pith. sign in

arxiv: 2606.30861 · v1 · pith:OBXWQO7Gnew · submitted 2026-06-29 · 💻 cs.GR

Editable Physically-based Reflections in Raytraced Gaussian Radiance Fields

Yohan Poirier-Ginter , Jeffrey Hu , Jean-Fran\c{c}ois Lalonde , George Drettakis This is my paper

Pith reviewed 2026-07-01 01:21 UTC · model grok-4.3

classification 💻 cs.GR
keywords 3D Gaussian Splattingspecular reflectionspath tracingscene editingradiance fieldsphysically based renderingray tracingmulti-bounce reflections
0
0 comments X

The pith

Optimizing a diffuse Gaussian scene and applying path tracing reconstructs true reflectors and reflected objects for real-time editable specular reflections.

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

The paper sets out to replace the common practice of embedding fake reflected geometry behind surfaces in radiance fields with a representation that keeps the actual reflector and reflected content separate. It extracts diffuse and specular buffers from the input views, optimizes a Gaussian model that contains only the diffuse component, then uses path tracing on top of that model to produce the reflections. A custom training schedule makes the optimization converge, and a fast ray tracer for Gaussian primitives handles multi-bounce paths efficiently enough for real-time use. The result is a single scene model in which specular reflections, including those of unseen objects, can be edited while remaining physically consistent across viewpoints.

Core claim

By using diffuse and specular buffers of input training views, we optimize a diffuse version of the scene and use path tracing to efficiently generate physically based specular reflections. A specialized training method allows this process to converge. A fast ray tracing algorithm for 3D Gaussian primitives enables efficient multi-bounce reflections. Our method reconstructs reflectors and reflected objects, including those not seen in the input images, in a unique scene representation. Our solution allows real-time, consistent editing of captured scenes with specular reflections, including multi-bounce effects, changing roughness, and more.

What carries the argument

A diffuse-only 3D Gaussian scene representation optimized from specular-free buffers, augmented at render time by path tracing and a specialized ray tracer for Gaussian primitives.

If this is right

  • Multi-bounce specular reflections become available without storing fake geometry behind reflectors.
  • Surface properties such as roughness can be changed after capture while reflections update consistently across all views.
  • Objects visible only through reflections can be reconstructed and edited even if they never appear directly in the input photographs.
  • Editing operations remain real-time because the diffuse base and the path-traced overlay are decoupled.

Where Pith is reading between the lines

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

  • If buffer extraction improves, the same separation could support editing of other view-dependent effects such as refraction.
  • The current reliance on ground-truth buffers for the main results suggests that real-scene performance will track advances in learning-based decomposition methods.
  • Because the diffuse model is trained separately, the approach may be combinable with existing diffuse-only editing tools without retraining the entire radiance field.

Load-bearing premise

The method assumes that sufficiently accurate diffuse and specular buffers can be obtained from the input views via learning-based extraction and that the specialized training procedure converges to a usable diffuse scene representation.

What would settle it

A controlled test on synthetic data where the generated reflections after material edits fail to match ground-truth path-traced images or become inconsistent when the viewpoint changes.

Figures

Figures reproduced from arXiv: 2606.30861 by George Drettakis, Jean-Fran\c{c}ois Lalonde, Jeffrey Hu, Yohan Poirier-Ginter.

