pith. sign in

arxiv: 2503.22676 · v5 · submitted 2025-03-28 · 💻 cs.CV

TranSplat: Instant Object Relighting in Gaussian Splatting via Spherical Harmonic Radiance Transfer

Boyang Tony Yu , Yanlin Jin , Yun He , Akshat Dave , Ravi Ramamoorthi , Guha Balakrishnan This is my paper

Pith reviewed 2026-05-22 22:01 UTC · model grok-4.3

classification 💻 cs.CV
keywords Gaussian Splattingobject relightingspherical harmonicsradiance transferenvironment mapsBRDF-freeself-shadowingpost-processing
0
0 comments X

The pith

A BRDF-free radiance transfer method relights objects in Gaussian Splatting instantly by modulating spherical harmonic coefficients with per-normal irradiance ratios from environment maps.

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

The paper presents TranSplat as a method for instant object relighting inside the Gaussian Splatting framework that avoids inverse rendering and material estimation entirely. It works by analytically modulating the spherical harmonic appearance coefficients of each 2D Gaussian surfel according to irradiance ratios computed between a source and target environment map. A specularity-aware dual-path transfer handles higher-order bands for glossy effects while a lightweight self-shadowing module adds occlusion in the SH domain. The entire process runs as post-processing with no retraining and finishes in under one second. If the approach holds, it supplies a lightweight route to change lighting on captured Gaussian scenes while preserving view-dependent appearance.

Core claim

TranSplat introduces a BRDF-free radiance transfer strategy that analytically modulates the spherical harmonic appearance coefficients of an object's 2D Gaussian surfels using per-normal irradiance ratios derived from source and target environment maps. It adds a specularity-aware dual-path SH transfer that adapts higher-order bands in the reflection domain and a lightweight SH-domain self-shadowing module that produces realistic occlusion without mesh raycasting. The method operates as a post-processing step requiring no additional Gaussian Splatting retraining and completes relighting in under one second while outperforming inverse-rendering and diffusion-based baselines on synthetic and实物

What carries the argument

The BRDF-free radiance transfer strategy that analytically modulates SH appearance coefficients of 2D Gaussian surfels using per-normal irradiance ratios derived from source and target environment maps.

Load-bearing premise

Radially symmetric BRDF approximations and the low-pass filtering of the spherical harmonic basis are sufficient to produce perceptually realistic renderings for glossy and complex materials.

What would settle it

A quantitative or visual comparison on a glossy object under a sharp lighting change where the TranSplat output deviates measurably from a ground-truth path-traced reference of the same geometry and materials.

Figures

Figures reproduced from arXiv: 2503.22676 by Akshat Dave, Boyang Tony Yu, Guha Balakrishnan, Ravi Ramamoorthi, Yanlin Jin, Yun He.

