pith. machine review for the scientific record. sign in

arxiv: 2604.27871 · v1 · submitted 2026-04-30 · 💻 cs.GR

Recognition: unknown

D-Rex : Diffusion Rendering for Relightable Expressive Avatars

Timo Teufel , Xilong Zhou , Umar Iqbal , Jan Kautz , Marc Habermann , Vladislav Golyanik , Christian Theobalt
Authors on Pith no claims yet

Pith reviewed 2026-05-07 07:43 UTC · model grok-4.3

classification 💻 cs.GR
keywords relightable avatarsvideo diffusion modelsexpressive avatarsfull-body avatarsimage relightinglight stage captureavatar animation
0
0 comments X

The pith

D-Rex decouples relighting from avatar modeling by using a fine-tuned video diffusion model to translate flat-lit renders into target-illuminated images.

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

The paper establishes that relighting can be separated from creating expressive full-body avatars. Instead of building relighting into the 3D model with reflectance and geometry, D-Rex applies relighting as a post-processing step. It fine-tunes a pre-trained video diffusion model on pairs of flat-lit and relit frames from a light stage. The model learns to add realistic lighting effects to images rendered from an independent avatar model trained only under white light. This separation allows any existing avatar system to gain relighting capability while maintaining high expressiveness in motion and facial details, which matters for applications like virtual meetings and film production where both animation quality and lighting realism are needed.

Core claim

D-Rex is a person-specific framework for photorealistic, relightable, expressive, and animatable full-body human avatars. Relighting is treated as an image-space post-process that translates flat-lit, albedo-like renderings to a target HDR illumination using a LoRA-fine-tuned pre-trained video diffusion relighting model. The driving frames come from an independent expressive full-body avatar framework trained under white-light conditions, requiring no modification to support relighting.

What carries the argument

A LoRA-fine-tuned pre-trained video diffusion model that performs the image-space translation from flat-lit avatar renderings to relit images under arbitrary target HDR illumination.

If this is right

  • Relighting no longer requires explicit 3D intrinsic decomposition or analytic reflectance models.
  • Any white-light avatar system can be made relightable without changes to its training or architecture.
  • View-consistent and temporally consistent relighting is achieved while preserving expressive motion and fine facial details.
  • Performance exceeds that of physically-based relightable avatar baselines in quality and consistency.

Where Pith is reading between the lines

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

  • This decoupling suggests that similar post-process diffusion models could add other effects like weather or time-of-day changes to avatar renders.
  • Future work might explore whether the diffusion model can generalize to illuminations and poses far outside the light-stage training distribution.
  • Real-time applications could become feasible if the diffusion inference is distilled or accelerated for interactive use.

Load-bearing premise

Fine-tuning a pre-trained video diffusion model on paired flat-lit and relit light-stage frames produces artifact-free, temporally consistent translations for arbitrary new illuminations and novel poses.

What would settle it

Generate relit videos of the avatar in novel poses and under illuminations absent from the fine-tuning pairs, then check whether lighting artifacts appear or motion becomes inconsistent across frames.

Figures

Figures reproduced from arXiv: 2604.27871 by Christian Theobalt, Jan Kautz, Marc Habermann, Timo Teufel, Umar Iqbal, Vladislav Golyanik, Xilong Zhou.