Figure 1
Figure 1. Figure 1: Our Gaussian-based radiance-field method allows interactive editing of path traced reflections, with consistent updates. We develop our proof-of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our method takes as input several buffers for every training view (left), and reconstructs a unique gaussian-based scene where reflections in novel [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: We showcase reconstruction of objects viewed indirectly by recovering the cover of a book only visible indirectly through a mirror in the training views [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Radiance fields, recovered for example with EnvGS [Xie et al [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: We support physically-based reflections with different levels of rough [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: We support multi-bounce reflections during optimization. The num [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: We accelerate integration by approximating Gaussians of low contri [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: We encapsulate Gaussians with oriented bounding boxes (OBB) [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: We fine-tune a pretrained Stable Diffusion 2 network into a one-step [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Example network predictions (left) for a given input view (top), and [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Editing a synthetic scene: changing the base reflectance ( [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
Figure 11
Figure 11. Figure 11: Editing two synthetic scenes: making a typewriter diffuse (first row), [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: Our method reconstructs the environment behind the camera with [PITH_FULL_IMAGE:figures/full_fig_p009_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Disentanglement results of different methods on [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Our method compared against EnvGS [Xie et al [PITH_FULL_IMAGE:figures/full_fig_p012_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Our method leverages hardware-accelerated transforms, providing [PITH_FULL_IMAGE:figures/full_fig_p013_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Training with just 4 views still results in clean reflections provided [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Result of fitting our method (left) to network predicted buffers (right). [PITH_FULL_IMAGE:figures/full_fig_p020_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Disentanglement results for the Shiny Kitchen scene. SA Conference Papers ’25, December 15–18, 2025, Hong Kong, Hong Kong [PITH_FULL_IMAGE:figures/full_fig_p021_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Disentanglement results for the Shiny Livingroom scene. SA Conference Papers ’25, December 15–18, 2025, Hong Kong, Hong Kong [PITH_FULL_IMAGE:figures/full_fig_p022_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Disentanglement results for the Shiny Office scene. SA Conference Papers ’25, December 15–18, 2025, Hong Kong, Hong Kong [PITH_FULL_IMAGE:figures/full_fig_p023_21.png] view at source ↗
read the original abstract

Radiance fields such as 3D Gaussian Splatting allow real-time rendering of scenes captured from photos. They also reconstruct most specular reflections with high visual quality, but typically model them with "fake" reflected geometry, using primitives behind the reflector. Our goal is to correctly reconstruct the reflector and the reflected objects such as to make specular reflections editable. We present a proof of concept which exploits promising learning-based methods to extract diffuse and specular buffers from photos, as well as geometry and BRDF buffers. Our method builds on three key components. First, by using diffuse and specular buffers of input training views, we optimize a diffuse version of the scene and use path tracing to efficiently generate physically based specular reflections. Second, we present a specialized training method that allows this process to converge. Finally, we present a fast ray tracing algorithm for 3D Gaussian primitives that enables efficient multi-bounce reflections. Our method reconstructs reflectors and reflected objects, including those not seen in the input images, in a unique scene representation. Our solution allows real-time, consistent editing of captured scenes with specular reflections, including multi-bounce effects, changing roughness, and more. We mainly show results using ground truth buffers from synthetic scenes, and also preliminary results in real scenes with currently imperfect learning-based buffers. Code and data are available at: https://repo-sam.inria.fr/nerphys/editable-gaussian-reflections/

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 paper presents a proof-of-concept method for editable, physically-based specular reflections in 3D Gaussian radiance fields. It extracts diffuse and specular (plus geometry/BRDF) buffers from input views via learning-based methods, optimizes a diffuse Gaussian scene representation, uses path tracing to generate the specular component, applies a specialized training procedure to ensure convergence, and introduces a fast ray-tracing algorithm over Gaussian primitives to support efficient multi-bounce reflections. The central claim is that this yields a unique scene representation that correctly reconstructs both reflectors and reflected objects (including those unseen in the input images) and enables real-time, consistent editing operations such as changing roughness or multi-bounce effects. Main results use ground-truth buffers on synthetic scenes; real-scene results are described as preliminary with imperfect learned buffers. Code and data are released.

Significance. If the claims hold, the work would be significant for radiance-field research in computer graphics: it replaces the common practice of modeling reflections via “fake” geometry behind the reflector with a physically-based, editable representation that supports path-traced multi-bounce effects. The combination of buffer separation, specialized optimization, and custom Gaussian ray tracing is a novel integration. Public release of code and data is a clear strength that aids reproducibility and follow-up work.

major comments (2)
  1. [Method overview and results sections] The central claim that reflected objects not visible in the input images can be reconstructed relies on the path-tracing component and the assumption that diffuse/specular buffers are sufficiently accurate. The manuscript provides no quantitative error analysis or ablation on buffer quality (synthetic GT vs. learned real buffers), which is load-bearing for the reconstruction and editing claims; without such analysis the extent to which unseen geometry is truly recovered versus hallucinated remains unclear.
  2. [Training procedure description] The specialized training procedure is presented as essential for convergence, yet no derivation, loss formulation, or comparison against standard 3DGS optimization is supplied in the high-level description. This omission makes it impossible to verify why the procedure succeeds where conventional training fails, undermining assessment of the method’s novelty and robustness.
minor comments (2)
  1. [Real-scene results] The abstract states that results on real scenes are “preliminary”; the corresponding section should explicitly list the current failure modes and buffer-error sensitivity to guide readers on the practical scope of the approach.
  2. [Ray-tracing algorithm] Notation for the diffuse versus specular Gaussian primitives and the ray-tracing acceleration structure should be introduced with a small diagram or pseudocode for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment of the work's significance and for highlighting the value of the public code and data release. We address each major comment below and will revise the manuscript accordingly to improve clarity and support for the claims.