Figure 1
Figure 1. Figure 1: Example demonstrations of TranSplat relighting a LEGO bulldozer from a source environment (top left) to target en￾vironments (bottom left). We assume that 3D Gaussian Splatting (GS) [18] representations may be built for both the source and target scenes from multiple views (not shown here), and that the object may be extracted (segmented) and placed into the target scene with either automatic or manual sup… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of TranSplat. Given source and target scenes, TranSplat first fits a Gaussian Surfels [9] model to the source scene (S) using input images and object masks, and a standard 3D Gaussian Splatting model to the target scene (T). From each trained GS representation, TranSplat renders cube maps centered at the object’s location to estimate spatially varying environment maps (LS, LT ). Using these enviro… view at source ↗
Figure 3
Figure 3. Figure 3: Capture an explicit environment map directly from 3D Gaussian fields. We render six 90° views around a speci￾fied position to obtain a cubemap, which is then converted into an equirectangular environment map. The position corresponds to the object’s original location to sample the source lighting or the target location where the object will be inserted. 3.3. Lighting Estimation The incident illumination ra… view at source ↗
Figure 4
Figure 4. Figure 4: Conceptual schematic of the shading map computa￾tion for a simple scene. (a) The shading map measures the contri￾bution of each point in an environment map on each Gaussian sur￾fel j of the object (here, a bunny). The contribution is based on the angular alignment of the surfel’s normal n j with the point. In this example, Gaussian j would not be strongly affected by the point along u with orientation (θ, … view at source ↗
Figure 5
Figure 5. Figure 5: Relighting results of TranSplat across various environments. We use city and sunset environments as the source lighting conditions. In addition to three target environment with distinct color tones and directional lighting, we also include, in the left four columns, the results of swapping the city and sunset lighting conditions. TranSplat produces relighting results that closely match the GT shading and c… view at source ↗
Figure 6
Figure 6. Figure 6: Demonstration of Spatially-Variant Relighting and Dynamic Shadows (Sec. 3.4 & 3.5). (Left) The source object under ‘City’ lighting environment map. (Right) The object is relit using the ‘Garage’ environment map. Our method uses per-Gaussian shading maps (Sec. 3.4) to ensure the object’s appearance correctly responds to the lighting’s spatial directionality. Our shadow module (Sec. 3.5) derives its casting … view at source ↗
Figure 7
Figure 7. Figure 7: Relighting results using two Blender-generated (Dragon [33] and Tower) and two TensoIR [15] objects. We apply the same (HDR) source environment map (City) to all objects, and then relight them to three other environment maps: Sunset, Fireplace, and Forest. We recommend zooming in to inspect fine-grained shading detail. TranSplat achieves the best visual quality in terms of color and shading effects compare… view at source ↗
Figure 8
Figure 8. Figure 8: Sample results demonstrating TranSplat’s ability at object transfer into real scenes. Injecting objects into another scene without relighting renders incongruous appearances with unnatural color and illumination, yet with the addition of TranSplat’s relighting optimization, renderings appear more natural. Details are best observed in Supplementary videos [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Failure cases due to non-Lambertian effects. We present relighting results using TranSplat on two objects from the TensoIR [15], relit from the City source environment to Fireplace target. For non-specular regions such as the hotdog and branches of the Ficus, TranSplat deliver reasonable color tones. However, TranSplat fails to recover accurate colors for specular regions such as the plate and sauces near … view at source ↗
Figure 10
Figure 10. Figure 10: Example of TranSplat’s full pipeline extracting fine details of an object when fitting a GS model to a source scene. TranSplat takes input 2D masks, such as the one on the left (generated by SAM [19]), which do not capture fine details such as the flower tendrils. TranSplat learns to associate the tendrils with the rest of the vase during optimization due to passing object labels from parent to child Gaus… view at source ↗
Figure 11
Figure 11. Figure 11: Object extraction/segmentation results. This figure presents examples of 3DGS object segmentation, comparing TranSplat to two segmentation baselines [7, 8] [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Experiments on the choice of color sampling method. The original sphere was under the city environment condition, and we use TranSplat to relight it to match the appearance under the fireplace environment. However, uniform sampling leads to inaccuracies when converting the environment map to SH, causing larger color deviations and the appearance of unintended green hues. D. Scene-Level Relighting In [PIT… view at source ↗
Figure 13
Figure 13. Figure 13: The effect of normal quality on relighting with TranSplat. Removing the Omnidata [12] normal prior leads to noticeable holes in the predicted surfel object normals. This degradation significantly affects relighting, even though the source renderings are almost identical to the GT. Recap 59m 7s for GS training, lighting and material estimation Transplat 4m 45s for GS training, 5s for post-relighting Recap … view at source ↗
Figure 14
Figure 14. Figure 14: Experiments on indoor scene relighting. Although minor artifacts appear due to imperfect normal estimation and complex material properties, TranSplat still produces lighting effects that are largely consistent with inverse-rendering-based methods. Moreover, its runtime increases only modestly as the number of Gaussians grows. The two room scenes are from Mip-NeRF360 [2] and MuSHRoom [29] respectively, and… view at source ↗
Figure 15
Figure 15. Figure 15: Examples of TranSplat’s sampled environment maps within the 3DGS scenes. A cubemap-based capture followed by conversion to an equirectangular map is a straightforward and commonly used approach for generating panoramic images. We novelly apply this directly within a 3D Gaussian Splatting scene, which allows the captured environment map to vary with the sampling location. After converting the map into sphe… view at source ↗
read the original abstract

We present TranSplat, a method for instant, accurate object relighting within the Gaussian Splatting (GS) framework. Rather than relying on costly inverse rendering routines, we propose a BRDF-free radiance transfer strategy that analytically modulates the spherical harmonic (SH) appearance coefficients of an object's 2D Gaussian surfels using per-normal irradiance ratios derived from source and target environment maps. To handle view-dependent and glossy appearances without explicit material estimation, we introduce a specularity-aware dual-path SH transfer strategy that adapts higher-order SH bands in the reflection domain. Additionally, we propose a lightweight SH-domain self-shadowing module to ensure physically realistic occlusion without explicit mesh raycasting. Operating as a post-processing step, TranSplat requires no additional GS retraining for a pair of source and target scenes. Evaluations on synthetic and real-world objects demonstrate state-of-the-art accuracy, outperforming recent inverse-rendering and diffusion-based GS relighting methods across most conditions, all while completing relighting operations in under one second. Although bounded by radially symmetric BRDF approximations and the low-pass nature of the SH basis, TranSplat produces perceptually realistic renderings even for glossy, complex materials, establishing a valuable, lightweight path forward for GS relighting.

