Pith. sign in

REVIEW 2 major objections 5 minor 64 references

HumanOLAT introduces the first publicly accessible large-scale dataset of multi-view OLAT captures of full-body humans, with roughly 850K frames that enable physically correct relighting under arbitrary illumination.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

HumanOLAT is the first public full-body OLAT dataset: 21 subjects, 3 poses, 40 views, 331 single-light captures, plus environment maps, color gradients, meshes and normals.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Genuinely useful full-body OLAT dataset; the motion-compensation and radiometric-linearity gaps are real but the dataset's value survives them. the 2 major comments →

arxiv 2508.09137 v1 pith:ECDFYPM4 submitted 2025-08-12 cs.CV

HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis

Timo Teufel , Pulkit Gera , Xilong Zhou , Umar Iqbal , Pramod Rao , Jan Kautz , Vladislav Golyanik , Christian Theobalt This is my paper
classification cs.CV
keywords full-body relightingOLAT datasetlightstage capturenovel-view synthesisphotometric normalsimage-based relightingmulti-view captureillumination harmonization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

HumanOLAT is a dataset paper whose central claim is that it provides the first publicly accessible large-scale multi-view One-Light-at-a-Time (OLAT) captures of full-body humans: 21 subjects, three poses each, 40 synchronized views, 331 individually controlled LED illuminations, ten environment-map illuminations, color gradients, and white light — roughly 850K HDR frames along with calibrations, meshes, masks, normals, and pose/SMPL-X annotations. The reason this matters is that light transport is linear, so any OLAT frame set can be combined additively to synthesize physically correct relighting under an arbitrary target environment (the paper's Eq. (2)). Prior public data covered faces, hands, objects, or bodies with only a few illuminations; full-body OLAT ground truth has been missing. The paper also evaluates four inverse-rendering methods on the dataset and reports that even the best reaches only 30.04 PSNR while remaining blurry and missing specular highlights and sharp shadows, positioning HumanOLAT as a stress test for current relighting.

Core claim

The core discovery is the dataset itself. HumanOLAT records 21 diverse subjects in three static poses through 40 cameras inside a lightstage with 331 individually controllable RGBAW LEDs, capturing white-light, color-gradient, ten environment-map, and 331 OLAT illuminations. After photogrammetric calibration and multi-view stereo mesh reconstruction, the pipeline estimates pixel-wise photometric normals from color gradients and makes all OLAT frames pixel-aligned to a reference white-light frame via sparse point tracking and interpolated warp (motion compensation). Because the camera and light positions are calibrated, any subset of OLAT frames can be weighted by the target environment and s

What carries the argument

The load-bearing mechanism is image-based relighting from OLAT captures, formalized by the paper's reference [11]: light transport is linear, so a target illumination is obtained by masking the environment map with each single-light mask, averaging per-channel to get weights $c_i$, and summing the motion-corrected OLAT frames $I_i$: $I_{\mathrm{target}}=\sum_i c_i I_i$. To make that sum valid on moving subjects, the pipeline injects a white-light tracking frame every 21st OLAT frame, tracks about 12,000 sparse grid points with a point-tracking model, linearly interpolates dense flow, and warps every OLAT frame to the reference frame. The dataset's utility rests on that alignment: uncorrected

Load-bearing premise

Subjects sway during the roughly 11-second capture, and the pipeline assumes that tracking about 12,000 sparse points and linearly interpolating their flow removes that sway well enough that every OLAT frame is pixel-aligned to the reference frame; the paper validates this only qualitatively.

What would settle it

For any environment map in the dataset, compare the actual captured environment-map frame with the OLAT-summed image for the same lighting computed from Eq. (2), and compute per-pixel residuals on the body (excluding sensor noise). If residual error is large in regions with sharp edges, specular highlights, or self-shadow boundaries, the motion compensation has not achieved pixel alignment; the paper reports no such quantitative residual, only the qualitative visual in its Figure 5.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A community benchmark for full-body relighting and novel-view synthesis now has real OLAT ground truth, not synthetic or single-illumination data.
  • Any method that assumes known lighting can train on 331 single-light observations per view and test on held-out views and illuminations.
  • The dataset's environment-map captures provide a direct check on the linear-combination assumption: image-based relighting from OLAT should reproduce them.
  • Current inverse-rendering baselines plateau around 30 PSNR and fail on specular and shadow effects, so the dataset defines a concrete target for improvement.
  • The released masks, photometric normals, MVS meshes, and pose/SMPL-X annotations extend the same captures to human avatar and illumination-harmonization research.