read point-by-point responses

  1. Referee: [Method overview and results sections] The central claim that reflected objects not visible in the input images can be reconstructed relies on the path-tracing component and the assumption that diffuse/specular buffers are sufficiently accurate. The manuscript provides no quantitative error analysis or ablation on buffer quality (synthetic GT vs. learned real buffers), which is load-bearing for the reconstruction and editing claims; without such analysis the extent to which unseen geometry is truly recovered versus hallucinated remains unclear.

    Authors: We agree that quantitative analysis of buffer quality is important for substantiating the reconstruction claims. The manuscript explicitly states that main results use ground-truth buffers on synthetic scenes to demonstrate the core method in isolation, while real-scene results are presented as preliminary due to limitations in current learning-based buffers. To strengthen the paper, we will add a dedicated ablation subsection with quantitative metrics (e.g., PSNR/SSIM on reconstructed unseen geometry and reflection accuracy) comparing ground-truth versus learned buffers, including error propagation analysis from buffer inaccuracies. revision: yes

  2. Referee: [Training procedure description] The specialized training procedure is presented as essential for convergence, yet no derivation, loss formulation, or comparison against standard 3DGS optimization is supplied in the high-level description. This omission makes it impossible to verify why the procedure succeeds where conventional training fails, undermining assessment of the method’s novelty and robustness.

    Authors: The manuscript describes the specialized training at a high level in the method overview because the full derivation, loss terms (including the custom regularization for diffuse/specular separation and path-traced supervision), and direct comparisons to vanilla 3DGS optimization are provided in the supplementary material and implementation details section. However, we acknowledge that the main text could better highlight these elements for readers. We will expand the main-text description with a concise derivation outline, explicit loss formulation, and a short comparison table or paragraph against standard 3DGS to improve accessibility without altering the core contribution. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes a proof-of-concept pipeline that takes diffuse/specular buffers (extracted via external learning-based methods or ground truth) as input, optimizes a diffuse Gaussian scene representation, and applies path tracing plus a custom ray-tracing algorithm for Gaussians to produce editable specular reflections. No equations, fitted parameters, or self-citations are presented that reduce the claimed reconstruction of reflectors and unseen reflected objects to a definitional identity or a statistically forced prediction. The method explicitly qualifies results as dependent on buffer quality and training convergence, treating these as external prerequisites rather than deriving them internally. The central claims therefore remain self-contained against the stated inputs and do not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The approach implicitly relies on the existence of accurate learned buffers and convergence of the custom training schedule, but these are not formalized.

pith-pipeline@v0.9.1-grok · 5799 in / 1210 out tokens · 33203 ms · 2026-07-01T01:21:07.995286+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

95 extracted references · 3 canonical work pages

  1. [1]

    Srinivasan and Matthew Tancik and Jonathan T

    Ben Mildenhall and Pratul P. Srinivasan and Matthew Tancik and Jonathan T. Barron and Ravi Ramamoorthi and Ren Ng , year=

  2. [2]

    Advances in neural rendering , author=

  3. [3]

    Structure-from-motion revisited , author=

  4. [4]

    Dust3r: Geometric

    Wang, Shuzhe and Leroy, Vincent and Cabon, Yohann and Chidlovskii, Boris and Revaud, Jerome , booktitle=CVPR, year=. Dust3r: Geometric

  5. [5]

    Grounding image matching in 3d with

    Leroy, Vincent and Cabon, Yohann and Revaud, J. Grounding image matching in 3d with

  6. [6]

    Wang, Jianyuan and Chen, Minghao and Karaev, Nikita and Vedaldi, Andrea and Rupprecht, Christian and Novotny, David , booktitle=CVPR, year=

  7. [7]

    Barron, Jonathan T and Mildenhall, Ben and Verbin, Dor and Srinivasan, Pratul P and Hedman, Peter , booktitle=CVPR, year=

  8. [8]

    Barron, Jonathan T and Mildenhall, Ben and Tancik, Matthew and Hedman, Peter and Martin-Brualla, Ricardo and Srinivasan, Pratul P , booktitle=ICCV, year=

