection and refraction aware neural radiance field. In2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7196–7203, 2024. 3

    work page 2024

  20. [20]

    Segment Anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick. Segment anything.arXiv:2304.02643, 2023. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  21. [21]

    Tanks and temples: benchmarking large-scale scene reconstruction.ACM Trans

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: benchmarking large-scale scene reconstruction.ACM Trans. Graph., 36(4), 2017. 7, 3

    work page 2017

  22. [22]

    Rgs-dr: Reflective gaussian surfels with deferred rendering for shiny objects.arXiv preprint arXiv:2504.18468, 2025

    Georgios Kouros, Minye Wu, and Tinne Tuytelaars. Rgs-dr: Reflective gaussian surfels with deferred rendering for shiny objects.arXiv preprint arXiv:2504.18468, 2025. 2, 3

    work page arXiv 2025

  23. [23]

    Tsgs: Improving gaussian splatting for transparent surface reconstruction via normal and de-lighting priors

    Mingwei Li, Pu Pang, Hehe Fan, Hua Huang, and Yi Yang. Tsgs: Improving gaussian splatting for transparent surface reconstruction via normal and de-lighting priors. InProceed- ings of the 33rd ACM International Conference on Multime- 9 dia, page 7220–7229, New York, NY , USA, 2025. Associa- tion for Computing Machinery. 3, 6

    work page 2025

  24. [24]

    Envidr: Implicit differentiable renderer with neural environment lighting

    Ruofan Liang, Huiting Chen, Chunlin Li, Fan Chen, Sel- vakumar Panneer, and Nandita Vijaykumar. Envidr: Implicit differentiable renderer with neural environment lighting. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 79–89, 2023. 2

    work page 2023

  25. [25]

    Gs-ir: 3d gaussian splatting for inverse rendering

    Zhihao Liang, Qi Zhang, Ying Feng, Ying Shan, and Kui Jia. Gs-ir: 3d gaussian splatting for inverse rendering. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21644–21653, 2024. 2

    work page 2024

  26. [26]

    Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images.ACM Trans

    Yuan Liu, Peng Wang, Cheng Lin, Xiaoxiao Long, Jiepeng Wang, Lingjie Liu, Taku Komura, and Wenping Wang. Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images.ACM Trans. Graph., 42(4), 2023. 2

    work page 2023

  27. [27]

    Specnerf: Gaussian directional encoding for spec- ular reflections

    Li Ma, Vasu Agrawal, Haithem Turki, Changil Kim, Chen Gao, Pedro Sander, Michael Zollh ¨ofer, and Christian Richardt. Specnerf: Gaussian directional encoding for spec- ular reflections. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21188–21198, 2024

    work page 2024

  28. [28]

    Neural microfacet fields for inverse render- ing

    Alexander Mai, Dor Verbin, Falko Kuester, and Sara Fridovich-Keil. Neural microfacet fields for inverse render- ing. InProceedings of the IEEE/CVF International Confer- ence on Computer Vision (ICCV), pages 408–418, 2023. 2

    work page 2023

  29. [29]

    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 synthe- sis.Commun. ACM, 65(1):99–106, 2021. 3

    work page 2021

  30. [30]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Rai- son, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. Pytorch: An imperative style, high-per...

    work page 2019

  31. [31]

    SAM 2: Segment Anything in Images and Videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross Girshick, Piotr Doll´ar, and Christoph Feicht- enhofer. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  32. [32]

    Normal-nerf: Ambiguity-robust normal es- timation for highly reflective scenes.Proceedings of the AAAI Conference on Artificial Intelligence, 39(7):6869– 6877, 2025

    Ji Shi, Xianghua Ying, Ruohao Guo, Bowei Xing, and Wenzhen Yue. Normal-nerf: Ambiguity-robust normal es- timation for highly reflective scenes.Proceedings of the AAAI Conference on Artificial Intelligence, 39(7):6869– 6877, 2025. 2

    work page 2025

  33. [33]

    Gir: 3d gaussian inverse rendering for relightable scene factorization.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, pages 1–12, 2025

    Yahao Shi, Yanmin Wu, Chenming Wu, Xing Liu, Chen Zhao, Haocheng Feng, Jian Zhang, Bin Zhou, Errui Ding, and Jingdong Wang. Gir: 3d gaussian inverse rendering for relightable scene factorization.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, pages 1–12, 2025. 2

    work page 2025

  34. [34]

    Spectre-gs: Modeling highly specular sur- faces with reflected nearby objects by tracing rays in 3d gaus- sian splatting

    Jiajun Tang, Fan Fei, Zhihao Li, Xiao Tang, Shiyong Liu, Youyu Chen, Binxiao Huang, Zhenyu Chen, Xiaofei Wu, and Boxin Shi. Spectre-gs: Modeling highly specular sur- faces with reflected nearby objects by tracing rays in 3d gaus- sian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16133–16142, ...

    work page 2025

  35. [35]

    Barron, and Pratul P

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T. Barron, and Pratul P. Srinivasan. Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5491–5500,

    work page

  36. [36]

    Srinivasan, Peter Hedman, Ben Milden- hall, Benjamin Attal, Richard Szeliski, and Jonathan T

    Dor Verbin, Pratul P. Srinivasan, Peter Hedman, Ben Milden- hall, Benjamin Attal, Richard Szeliski, and Jonathan T. Bar- ron. Nerf-casting: Improved view-dependent appearance with consistent reflections. InSIGGRAPH Asia 2024 Con- ference Papers, New York, NY , USA, 2024. Association for Computing Machinery. 2, 7, 3

    work page 2024

  37. [37]