Where Pith is reading between the lines

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

  • If the motion compensation is as clean as claimed, the same capture pipeline could plausibly be extended to sequences with deliberate motion, but the sparse-flow interpolation would need validation against dense ground truth before that extension is trustworthy.
  • The environment-map versus OLAT-sum consistency check described in the falsifier could be run by the dataset authors as a quantitative quality metric; the paper currently reports only qualitative evidence for alignment.
  • Because 16 of the 21 subjects wear loose clothing, the dataset stresses self-shadowing and cloth-related artifacts more than face or hand datasets, making it a harder and more realistic benchmark for material-aware relighting.
  • The 331 finely spaced OLAT directions could support learning-based relighting priors that interpolate between lights, potentially reducing the capture cost for future full-body OLAT datasets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper introduces HumanOLAT, a multi-view, multi-illumination lightstage dataset of 21 full-body subjects in three poses, captured by 40 cameras under 331 one-light-at-a-time (OLAT) illuminations, 10 environment maps, white light, and color gradients, totaling roughly 850K frames. The release includes camera and light calibration, MVS meshes, segmentation masks, photometric normals, OpenPose, and SMPL-X annotations. The central claim is that, because OLAT images are linear in light transport, arbitrary target illumination can be synthesized as a weighted sum of the OLAT frames (Eq. 2), making the dataset physically correct ground truth for full-body relighting and a benchmark for relighting methods. The authors evaluate four Gaussian-based relighting baselines (PRT-Gaussian, GS3, RNG, BiGS) and IC-Light, finding that current methods underperform on full-body scenes (GS3 is best at 30.04 dB PSNR but still blurry), which supports the dataset's utility as a benchmark.

Significance. If the released data live up to the description, HumanOLAT would fill a genuine gap: existing public lightstage datasets cover objects, faces, or hands, and Ultrastage, the most similar full-body dataset, lacks OLAT illumination. The authors are appropriately transparent about known limitations, such as the ~20 hatch LEDs with uncertain positions, and provide a reasonable geometric calibration validation at 0.819 px reprojection error. The use of standard light-transport identities from Debevec et al. [11] is not circular. The baseline evaluation is a useful stress test showing that state-of-the-art inverse rendering methods struggle on full-body OLAT data. However, the manuscript does not currently establish the key radiometric precondition for Eq. (2), and the motion-compensation validation is only qualitative. These are load-bearing gaps for the core claim of physically correct relighting ground truth.

major comments (2)
  1. [Sec. 3.3.3 / Eq. (2)] The abstract and Sec. 3.2 state that HumanOLAT contains 'HDR RGB frames,' and Eq. (2) presents relit images as a linear combination of OLAT frames. This is only valid if every stored frame is proportional to incident radiance. However, the processing pipeline in Sec. 3.3 describes geometric calibration, mesh reconstruction, mask generation, and motion compensation, but no radiometric calibration: no camera response function estimation, no RAW-to-linear conversion, no exposure/HDR merging, no bit-depth or file-format specification, and no validation that pixel values are linear. If the released images are gamma-encoded or clipped, Eq. (2) does not produce physically correct relit pixels, and Eq. (1) photometric normals would also be biased. Since the dataset is access-gated, the manuscript itself must supply this evidence. Please add the full radiometric pipeline and a quantitative linear
  2. [Sec. 3.3.4 / Eq. (2)] Motion compensation is essential to the dataset's validity because Eq. (2) sums many OLAT frames. The current procedure tracks only ~12k sparse grid points with CoTracker3 on white-light frames injected every 21st OLAT frame and linearly interpolates to dense flow, yet the only validation is the qualitative side-by-side in Fig. 5. No residual error metric is reported. Residual sway therefore directly corrupts the relit ground truth. Please provide a quantitative evaluation of residual alignment error, for example on a rigid static object or via flow reprojection consistency, and report the expected worst-case residual.
minor comments (5)
  1. [Eq. (1)] The formula for photometric normals is written as 'n = d |d|' but should be 'n = d / |d|'. Please correct the notation.
  2. [Sec. 4.1] The baseline protocol uses six 'representative' captures, selects 100 lights from 32 cameras, and downscales to 1K, and the images are 'empirically brightened by a factor of 10.' Please state how the six captures were selected, provide the exact split (which subjects/poses/views/lights), and clarify whether the brightening factor is only a training convenience or reflects a property of the released frame values.
  3. [Fig. 4] Caption typo: 'one of the40 frames' should be 'one of the 40 frames'.
  4. [Table 3 caption] Typo: 'relighting methonds' should be 'relighting methods'.
  5. [Sec. 3.3.4] The weighting color c_i in Eq. (2) is described only as 'masking Etarget with each OLAT environment mask Ei and subsequent per-channel averaging.' Please define the OLAT environment masks precisely and state whether the LED angular response is included; otherwise, the accuracy of the weights is unclear.