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 / 1 minor

Summary. The paper claims to present TranSplat, a post-processing method for instant object relighting in Gaussian Splatting. It introduces a BRDF-free radiance transfer strategy that analytically modulates the spherical harmonic (SH) appearance coefficients of an object's 2D Gaussian surfels using per-normal irradiance ratios derived from source and target environment maps. A specularity-aware dual-path SH transfer strategy adapts higher-order bands for view-dependent glossy effects, and a lightweight SH-domain self-shadowing module handles occlusion without mesh raycasting. The method requires no GS retraining and claims state-of-the-art accuracy on synthetic and real-world objects while completing operations in under one second, producing perceptually realistic results for glossy materials despite radially symmetric BRDF approximations and SH low-pass limits.

Significance. If the analytical SH modulation and dual-path transfer hold without hidden parameters, the work would offer a significant lightweight alternative to inverse-rendering or diffusion-based GS relighting, enabling fast post-process relighting for AR/VR and content pipelines without retraining.

major comments (2)
  1. [Abstract] Abstract: The central claim of SOTA accuracy and perceptual realism for glossy, complex materials rests on the 'radially symmetric BRDF approximations' and analytical per-normal irradiance ratio modulation, but no equations, basis conversions, normalization details, or quantitative metrics (e.g., error on glossy objects) are provided, leaving the claim unverifiable and the low-pass SH sufficiency untested.
  2. [Abstract] Abstract: The 'specularity-aware dual-path SH transfer strategy' and 'lightweight SH-domain self-shadowing module' are described as load-bearing components for handling view-dependent effects and realistic occlusion, yet no derivation, pseudocode, or handling of higher-order SH bands is available to assess whether they avoid material parameters or explicit raycasting as stated.
minor comments (1)
  1. [Abstract] Abstract: The term '2D Gaussian surfels' is introduced without clarifying its relation to standard 3D Gaussian primitives or how surfel normals are obtained for the irradiance ratios.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the comments on the abstract. The points raised correctly note that the abstract is a high-level summary and does not contain the supporting equations, derivations, or metrics. Since only the abstract is available to us in this context, we cannot supply the requested technical details here.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of SOTA accuracy and perceptual realism for glossy, complex materials rests on the 'radially symmetric BRDF approximations' and analytical per-normal irradiance ratio modulation, but no equations, basis conversions, normalization details, or quantitative metrics (e.g., error on glossy objects) are provided, leaving the claim unverifiable and the low-pass SH sufficiency untested.

    Authors: The abstract is intentionally concise and omits mathematical derivations and quantitative results. Without access to the full manuscript, we cannot provide the irradiance ratio equations, basis conversions, normalization steps, or specific error metrics on glossy objects. We agree that the abstract alone does not allow verification of the SOTA claims or the sufficiency of low-order SH. revision: no

  2. Referee: [Abstract] Abstract: The 'specularity-aware dual-path SH transfer strategy' and 'lightweight SH-domain self-shadowing module' are described as load-bearing components for handling view-dependent effects and realistic occlusion, yet no derivation, pseudocode, or handling of higher-order SH bands is available to assess whether they avoid material parameters or explicit raycasting as stated.

    Authors: The abstract summarizes these components at a high level but provides no derivations or pseudocode. Since the full text is not available, we cannot supply the dual-path derivation, handling of higher-order bands, or the SH-domain self-shadowing formulation to demonstrate the absence of material parameters and raycasting. revision: no

standing simulated objections not resolved
  • Full derivations, equations, pseudocode, and quantitative metrics for the radiance transfer, dual-path strategy, and self-shadowing module are not present in the provided manuscript excerpt (limited to the abstract), so the specific technical claims cannot be substantiated or revised in this response.