    Nemto: Neural environment matting for novel view and re- lighting synthesis of transparent objects

    Dongqing Wang, Tong Zhang, and Sabine S ¨usstrunk. Nemto: Neural environment matting for novel view and re- lighting synthesis of transparent objects. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 317–327, 2023. 3

    work page 2023

  38. [38]

    Unisdf: Unifying neural representations for high- fidelity 3d reconstruction of complex scenes with reflec- tions

    Fangjinhua Wang, Marie-Julie Rakotosaona, Michael Niemeyer, Richard Szeliski, Marc Pollefeys, and Federico Tombari. Unisdf: Unifying neural representations for high- fidelity 3d reconstruction of complex scenes with reflec- tions. InAdvances in Neural Information Processing Sys- tems, pages 3157–3184. Curran Associates, Inc., 2024. 2

    work page 2024

  39. [39]

    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, 13(4): 600–612, 2004. 7, 1

    work page 2004

  40. [40]

    Deferredgs: Decoupled and editable gaussian splatting with deferred shading.arXiv preprint arXiv:2404.09412, 2024

    Tong Wu, Jia-Mu Sun, Yu-Kun Lai, Yuewen Ma, Leif Kobbelt, and Lin Gao. Deferredgs: Decoupled and editable gaussian splatting with deferred shading.arXiv preprint arXiv:2404.09412, 2024. 2, 3

    work page arXiv 2024

  41. [41]

    En- vgs: Modeling view-dependent appearance with environ- ment gaussian

    Tao Xie, Xi Chen, Zhen Xu, Yiman Xie, Yudong Jin, Yu- jun Shen, Sida Peng, Hujun Bao, and Xiaowei Zhou. En- vgs: Modeling view-dependent appearance with environ- ment gaussian. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5742–5751, 2025. 2, 4, 7, 1, 3

    work page 2025

  42. [42]

    Reflective gaussian splatting

    Yuxuan Yao, Zixuan Zeng, Chun Gu, Xiatian Zhu, and Li Zhang. Reflective gaussian splatting. InThe Thirteenth In- ternational Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025, 2025

    work page 2025

  43. [43]

    3d gaussian splat- ting with deferred reflection

    Keyang Ye, Qiming Hou, and Kun Zhou. 3d gaussian splat- ting with deferred reflection. InACM SIGGRAPH 2024 Con- ference Papers, New York, NY , USA, 2024. Association for Computing Machinery. 2, 4, 7, 1, 3

    work page 2024

  44. [44]

    Progressive ra- diance distillation for inverse rendering with gaussian splat- ting.arXiv preprint arXiv:2408.07595, 2024

    Keyang Ye, Qiming Hou, and Kun Zhou. Progressive ra- diance distillation for inverse rendering with gaussian splat- ting.arXiv preprint arXiv:2408.07595, 2024

    work page arXiv 2024

  45. [45]

    Geosplatting: Towards geometry guided gaus- sian splatting for physically-based inverse rendering

    Kai Ye, Chong Gao, Guanbin Li, Wenzheng Chen, and Bao- quan Chen. Geosplatting: Towards geometry guided gaus- sian splatting for physically-based inverse rendering. In 10 Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 28991–29000, 2025. 2, 3

    work page 2025

  46. [46]

    Nerfrac: Neural radiance fields through refractive surface

    Yifan Zhan, Shohei Nobuhara, Ko Nishino, and Yinqiang Zheng. Nerfrac: Neural radiance fields through refractive surface. InProceedings of the IEEE/CVF International Con- ference on Computer Vision (ICCV), pages 18402–18412,

    work page

  47. [47]

    Nerf++: Analyzing and improving neural radiance fields

    Kai Zhang, Gernot Riegler, Noah Snavely, and Vladlen Koltun. Nerf++: Analyzing and improving neural radiance fields.arXiv preprint arXiv:2010.07492, 2020. 2

    work page arXiv 2010

  48. [48]

    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 Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2018. 7

    work page 2018

  49. [49]

    Refgaussian: Dis- entangling reflections from 3d gaussian splatting for realistic rendering.arXiv preprint arXiv:2406.05852, 2024

    Rui Zhang, Tianyue Luo, Weidong Yang, Ben Fei, Jingyi Xu, Qingyuan Zhou, Keyi Liu, and Ying He. Refgaussian: Dis- entangling reflections from 3d gaussian splatting for realistic rendering.arXiv preprint arXiv:2406.05852, 2024. 3, 5

    work page arXiv 2024

  50. [50]

    Ref-gs: Directional factorization for 2d gaussian splatting

    Youjia Zhang, Anpei Chen, Yumin Wan, Zikai Song, Jun- qing Yu, Yawei Luo, and Wei Yang. Ref-gs: Directional factorization for 2d gaussian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 26483–26492, 2025. 2, 3, 4, 5, 7, 1

    work page 2025

  51. [51]

    Gs-ror2: Bidirectional-guided 3dgs and sdf for reflective object re- lighting and reconstruction.ACM Trans

    Zuoliang Zhu, Beibei Wang, and Jian Yang. Gs-ror2: Bidirectional-guided 3dgs and sdf for reflective object re- lighting and reconstruction.ACM Trans. Graph., 45(1),

    work page

  52. [52]

    2, 3 11 RT-Splatting: Joint Reflection-Transmission Modeling with Gaussian Splatting Supplementary Material In the supplementary material, we provide additional implementation details of our method (Sec. A). We also present an ablation study that examines the sensitivity of Specular-Aware Gradient Gating to the gating strengthk (Sec. B). Finally, we repor...

    work page