Circularity Check

0 steps flagged

No circularity: Eq. (1) and Eq. (2) are standard linear-light transport identities from cited prior work, and the benchmark evaluations use external methods.

full rationale

The paper's derivation chain consists of Eq. (1) for photometric normals from color-gradient illumination and Eq. (2) for image-based relighting by per-channel linear combination of OLAT frames. Both are standard physical identities taken from prior work [11, 17, 63]; neither is defined in terms of this paper's own outputs, and no parameter is fitted from the data and then presented as a prediction. The normals are computed directly from captured gradient images, and the relighting weights ci are computed from the target environment map and known OLAT masks. The baseline experiments use externally released methods (GS3, RNG, BiGS, PRT-Gaussian, IC-Light) and standard metrics, so the benchmark results are not forced by construction. Self-citations such as [19] and [36] are contextual only and do not carry the central argument. Potential concerns raised by a reader — the absence of a described radiometric calibration/HDR linearization pipeline and the qualitative-only validation of motion compensation — are data-quality or support issues, not circularity: the manuscript does not define Eq. (2) in terms of the released frames in a way that makes the claim true by construction. No circular step can be quoted, so the appropriate score is 0.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim rests on standard capture and calibration assumptions rather than fitted models. The free parameters listed are hand-chosen constants in the capture and evaluation pipeline; none is fitted to make a prediction appear correct. The axioms are standard light-transport and photometric identities plus a staticity assumption that the paper itself flags as imperfect (Sec. 3.3.3). No invented physical entities are introduced.

free parameters (3)
  • OLAT frame brightening factor = 10
    Sec. 4.1: frames are 'empirically brighten[ed] ... by a factor of 10 to strengthen the training signal'. Hand-chosen; affects baseline training, not dataset content.
  • White-light tracking frame injection interval = every 21st OLAT frame
    Sec. 3.3.3: design choice for motion compensation; affects optical-flow accuracy.
  • Sparse tracking point count = ~12k grid points
    Sec. 3.3.3: chosen to cap computational cost; density of interpolation affects the fidelity of motion compensation.
axioms (4)
  • domain assumption Light transport in the capture is linear, so OLAT frames can be weighted and summed per Eq. (2) to synthesize any target environment map.
    Sec. 3.3.4 and Sec. 2.1, following Debevec et al. [11]. Standard in image-based relighting; holds for fixed geometry with linear light sources, and is the load-bearing identity for the dataset's relighting ground truth.
  • domain assumption Subjects are sufficiently static across the ~11 s capture once the sparse-flow motion compensation is applied.
    Sec. 3.3.3: the paper acknowledges sway and compensates with ~12k tracked points (CoTracker3) plus linear interpolation. Residual motion would corrupt Eq. (2); the paper validates only qualitatively (Fig. 5). This is the weakest assumption.
  • domain assumption Pixel-wise normals are recoverable from two color-gradient captures via n = (g+ - g-)/(g+ + g-) normalized.
    Sec. 3.3.4, Eq. (1), following Guo et al. [17] and Zhou et al. [63]; assumes color-channel encoding of three gradient axes and diffuse-dominated response.
  • domain assumption A single Metashape camera calibration taken from the A-pose is valid for the other two poses of the same subject.
    Sec. 3.3.1: features are matched on white-light frames and one calibration per subject is used across poses, assuming the camera rig is rigid during the capture session.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis." pith.science (2026). https://pith.science/paper/ECDFYPM4

@misc{pith2026250809137,
  author       = {Pith},
  title        = {Pith review of: HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECDFYPM4}},
  note         = {Machine review of arXiv:2508.09137}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Simultaneous relighting and novel-view rendering of digital human representations is an important yet challenging task with numerous applications. Progress in this area has been significantly limited due to the lack of publicly available, high-quality datasets, especially for full-body human captures. To address this critical gap, we introduce the HumanOLAT dataset, the first publicly accessible large-scale dataset of multi-view One-Light-at-a-Time (OLAT) captures of full-body humans. The dataset includes HDR RGB frames under various illuminations, such as white light, environment maps, color gradients and fine-grained OLAT illuminations. Our evaluations of state-of-the-art relighting and novel-view synthesis methods underscore both the dataset's value and the significant challenges still present in modeling complex human-centric appearance and lighting interactions. We believe HumanOLAT will significantly facilitate future research, enabling rigorous benchmarking and advancements in both general and human-specific relighting and rendering techniques.