Circularity Check

0 steps flagged

No circularity detectable; abstract presents analytical modulation from external maps with no self-referential reductions

full rationale

Only the abstract is available, which describes the core method as 'a BRDF-free radiance transfer strategy that analytically modulates the spherical harmonic (SH) appearance coefficients of an object's 2D Gaussian surfels using per-normal irradiance ratios derived from source and target environment maps' and a 'specularity-aware dual-path SH transfer strategy'. No equations, derivations, fitted parameters, or self-citations are provided. No load-bearing step reduces by construction to inputs, self-definition, or prior author work. The description is self-contained as direct analytical processing of external environment maps, consistent with a non-circular claim. Abstract-only access prevents deeper inspection but does not create evidence of circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

Abstract-only review; limited visibility into parameters or background assumptions. Inferred components listed below.

axioms (2)
  • domain assumption Radially symmetric BRDF approximations suffice for target objects
    Explicitly stated as a bounding limitation in the abstract.
  • domain assumption Higher-order SH bands in the reflection domain can be adapted without explicit material parameters
    Core premise of the dual-path strategy described in the abstract.
invented entities (2)
  • specularity-aware dual-path SH transfer strategy no independent evidence
    purpose: Handle view-dependent and glossy appearances without explicit material estimation
    New component introduced in the abstract.
  • lightweight SH-domain self-shadowing module no independent evidence
    purpose: Ensure physically realistic occlusion without explicit mesh raycasting
    New component introduced in the abstract.

pith-pipeline@v0.9.0 · 5742 in / 1665 out tokens · 72452 ms · 2026-05-22T22:01:35.904780+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.

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

39 extracted references · 39 canonical work pages

  1. [1]

    Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields

    Jonathan T Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 5855–5864,

    work page

  2. [2]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. 2022. 3, 5, 7, 1, 2

    work page 2022

  3. [3]

    Lambertian reflectance and linear subspaces.IEEE transactions on pattern analysis and machine intelligence, 25(2):218–233, 2003

    Ronen Basri and David W Jacobs. Lambertian reflectance and linear subspaces.IEEE transactions on pattern analysis and machine intelligence, 25(2):218–233, 2003. 4

    work page 2003

  4. [4]

    Pearson Educa- tion, 2014

    Jeremy Birn.Digital lighting & rendering. Pearson Educa- tion, 2014. 2

    work page 2014

  5. [5]

    Image-based render- ing of diffuse, specular and glossy surfaces from a single im- age

    Samuel Boivin and Andr ´e Gagalowicz. Image-based render- ing of diffuse, specular and glossy surfaces from a single im- age. InProceedings of the 28th annual conference on Com- puter graphics and interactive techniques, pages 107–116,

    work page

  6. [6]

    Nerd: Neural reflectance decomposition from image collections

    Mark Boss, Raphael Braun, Varun Jampani, Jonathan T Bar- ron, Ce Liu, and Hendrik Lensch. Nerd: Neural reflectance decomposition from image collections. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 12684–12694, 2021. 3

    work page 2021

  7. [7]

    Segment any 3d gaussians.arXiv preprint arXiv:2312.00860, 2023

    Jiazhong Cen, Jiemin Fang, Chen Yang, Lingxi Xie, Xi- aopeng Zhang, Wei Shen, and Qi Tian. Segment any 3d gaussians.arXiv preprint arXiv:2312.00860, 2023. 2

    work page arXiv 2023

  8. [8]

    Gaussianeditor: Swift and control- lable 3d editing with gaussian splatting

    Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xi- aofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and control- lable 3d editing with gaussian splatting. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 21476–21485, 2024. 1, 2

    work page 2024

  9. [9]

    High-quality surface recon- struction using gaussian surfels

    Pinxuan Dai, Jiamin Xu, Wenxiang Xie, Xinguo Liu, Huamin Wang, and Weiwei Xu. High-quality surface recon- struction using gaussian surfels. InACM SIGGRAPH 2024 Conference Papers. Association for Computing Machinery,

    work page 2024

  10. [10]

    Estimating surface reflectance properties of a complex scene under cap- tured natural illumination.Conditionally Accepted to ACM Transactions on Graphics, 19:2, 2004

    Paul Debevec, Chris Tchou, Andrew Gardner, Tim Hawkins, Charis Poullis, Jessi Stumpfel, Andrew Jones, Nathaniel Yun, Per Einarsson, Therese Lundgren, et al. Estimating surface reflectance properties of a complex scene under cap- tured natural illumination.Conditionally Accepted to ACM Transactions on Graphics, 19:2, 2004. 2

    work page 2004

  11. [11]