  9. [9]

    Barron, Jonathan T and Mildenhall, Ben and Verbin, Dor and Srinivasan, Pratul P and Hedman, Peter , booktitle=ICCV, year=

  10. [10]

    Instant neural graphics primitives with a multiresolution hash encoding , author=

  11. [11]

    Plenoxels: Radiance fields without neural networks , author=

  12. [12]

    Verbin, Dor and Hedman, Peter and Mildenhall, Ben and Zickler, Todd and Barron, Jonathan T and Srinivasan, Pratul P , booktitle=CVPR, year=

  13. [13]

    Ma, Li and Agrawal, Vasu and Turki, Haithem and Kim, Changil and Gao, Chen and Sander, Pedro and Zollh

  14. [14]

    Liang, Ruofan and Chen, Huiting and Li, Chunlin and Chen, Fan and Panneer, Selvakumar and Vijaykumar, Nandita , booktitle=CVPR, year=

  15. [15]

    Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images , author=

  16. [16]

    UniSDF: Unifying neural representations for high-fidelity 3d reconstruction of complex scenes with reflections , author=

  17. [17]

    Verbin, Dor and Srinivasan, Pratul P and Hedman, Peter and Mildenhall, Ben and Attal, Benjamin and Szeliski, Richard and Barron, Jonathan T , booktitle=SIGGRAPHAsia, year=

  18. [18]

    Guo, Yuan-Chen and Kang, Di and Bao, Linchao and He, Yu and Zhang, Song-Hai , booktitle=CVPR, year=

  19. [19]

    Zeng, Junyi and Bao, Chong and Chen, Rui and Dong, Zilong and Zhang, Guofeng and Bao, Hujun and Cui, Zhaopeng , booktitle=ACMMM, year=

  20. [20]

    Yin, Ze-Xin and Jiao, Peng-Yi and Qiu, Jiaxiong and Cheng, Ming-Ming and Ren, Bo , journal=PAMI, year=

  21. [21]

    Liu, Jiayue and Tang, Xiao and Cheng, Freeman and Yang, Roy and Li, Zhihao and Liu, Jianzhuang and Huang, Yi and Lin, Jiaqi and Liu, Shiyong and Wu, Xiaofei and others , booktitle=ECCV, year=

  22. [22]

    Meng, Jiarui and Li, Haijie and Wu, Yanmin and Gao, Qiankun and Yang, Shuzhou and Zhang, Jian and Ma, Siwei , booktitle=

  23. [23]

    Kerbl, Bernhard and Kopanas, Georgios and Leimk

  24. [24]

    Huang, Binbin and Yu, Zehao and Chen, Anpei and Geiger, Andreas and Gao, Shenghua , booktitle=SIGGRAPH, year=

  25. [25]

    Relightable

    Gao, Jian and Gu, Chun and Lin, Youtian and Li, Zhihao and Zhu, Hao and Cao, Xun and Zhang, Li and Yao, Yao , booktitle=ECCV, year=. Relightable

  26. [26]

    Xie, Tao and Chen, Xi and Xu, Zhen and Xie, Yiman and Jin, Yudong and Shen, Yujun and Peng, Sida and Bao, Hujun and Zhou, Xiaowei , booktitle=CVPR, year=

  27. [27]

    arXiv preprint arXiv:2412.19282 , year=

    Reflective Gaussian Splatting , author=. arXiv preprint arXiv:2412.19282 , year=

    arXiv

  28. [28]

    3D Gaussian Ray Tracing: Fast Tracing of Particle Scenes , author=

  29. [29]

    Bi, Zoubin and Zeng, Yixin and Zeng, Chong and Pei, Fan and Feng, Xiang and Zhou, Kun and Wu, Hongzhi , booktitle=SIGGRAPHAsia, year=

  30. [30]

    Jiang, Yingwenqi and Tu, Jiadong and Liu, Yuan and Gao, Xifeng and Long, Xiaoxiao and Wang, Wenping and Ma, Yuexin , booktitle=CVPR, year=

  31. [31]

    Ye, Keyang and Hou, Qiming and Zhou, Kun , booktitle=SIGGRAPH, year=

  32. [32]

    Zhang, Youjia and Chen, Anpei and Wan, Yumin and Song, Zikai and Yu, Junqing and Luo, Yawei and Yang, Wei , booktitle=CVPR, year=