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

64 extracted references · 61 canonical work pages

  1. [1]

    Easymocap - make human motion capture easier. Github,

  2. [2]

    Metashape

    Agisoft LLC. Metashape. https://www.agisoft. com/downloads/installer/, 2025. Version retrieved July 2025. 4

    work page 2025

  3. [3]

    Deep relightable appearance models for animatable faces

    Sai Bi, Stephen Lombardi, Shunsuke Saito, Tomas Simon, Shih-En Wei, Kevyn Mcphail, Ravi Ramamoorthi, Yaser Sheikh, and Jason Saragih. Deep relightable appearance models for animatable faces. ACM Trans. Graph. , 40(4),

    work page

  4. [4]

    Gs3: Efficient relighting with triple gaussian splatting

    Zoubin Bi, Yixin Zeng, Chong Zeng, Fan Pei, Xiang Feng, Kun Zhou, and Hongzhi Wu. Gs3: Efficient relighting with triple gaussian splatting. In SIGGRAPH Asia, 2024. 2, 3, 6, 7, 8, 12, 15

    work page 2024

  5. [5]

    Bar- ron, Ce Liu, and Hendrik P.A

    Mark Boss, Raphael Braun, Varun Jampani, Jonathan T. Bar- ron, Ce Liu, and Hendrik P.A. Lensch. Nerd: Neural re- flectance decomposition from image collections. In Interna- tional Conference on Computer Vision (ICCV), 2021. 2

    work page 2021

  6. [6]

    Openpose: Realtime multi-person 2d pose estimation using part affinity fields

    Zhe Cao, Gines Hidalgo, Tomas Simon, Shih-En Wei, and Yaser Sheikh. Openpose: Realtime multi-person 2d pose estimation using part affinity fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019. 4, 12, 13

    work page 2019

  7. [7]

    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. In International Conference on Learning Repre- sentations (ICLR), 2025. 3, 7

    work page 2025

  8. [8]

    Meshavatar: Learning high-quality triangular human avatars from multi-view videos

    Yushuo Chen, Zerong Zheng, Zhe Li, Chao Xu, and Yebin Liu. Meshavatar: Learning high-quality triangular human avatars from multi-view videos. In European Conference on Computer Vision (ECCV), 2024. 3, 7

    work page 2024

  9. [9]

    Relighting4d: Neural re- lightable human from videos

    Zhaoxi Chen and Ziwei Liu. Relighting4d: Neural re- lightable human from videos. In European Conference on Computer Vision (ECCV), 2022. 2

    work page 2022

  10. [10]

    URhand: Universal relightable hands

    Zhaoxi Chen, Gyeongsik Moon, Kaiwen Guo, Chen Cao, Stanislav Pidhorskyi, Tomas Simon, Rohan Joshi, Yuan Dong, Yichen Xu, Bernardo Pires, He Wen, Lucas Evans, Bo Peng, Julia Buffalini, Autumn Trimble, Kevyn McPhail, Melissa Schoeller, Shoou-I Yu, Javier Romero, Michael Zollh¨ofer, Yaser Sheikh, Ziwei Liu, and Shunsuke Saito. URhand: Universal relightable h...

    work page 2024

  11. [11]

    Acquiring the reflectance field of a human face

    Paul Debevec, Tim Hawkins, Chris Tchou, Westley Sarokin, and Mark Sagar. Acquiring the reflectance field of a human face. In SIGGRAPH, 2000. 1, 2, 3, 5, 6

    work page 2000

  12. [12]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In Computer Vision and Pattern Recognition (CVPR), 2022. 3

    work page 2022

  13. [13]

    Subsurface scat- tering for gaussian splatting

    Jan-Niklas Dihlmann, Arjun Majumdar, Andreas Engel- hardt, Raphael Braun, and Hendrik Lensch. Subsurface scat- tering for gaussian splatting. In Neural Information Process- ing Systems (NeurIPS), 2024. 2, 3, 7

    work page 2024

  14. [14]

    Gs-id: Illumina- tion decomposition on gaussian splatting via diffusion prior and parametric light source optimization

    Kang Du, Zhihao Liang, and Zeyu Wang. Gs-id: Illumina- tion decomposition on gaussian splatting via diffusion prior and parametric light source optimization. In International Conference on Computer Vision (ICCV), 2025. 2, 3, 7

    work page 2025

  15. [15]