    A frequency analysis of light trans- port.ACM Transactions on Graphics (TOG), 24(3):1115– 1126, 2005

    Fr ´edo Durand, Nicolas Holzschuch, Cyril Soler, Eric Chan, and Franc ¸ois X Sillion. A frequency analysis of light trans- port.ACM Transactions on Graphics (TOG), 24(3):1115– 1126, 2005. 2

    work page 2005

  12. [12]

    Omnidata: A scalable pipeline for making multi- task mid-level vision datasets from 3d scans

    Ainaz Eftekhar, Alexander Sax, Jitendra Malik, and Amir Zamir. Omnidata: A scalable pipeline for making multi- task mid-level vision datasets from 3d scans. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 10786–10796, 2021. 1, 3

    work page 2021

  13. [13]

    Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray trac- ing.arXiv:2311.16043, 2023

    Jian Gao, Chun Gu, Youtian Lin, Hao Zhu, Xun Cao, Li Zhang, and Yao Yao. Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray trac- ing.arXiv:2311.16043, 2023. 2, 3, 5, 7

    work page arXiv 2023

  14. [14]

    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, pages 5322–5332, 2024. 2, 3, 5

    work page 2024

  15. [15]

    Tensoir: Tensorial inverse rendering

    Haian Jin, Isabella Liu, Peijia Xu, Xiaoshuai Zhang, Song- fang Han, Sai Bi, Xiaowei Zhou, Zexiang Xu, and Hao Su. Tensoir: Tensorial inverse rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2, 3, 5, 7, 8

    work page 2023

  16. [16]

    Neural gaffer: Relighting any object via diffusion

    Haian Jin, Yuan Li, Fujun Luan, Yuanbo Xiangli, Sai Bi, Kai Zhang, Zexiang Xu, Jin Sun, and Noah Snavely. Neural gaffer: Relighting any object via diffusion. InAdvances in Neural Information Processing Systems, 2024. 2, 3, 5, 7

    work page 2024

  17. [17]

    James T. Kajiya. The rendering equation. page 4, 1986. 4

    work page 1986

  18. [18]

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

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42 (4), 2023. 1, 2, 3, 4

    work page 2023

  19. [19]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. InProceedings of the IEEE/CVF international confer- ence on computer vision, pages 4015–4026, 2023. 1

    work page 2023

  20. [20]

    Recap: Better gaussian relighting with cross-environment captures

    Jingzhi Li, Zongwei Wu, Eduard Zamfir, and Radu Timofte. Recap: Better gaussian relighting with cross-environment captures. InCVPR, 2025. 2, 3, 5

    work page 2025

  21. [21]

    Crowdsampling the plenoptic function

    Zhengqi Li, Wenqi Xian, Abe Davis, and Noah Snavely. Crowdsampling the plenoptic function. InComputer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, Au- gust 23–28, 2020, Proceedings, Part I 16, pages 178–196. Springer, 2020. 3

    work page 2020

  22. [22]

    Gs-ir: 3d gaussian splatting for inverse rendering.arXiv preprint arXiv:2311.16473, 2023

    Zhihao Liang, Qi Zhang, Ying Feng, Ying Shan, and Kui Jia. Gs-ir: 3d gaussian splatting for inverse rendering.arXiv preprint arXiv:2311.16473, 2023. 2, 3, 5, 6, 7

    work page arXiv 2023

  23. [23]

    Single-image hdr reconstruction by learning to reverse the camera pipeline

    Yu-Lun Liu, Wei-Sheng Lai, Yu-Sheng Chen, Yi-Lung Kao, Ming-Hsuan Yang, Yung-Yu Chuang, and Jia-Bin Huang. Single-image hdr reconstruction by learning to reverse the camera pipeline. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 1651–1660, 2020. 4

    work page 2020

  24. [24]

    A theory of spherical harmonic identities for brdf/lighting transfer and image consistency

    Dhruv Mahajan, Ravi Ramamoorthi, and Brian Curless. A theory of spherical harmonic identities for brdf/lighting transfer and image consistency. InComputer Vision – ECCV 2006, pages 41–55. Springer, Heidelberg, 2006. 2, 3, 4

    work page 2006

  25. [25]