  33. [33]

    Interactive indirect illumination using voxel cone tracing , author=

  34. [34]

    Zeng, Zheng and Deschaintre, Valentin and Georgiev, Iliyan and Hold-Geoffroy, Yannick and Hu, Yiwei and Luan, Fujun and Yan, Ling-Qi and Ha

  35. [35]

    arXiv preprint arXiv:2502.19318 , year=

    Does 3D Gaussian Splatting Need Accurate Volumetric Rendering? , author=. arXiv preprint arXiv:2502.19318 , year=

    arXiv

  36. [36]

    Mallick, Saswat Subhajyoti and Goel, Rahul and Kerbl, Bernhard and Steinberger, Markus and Carrasco, Francisco Vicente and De La Torre, Fernando , booktitle=SIGGRAPH, year=. Taming

  37. [37]

    ACM Transactions on Graphics (ToG) , volume=

    Mitsuba 2: A retargetable forward and inverse renderer , author=. ACM Transactions on Graphics (ToG) , volume=. 2019 , publisher=

    2019

  38. [38]

    ACM Transactions on Graphics (TOG) , volume=

    Differentiable monte carlo ray tracing through edge sampling , author=. ACM Transactions on Graphics (TOG) , volume=. 2018 , publisher=

    2018

  39. [39]

    ACM Transactions on Graphics , volume=

    Neural Point Catacaustics for Novel-View Synthesis of Reflections , author=. ACM Transactions on Graphics , volume=

  40. [40]

    2025 , organization=

    Efficient Perspective-Correct 3D Gaussian Splatting Using Hybrid Transparency , author=. 2025 , organization=

    2025

  41. [41]

    Proceedings of the Royal Society of London

    The interpretation of structure from motion , author=. Proceedings of the Royal Society of London. Series B. Biological Sciences , volume=. 1979 , publisher=

    1979

  42. [42]

    Hypersim: A photorealistic synthetic dataset for holistic indoor scene understanding , author=

  43. [43]

    High-resolution image synthesis with latent diffusion models , author=

  44. [44]

    2013 , url =

    Brian Karis , title =. 2013 , url =

    2013

  45. [45]

    Adaptive radiosity textures for bidirectional ray tracing , author=. Proc. 17th Ann. Conf. Comp. Graph. Interac. Tech. , pages=

  46. [46]

    Neural rays for occlusion-aware image-based rendering , author=

  47. [47]

    Proceedings of the 25th annual conference on Computer graphics and interactive techniques , pages=

    Rendering synthetic objects into real scenes: Bridging traditional and image-based graphics with global illumination and high dynamic range photography , author=. Proceedings of the 25th annual conference on Computer graphics and interactive techniques , pages=. 1998 , organization=

    1998

  48. [48]

    2023 , doi =

    Choi, Changwoon and Kim, Juhyeon and Kim, Young Min , journal = CGF, volume =. 2023 , doi =

    2023

  49. [49]

    Extracting triangular

    Munkberg, Jacob and Hasselgren, Jon and Shen, Tianchang and Gao, Jun and Chen, Wenzheng and Evans, Alex and M. Extracting triangular

  50. [50]

    Wang, Peng and Liu, Lingjie and Liu, Yuan and Theobalt, Christian and Komura, Taku and Wang, Wenping , booktitle=NeurIPS, year=

  51. [51]

    Neuralangelo: High-fidelity neural surface reconstruction , author=

  52. [52]

    Neural directional encoding for efficient and accurate view-dependent appearance modeling , author=

  53. [53]

    Zhuang, Yiyu and Zhang, Qi and Wang, Xuan and Zhu, Hao and Feng, Ying and Li, Xiaoyu and Shan, Ying and Cao, Xun , journal=

  54. [54]

    1982 , publisher =

    A Reflectance Model for Computer Graphics , author =. 1982 , publisher =

    1982

  55. [55]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A Method for Stochastic Optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    Pith/arXiv arXiv

  56. [56]

    Mip-splatting: Alias-free

    Yu, Zehao and Chen, Anpei and Huang, Binbin and Sattler, Torsten and Geiger, Andreas , booktitle=CVPR, year=. Mip-splatting: Alias-free

  57. [57]

    2024 , url =

    Kerbl, Bernhard and Meuleman, Andreas and Kopanas, Georgios and Wimmer, Michael and Lanvin, Alexandre and Drettakis, George , title =. 2024 , url =