    Rng: Relightable neural gaussians

    Jiahui Fan, Fujun Luan, Jian Yang, Milos Hasan, and Beibei Wang. Rng: Relightable neural gaussians. In Computer Vi- sion and Pattern Recognition (CVPR), 2025. 2, 6, 7, 8, 12, 15

    work page 2025

  16. [16]

    Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray trac- ing

    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. In European Conference on Computer Vision (ECCV),

    work page

  17. [17]

    The re- lightables: volumetric performance capture of humans with realistic relighting

    Kaiwen Guo, Peter Lincoln, Philip Davidson, Jay Busch, Xueming Yu, Matt Whalen, Geoff Harvey, Sergio Orts- Escolano, Rohit Pandey, Jason Dourgarian, Danhang Tang, Anastasia Tkach, Adarsh Kowdle, Emily Cooper, Ming- song Dou, Sean Fanello, Graham Fyffe, Christoph Rhemann, Jonathan Taylor, Paul Debevec, and Shahram Izadi. The re- lightables: volumetric perf...

    work page 2019

  18. [18]

    Diffre- light: Diffusion-based facial performance relighting

    Mingming He, Pascal Clausen, Ahmet Levent Tas ¸el, Li Ma, Oliver Pilarski, Wenqi Xian, Laszlo Rikker, Xueming Yu, Ryan Burgert, Ning Yu, and Paul Debevec. Diffre- light: Diffusion-based facial performance relighting. In SIG- GRAPH Asia, 2024. 3, 7

    work page 2024

  19. [19]

    Rana: Relightable artic- ulated neural avatars

    Umar Iqbal, Akin Caliskan, Koki Nagano, Sameh Khamis, Pavlo Molchanov, and Jan Kautz. Rana: Relightable artic- ulated neural avatars. In International Conference on Com- puter Vision (ICCV), 2023. 2, 3, 7

    work page 2023

  20. [20]

    Re- lightablehands: Efficient neural relighting of articulated hand models

    Shun Iwase, Shunsuke Saito, Tomas Simon, Stephen Lom- bardi, Timur Bagautdinov, Rohan Joshi, Fabian Prada, Takaaki Shiratori, Yaser Sheikh, and Jason Saragih. Re- lightablehands: Efficient neural relighting of articulated hand models. In Computer Vision and Pattern Recognition (CVPR), 2023. 3

    work page 2023

  21. [21]

    Geometry-aware single-image full-body human relight- ing

    Chaonan Ji, Tao Yu, Kaiwen Guo, Jingxin Liu, and Yebin Liu. Geometry-aware single-image full-body human relight- ing. In European Conference on Computer Vision (ECCV),

    work page

  22. [22]

    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. In Computer Vision and Pattern Recogni- tion (CVPR), 2024. 2, 3

    work page 2024

  23. [23]

    Relighting humans: occlusion-aware inverse rendering for full-body human im- ages

    Yoshihiro Kanamori and Yuki Endo. Relighting humans: occlusion-aware inverse rendering for full-body human im- ages. 37(6), 2018. 3, 7

    work page 2018

  24. [24]

    Co- tracker3: Simpler and better point tracking by pseudo- labelling real videos

    Nikita Karaev, Iurii Makarov, Jianyuan Wang, Natalia Neverova, Andrea Vedaldi, and Christian Rupprecht. Co- tracker3: Simpler and better point tracking by pseudo- labelling real videos. In International Conference on Com- puter Vision (ICCV), 2025. 5

    work page 2025

  25. [25]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1,

  26. [26]

    Sapiens: Foundation for human vision mod- els

    Rawal Khirodkar, Timur Bagautdinov, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik, Stuart Anderson, and 9 Shunsuke Saito. Sapiens: Foundation for human vision mod- els. In European Conference on Computer Vision (ECCV) ,

    work page

  27. [27]

    Switchlight: Co-design of physics- driven architecture and pre-training framework for human portrait relighting

    Hoon Kim, Minje Jang, Wonjun Yoon, Jisoo Lee, Donghyun Na, and Sanghyun Woo. Switchlight: Co-design of physics- driven architecture and pre-training framework for human portrait relighting. In Computer Vision and Pattern Recogni- tion (CVPR), 2024. 2, 3, 7

    work page 2024

  28. [28]

    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, et al. Segment anything. In International Conference on Computer Vision (ICCV), 2023. 5

    work page 2023

  29. [29]