    Cornell University, 1998

    Stephen Robert Marschner.Inverse rendering for computer graphics. Cornell University, 1998. 2

    work page 1998

  26. [26]

    Nerf in the wild: Neural radiance fields for uncon- strained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM 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 recognition, pages 7210–7219, 2021. 3

    work page 2021

  27. [27]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

    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. 3

    work page 2021

  28. [28]

    A signal-processing framework for inverse rendering

    Ravi Ramamoorthi and Pat Hanrahan. A signal-processing framework for inverse rendering. InProceedings of the 28th annual conference on Computer graphics and interactive techniques, pages 117–128, 2001. 2, 4

    work page 2001

  29. [29]

    Mushroom: Multi-sensor hy- brid room dataset for joint 3d reconstruction and novel view synthesis

    Xuqian Ren, Wenjia Wang, Dingding Cai, Tuuli Tuominen, Juho Kannala, and Esa Rahtu. Mushroom: Multi-sensor hy- brid room dataset for joint 3d reconstruction and novel view synthesis. InProceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision, pages 4508–4517,

    work page

  30. [30]

    Relightable gaussian codec avatars

    Shunsuke Saito, Gabriel Schwartz, Tomas Simon, Junxuan Li, and Giljoo Nam. Relightable gaussian codec avatars. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 130–141, 2024. 2, 3

    work page 2024

  31. [31]

    Single-shot neural relighting and svbrdf estimation

    Shen Sang and Manmohan Chandraker. Single-shot neural relighting and svbrdf estimation. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part XIX 16, pages 85–101. Springer,

    work page 2020

  32. [32]

    Nerv: Neural reflectance and visibility fields for relighting and view synthesis

    Pratul P Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T Barron. Nerv: Neural reflectance and visibility fields for relighting and view synthesis. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 7495–7504, 2021. 3

    work page 2021

  33. [33]

    Using shape to categorize: Low-shot learning with an explicit shape bias

    Stefan Stojanov, Anh Thai, and James M Rehg. Using shape to categorize: Low-shot learning with an explicit shape bias. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 1798–1808, 2021. 5, 7

    work page 2021

  34. [34]

    Neural-pbir reconstruction of shape, material, and illumination, 2024

    Cheng Sun, Guangyan Cai, Zhengqin Li, Kai Yan, Cheng Zhang, Carl Marshall, Jia-Bin Huang, Shuang Zhao, and Zhao Dong. Neural-pbir reconstruction of shape, material, and illumination, 2024. 3

    work page 2024

  35. [35]

    Neilf: Neural incident light field for physically-based mate- rial estimation

    Yao Yao, Jingyang Zhang, Jingbo Liu, Yihang Qu, Tian Fang, David McKinnon, Yanghai Tsin, and Long Quan. Neilf: Neural incident light field for physically-based mate- rial estimation. InEuropean conference on computer vision, pages 700–716. Springer, 2022. 3

    work page 2022

  36. [36]

    Inverserendernet: Learn- ing single image inverse rendering

    Ye Yu and William AP Smith. Inverserendernet: Learn- ing single image inverse rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3155–3164, 2019. 3

    work page 2019

  37. [37]

    Neilf++: Inter-reflectable light fields for geometry and material esti- mation

    Jingyang Zhang, Yao Yao, Shiwei Li, Jingbo Liu, Tian Fang, David McKinnon, Yanghai Tsin, and Long Quan. Neilf++: Inter-reflectable light fields for geometry and material esti- mation. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 3601–3610, 2023. 3

    work page 2023

  38. [38]

    Ner- factor: Neural factorization of shape and reflectance under an unknown illumination.ACM Transactions on Graphics (ToG), 40(6):1–18, 2021

    Xiuming Zhang, Pratul P Srinivasan, Boyang Deng, Paul De- bevec, William T Freeman, and Jonathan T Barron. Ner- factor: Neural factorization of shape and reflectance under an unknown illumination.ACM Transactions on Graphics (ToG), 40(6):1–18, 2021. 3

    work page 2021

  39. [39]

    densification

    Xiaoming Zhao, Pratul P. Srinivasan, Dor Verbin, Keunhong Park, Ricardo Martin Brualla, and Philipp Henzler. Illumin- erf: 3d relighting without inverse rendering, 2024. 3 Supplementary Material for TranSplat A. Object Segmentation The full workflow ofTranSplatincludes a segmentation step to extract the Gaussian objects from the source scene. For instance...

    work page 2024