    2024

  58. [58]

    A Diffusion Approach to Radiance Field Relighting using Multi-Illumination Synthesis , author=

  59. [59]

    Tang, Zhe Jun and Cham, Tat-Jen , booktitle=ECCV, year=

  60. [60]

    Chen, Anpei and Xu, Zexiang and Geiger, Andreas and Yu, Jingyi and Su, Hao , booktitle=ECCV, year=

  61. [61]

    Jin, Haian and Liu, Isabella and Xu, Peijia and Zhang, Xiaoshuai and Han, Songfang and Bi, Sai and Zhou, Xiaowei and Xu, Zexiang and Su, Hao , booktitle=CVPR, year=

  62. [62]

    2025 , publisher=

    Don't Splat your Gaussians: Volumetric Ray-Traced Primitives for Modeling and Rendering Scattering and Emissive Media , author=. 2025 , publisher=

    2025

  63. [63]

    Raygauss: Volumetric gaussian-based ray casting for photorealistic novel view synthesis , author=

  64. [64]

    Steven G. Parker and James Bigler and Andreas Dietrich and Heiko Friedrich and Jared Hoberock and David Luebke and David McAllister and Morgan McGuire and Keith Morley and Austin Robison and Martin Stich , title =. 2010 , month =

    2010

  65. [65]

    Real-time concurrent linked list construction on the

    Yang, Jason C and Hensley, Justin and Gr. Real-time concurrent linked list construction on the

  66. [66]

    Wald, Ingo and Morrical, Nate and Zellmann, Stefan and Ma, Lei and Usher, Will and Huang, Tiejun and Pascucci, Valerio , title =. Proc. ACM Comput. Graph. Interact. Tech. , month = aug, articleno =. 2020 , issue_date =. doi:10.1145/3406179 , abstract =

    work page doi:10.1145/3406179 2020

  67. [67]

    arXiv preprint arXiv:2501.19196 , year=

    RaySplats: Ray Tracing based Gaussian Splatting , author=. arXiv preprint arXiv:2501.19196 , year=

    arXiv

  68. [68]

    arXiv preprint arXiv:2503.12284 , year=

    Byrski, Krzysztof and Wilczy. arXiv preprint arXiv:2503.12284 , year=

    arXiv

  69. [69]

    The Visual Computer , volume=

    Fast sorting for exact OIT of complex scenes , author=. The Visual Computer , volume=. 2014 , publisher=

    2014

  70. [70]

    Stopthepop: Sorted gaussian splatting for view-consistent real-time rendering , author=

  71. [71]

    Marigold: Repurposing diffusion-based image generators for monocular depth estimation , author=

  72. [72]

    Fine-tuning image-conditional diffusion models is easier than you think , author=

  73. [73]

    What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? , author=

  74. [74]

    Learning-based inverse rendering of complex indoor scenes with differentiable

    Zhu, Jingsen and Luan, Fujun and Huo, Yuchi and Lin, Zihao and Zhong, Zhihua and Xi, Dianbing and Wang, Rui and Bao, Hujun and Zheng, Jiaxiang and Tang, Rui , booktitle=SIGGRAPHAsia, year=. Learning-based inverse rendering of complex indoor scenes with differentiable

  75. [75]

    2012 , url=

    Physically-Based Shading at Disney , author=. 2012 , url=

    2012

  76. [76]

    2025 , url =

    Blender - a. 2025 , url =

    2025

  77. [77]

    EDGS: Eliminating Densification for Efficient Convergence of

    Kotovenko, Dmytro and Grebenkova, Olga and Ommer, Bj. EDGS: Eliminating Densification for Efficient Convergence of. arXiv preprint arXiv:2504.13204 , year=

    arXiv

  78. [78]

    1998 , publisher=

    Inverse rendering for computer graphics , author=. 1998 , publisher=

    1998

  79. [79]

    Srinivasan, Pratul P and Deng, Boyang and Zhang, Xiuming and Tancik, Matthew and Mildenhall, Ben and Barron, Jonathan T , booktitle=CVPR, year=

  80. [80]

    2021 , publisher=

    Zhang, Xiuming and Srinivasan, Pratul P and Deng, Boyang and Debevec, Paul and Freeman, William T and Barron, Jonathan T , journal=TOG, volume=. 2021 , publisher=

    2021

Showing first 80 references.