    Olat gaussians for generic re- lightable appearance acquisition

    Zhiyi Kuang, Yanchao Yang, Siyan Dong, Jiayue Ma, Hongbo Fu, and Youyi Zheng. Olat gaussians for generic re- lightable appearance acquisition. In SIGGRAPH Asia, 2024. 2, 3, 7

    work page 2024

  30. [30]

    Uravatar: Universal relightable gaussian codec avatars

    Junxuan Li, Chen Cao, Gabriel Schwartz, Rawal Khirodkar, Christian Richardt, Tomas Simon, Yaser Sheikh, and Shun- suke Saito. Uravatar: Universal relightable gaussian codec avatars. In SIGGRAPH, 2024. 2, 3, 7

    work page 2024

  31. [31]

    Animatable and re- lightable gaussians for high-fidelity human avatar modeling

    Zhe Li, Yipengjing Sun, Zerong Zheng, Lizhen Wang, Shengping Zhang, and Yebin Liu. Animatable and re- lightable gaussians for high-fidelity human avatar modeling. arXiv preprint arXiv:2311.16096v4, 2024. 3, 7

    Pith/arXiv arXiv 2024

  32. [32]

    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. In Computer Vision and Pattern Recognition (CVPR), 2024. 2, 3, 7

    work page 2024

  33. [33]

    Gus-ir: Gaussian splatting with unified shading for inverse rendering

    Zhihao Liang, Hongdong Li, Kui Jia, Kailing Guo, and Qi Zhang. Gus-ir: Gaussian splatting with unified shading for inverse rendering. IEEE Transactions on Pattern Analysis and Machine Intelligence, PP:1–15, 2025. 2, 7

    work page 2025

  34. [34]

    Relightable and animatable neural avatars from videos

    Wenbin Lin, Chengwei Zheng, Jun-Hai Yong, and Feng Xu. Relightable and animatable neural avatars from videos. In Association for the Advancement of Artificial Intelligence (AAAI), 2024. 3, 7

    work page 2024

  35. [35]

    Openillumination: A multi-illumination dataset for inverse rendering evaluation on real objects

    Isabella Liu, Linghao Chen, Ziyang Fu, Liwen Wu, Ha- ian Jin, Zhong Li, Chin Ming Ryan Wong, Yi Xu, Ravi Ramamoorthi, Zexiang Xu, et al. Openillumination: A multi-illumination dataset for inverse rendering evaluation on real objects. In Neural Information Processing Systems (NeurIPS), 2023. 2, 3, 6, 7

    work page 2023

  36. [36]

    Relightable neu- ral actor with intrinsic decomposition and pose control

    Diogo Luvizon, Vladislav Golyanik, Adam Kortylewski, Marc Habermann, and Christian Theobalt. Relightable neu- ral actor with intrinsic decomposition and pose control. In European Conference on Computer Vision (ECCV), 2024. 2, 3, 7

    work page 2024

  37. [37]

    Rapid acquisition of specular and diffuse normal maps from polarized spher- ical gradient illumination

    Wan-Chun Ma, Tim Hawkins, Pieter Peers, Charles-Felix Chabert, Malte Weiss, and Paul Debevec. Rapid acquisition of specular and diffuse normal maps from polarized spher- ical gradient illumination. In Eurographics Symposium on Rendering (EGSR), 2007. 6, 7

    work page 2007

  38. [38]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In European Conference on Computer Vision (ECCV), 2020. 2

    work page 2020

  39. [39]

    Total relighting: learning to relight portraits for background replacement

    Rohit Pandey, Sergio Orts-Escolano, Chloe Legendre, Chris- tian Haene, Sofien Bouaziz, Christoph Rhemann, Paul E De- bevec, and Sean Ryan Fanello. Total relighting: learning to relight portraits for background replacement. ACM Trans. Graph., 40(4):43–1, 2021. 2, 3, 7

    work page 2021

  40. [40]

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black. Expressive body capture: 3D hands, face, and body from a single image. In Computer Vision and Pat- tern Recognition (CVPR). 4, 12

    work page

  41. [41]

    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. In International Conf...

    work page 2025

  42. [42]

    Relightable gaussian codec avatars

    Shunsuke Saito, Gabriel Schwartz, Tomas Simon, Junxuan Li, and Giljoo Nam. Relightable gaussian codec avatars. In Computer Vision and Pattern Recognition (CVPR), 2024. 2, 3, 6, 7

    work page 2024

  43. [43]

    Gir: 3d gaussian inverse rendering for relightable scene factorization

    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, 3, 7

    work page 2025

  44. [44]

    Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T

    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. In Computer Vision and Pattern Recog- nition (CVPR), 2021. 2

    work page 2021

  45. [45]

    Effect of illumination on automatic ex- pression recognition: A novel 3d relightable facial database

    Giota Stratou, Abhijeet Ghosh, Paul Debevec, and Louis- Philippe Morency. Effect of illumination on automatic ex- pression recognition: A novel 3d relightable facial database. In IEEE International Conference on Automatic Face & Ges- ture Recognition (FG), 2011. 2, 6, 7

    work page 2011

  46. [46]

    Barron, Yun-Ta Tsai, Zexiang Xu, Xueming Yu, Graham Fyffe, Christoph Rhemann, Jay Busch, Paul Debevec, and Ravi Ramamoorthi

    Tiancheng Sun, Jonathan T. Barron, Yun-Ta Tsai, Zexiang Xu, Xueming Yu, Graham Fyffe, Christoph Rhemann, Jay Busch, Paul Debevec, and Ravi Ramamoorthi. Single image portrait relighting. ACM Trans. Graph., 38(4), 2019. 3

    work page 2019

  47. [47]

    Relight my nerf: A dataset for novel view synthesis and re- lighting of real world objects

    Marco Toschi, Riccardo De Matteo, Riccardo Spezialetti, Daniele De Gregorio, Luigi Di Stefano, and Samuele Salti. Relight my nerf: A dataset for novel view synthesis and re- lighting of real world objects. In Computer Vision and Pat- tern Recognition (CVPR), 2023. 2, 6, 7

    work page 2023

  48. [48]

    Intrinsicavatar: Physically based inverse rendering of dynamic humans from monocular videos via explicit ray tracing

    Shaofei Wang, Bo ˇzidar Anti ´c, Andreas Geiger, and Siyu Tang. Intrinsicavatar: Physically based inverse rendering of dynamic humans from monocular videos via explicit ray tracing. In Computer Vision and Pattern Recognition (CVPR), 2024. 3, 6, 7, 12, 13

    work page 2024

  49. [49]

    10 Relightable full-body gaussian codec avatars

    Shaofei Wang, Tomas Simon, Igor Santesteban, Timur Bagautdinov, Junxuan Li, Vasu Agrawal, Fabian Prada, Shoou-I Yu, Pace Nalbone, Matt Gramlich, Roman Lubach- ersky, Chenglei Wu, Javier Romero, Jason Saragih, Michael Zollhoefer, Andreas Geiger, Siyu Tang, and Shunsuke Saito. 10 Relightable full-body gaussian codec avatars. In SIG- GRAPH, 2025. 3, 7

    work page 2025

  50. [50]

    Perfor- mance relighting and reflectance transformation with time- multiplexed illumination

    Andreas Wenger, Andrew Gardner, Chris Tchou, Jonas Unger, Tim Hawkins, and Paul Debevec. Perfor- mance relighting and reflectance transformation with time- multiplexed illumination. ACM Transactions on Graphics (TOG), 24:756–764, 2005. 5

    work page 2005

  51. [51]

    Deferredgs: Decoupled and re- lightable gaussian splatting with deferred shading

    Tong Wu, Jia-Mu Sun, Yu-Kun Lai, Yuewen Ma, Leif Kobbelt, and Lin Gao. Deferredgs: Decoupled and re- lightable gaussian splatting with deferred shading. IEEE Transactions on Pattern Analysis and Machine Intelligence, PP, 2025. 2, 3, 7

    work page 2025

  52. [52]

    Mvhumannet: A large-scale dataset of multi-view daily dressing human captures

    Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu, Chongjie Wang, Shijie Wang, Shuguang Cui, and Xiaoguang Han. Mvhumannet: A large-scale dataset of multi-view daily dressing human captures. In Computer Vision and Pattern Recognition (CVPR), 2024. 3

    work page 2024

  53. [53]

    Relightable and animatable neural avatar from sparse-view video

    Zhen Xu, Sida Peng, Chen Geng, Linzhan Mou, Zihan Yan, Jiaming Sun, Hujun Bao, and Xiaowei Zhou. Relightable and animatable neural avatar from sparse-view video. In Computer Vision and Pattern Recognition (CVPR), 2024. 3, 7

    work page 2024

  54. [54]

    Towards practical capture of high-fidelity relightable avatars

    Haotian Yang, Mingwu Zheng, Wanquan Feng, Haibin Huang, Yu-Kun Lai, Pengfei Wan, Zhongyuan Wang, and Chongyang Ma. Towards practical capture of high-fidelity relightable avatars. In SIGGRAPH Asia, 2023. 3

    work page 2023

  55. [55]

    Omg: Opacity matters in material model- ing with gaussian splatting

    Silong Yong, Venkata Nagarjun Pudureddiyur Manivannan, Bernhard Kerbl, Zifu Wan, Simon Stepputtis, Katia Sycara, and Yaqi Xie. Omg: Opacity matters in material model- ing with gaussian splatting. In International Conference on Learning Representations (ICLR), 2025. 2, 3, 7

    work page 2025

  56. [56]

    PhySG: Inverse rendering with spherical gaussians for physics-based material editing and relighting

    Kai Zhang, Fujun Luan, Qianqian Wang, Kavita Bala, and Noah Snavely. PhySG: Inverse rendering with spherical gaussians for physics-based material editing and relighting. In Computer Vision and Pattern Recognition (CVPR), 2021. 2

    work page 2021

  57. [57]

    Neural video portrait relighting in real-time via con- sistency modeling

    Longwen Zhang, Qixuan Zhang, Minye Wu, Jingyi Yu, and Lan Xu. Neural video portrait relighting in real-time via con- sistency modeling. InInternational Conference on Computer Vision (ICCV), 2021. 2, 6, 7

    work page 2021

  58. [58]

    Prtgaussian: Efficient relighting using 3d gaus- sians with precomputed radiance transfer

    Libo Zhang, Yuxuan Han, Wenbin Lin, Jingwang Ling, and Feng Xu. Prtgaussian: Efficient relighting using 3d gaus- sians with precomputed radiance transfer. In 2024 Asia Pa- cific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), 2024. 2, 3, 6, 7, 8, 12, 15

    work page 2024

  59. [59]

    Scal- ing in-the-wild training for diffusion-based illumination har- monization and editing by imposing consistent light trans- port

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Scal- ing in-the-wild training for diffusion-based illumination har- monization and editing by imposing consistent light trans- port. In International Conference on Learning Representa- tions (ICLR), 2025. 3, 7, 8, 12, 15

    work page 2025

  60. [60]

    Srinivasan, Boyang Deng, Paul Debevec, William T

    Xiuming Zhang, Pratul P. Srinivasan, Boyang Deng, Paul Debevec, William T. Freeman, and Jonathan T. Barron. Ner- factor: neural factorization of shape and reflectance under an unknown illumination. ACM Trans. Graph., 40(6), 2021. 2

    work page 2021

  61. [61]

    Physavatar: Learning the physics of dressed 3d avatars from visual observations

    Yang Zheng, Qingqing Zhao, Guandao Yang, Wang Yi- fan, Donglai Xiang, Florian Dubost, Dmitry Lagun, Thabo Beeler, Federico Tombari, Leonidas Guibas, and Gordon Wetzstein. Physavatar: Learning the physics of dressed 3d avatars from visual observations. In European Conference on Computer Vision (ECCV), 2024. 3, 7

    work page 2024

  62. [62]

    Bigs: Bidirectional gaussian primitives for re- lightable 3d gaussian splatting

    Liu Zhenyuan, Yu Guo, Xinyuan Li, Bernd Bickel, and Ran Zhang. Bigs: Bidirectional gaussian primitives for re- lightable 3d gaussian splatting. In International Conference on 3D Vision (3DV), 2025. 2, 3, 6, 7, 8, 12, 15

    work page 2025

  63. [63]

    Re- lightable neural human assets from multi-view gradient il- luminations

    Taotao Zhou, Kai He, Di Wu, Teng Xu, Qixuan Zhang, Kuix- iang Shao, Wenzheng Chen, Lan Xu, and Jingyi Yu. Re- lightable neural human assets from multi-view gradient il- luminations. In Computer Vision and Pattern Recognition (CVPR), 2023. 2, 3, 4, 5, 6, 7

    work page 2023

  64. [64]

    Gs-ror 2: Bidirectional-guided 3dgs and sdf for reflective object re- lighting and reconstruction, 2025

    Zuo-Liang Zhu, Beibei Wang, and Jian Yang. Gs-ror 2: Bidirectional-guided 3dgs and sdf for reflective object re- lighting and reconstruction, 2025. 2, 3, 7 11 HumanOLAT: A Large-Scale Dataset for Full-Body Human Relighting and Novel-View Synthesis Supplementary Material This supplement discusses the ethical concerns in App. A, provides additional results ...

    work page 2025

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.