Figure 1
Figure 1. Figure 1: Given 3D body motion, facial expressions, a target viewpoint, and a High Dynamic Range (HDR) environment map, D-Rex renders photorealistic, relightable images of the person under novel views and illuminations. We show that fine-tuning a pre-trained video diffusion relighting model solely on single-image flat-lit to relit frame pairs is sufficient to achieve convincing view and temporal consistency across f… view at source ↗
Figure 2
Figure 2. Figure 2: D-Rex overview. Given a calibrated sequence of flat-lit and HDR-illuminated multi-view frame pairs (Sec. 3.1), D-Rex trains two independent components. An ex￾pressive, controllable albedo avatar FWL (EVA [17]) is trained on the flat-lit frames to render albedo-like images for arbitrary pose, expression, and viewpoint. In parallel, a video diffusion relighting model [25] is fine-tuned via LoRA on the flat-l… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison against our constructed baselines. The two MeshAvatar￾based baselines [7] are unable to reconstruct the face due to missing expression depen￾dence. The PBR version of MeshAvatar, adapted to our setting, is able to learn intrin￾sics for relighting, but optimization is difficult to constrain. In comparison, training this version of MeshAvatar of flat-lit frames only, with video diffusi… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results for the experiment on removing background elements. While the model achieves results closer to ground truth by removing the background, percep￾tually realistic relighting can be achieved only through fine-tuning on the unmasked frame pairs view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results for training the video diffusion model on EVA to relit (en￾hancement) vs. real flat-lit to relit (relight only) frame pairs. While the enhancement model can fix artifacts in EVA (blurry shoes) it also tends to hallucinate more view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results for different model training strategies. Due to the domain gap, without fine-tuning the diffusion model is unable to perform realistic relighting on the full-body human. Both LoRA and full model fine-tuning from the pretrained checkpoint allow the model to learn convincing relighting, while training from random initialization fails to properly converge. 5 Limitations and Future Work Whi… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative demonstration of consistency across inference runs. adjusting the data collection protocol to record only slow movement, integrating emerging technology [51], and applying motion-based filtering during diffusion model optimization. Moreover, D-Rex requires light stage capture to enable re￾lighting, limiting accessibility. However, as demonstrated in Sec. 4.3, the video diffusion model requires … view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results for consistent video diffusion relighting. Row 1: Results for a 180◦ camera rotation under a fixed OLAT light, demonstrating view consistency. Row 2: Results for a 180◦ OLAT light rotation under a fixed front-facing view, showing illumination consistency. Row 3: Results for a 57-frame sequence under a fixed view and OLAT light, illustrating motion consistency. Fine-tuning on frame pairs… view at source ↗
Figure 9
Figure 9. Figure 9: Results for direct HDRI relighting under diverse expressions and motions. D￾Rex enables rendering of a person under controllable motion, expression, view, and illumination view at source ↗
read the original abstract

We present D-Rex, a person-specific framework for photorealistic, relightable, expressive, and animatable full-body human avatars with free-viewpoint rendering. Existing methods for relightable full-body avatars rely on explicit 3D intrinsic decomposition with analytic reflectance models, which require accurate geometry registration and careful optimization to capture realistic light transport effects. This tight coupling of relighting with avatar modeling has hindered expressiveness: to our knowledge, no existing method demonstrates strong facial animation alongside relighting, limiting applicability in telepresence, gaming, and virtual production. We propose to decouple relighting entirely from avatar modeling by treating it as an image-space post-process: a learned translation from flat-lit, albedo-like renderings to a target HDR illumination. To this end, we leverage the strong generative prior of a pre-trained video diffusion relighting model, fine-tuned via LoRA on paired flat-lit and relit frames captured in a light stage. The flat-lit driving frames are produced by an independent expressive full-body avatar framework trained under white-light conditions, requiring no modification to support relighting, making D-Rex directly applicable to any white-light avatar system. We demonstrate that D-Rex enables view- and temporally consistent relighting while faithfully preserving expressive motion and fine-grained facial detail, outperforming physically-based relightable avatar baselines. Project page is https://vcai.mpi-inf.mpg.de/projects/DRex/

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

3 major / 3 minor

Summary. The paper presents D-Rex, a person-specific framework for photorealistic relightable expressive full-body avatars. It decouples relighting from avatar modeling by treating relighting as an image-space post-process: a LoRA-fine-tuned pre-trained video diffusion model translates flat-lit, albedo-like renderings (produced by an independent white-light expressive avatar) into images under target HDR illumination. The method is trained on paired flat-lit and relit frames captured in a light stage and claims to deliver view- and temporally-consistent relighting that preserves expressive motion and fine-grained facial detail while outperforming physically-based relightable avatar baselines.

Significance. If the central claims hold, the work would be significant for enabling relightable expressive avatars without requiring explicit 3D geometry registration, reflectance decomposition, or light-transport optimization inside the avatar pipeline. By leveraging a generative video diffusion prior via efficient LoRA adaptation on light-stage pairs, the approach potentially captures complex effects (shadows, inter-reflections, specular highlights) more faithfully than analytic models while remaining compatible with any existing white-light avatar system. This decoupling could broaden applicability in telepresence, gaming, and virtual production where both high-fidelity facial animation and dynamic relighting are required.

major comments (3)
  1. [§4] §4 (Experiments and Results): The central claim of outperformance and artifact-free generalization to arbitrary illuminations and novel expressive poses rests on the diffusion prior correctly inferring physics without explicit geometry or light transport. However, the evaluation provides only qualitative examples on seen or lightly varied conditions; no quantitative metrics (e.g., PSNR, SSIM, or perceptual scores) or explicit out-of-distribution tests on held-out lighting environments and avatar-driven novel poses are reported. This leaves the domain-gap risk between real light-stage flat-lit captures and synthetic avatar renders unquantified and undermines the assertion that the method works for arbitrary new inputs.
  2. [§3.2] §3.2 (LoRA Fine-Tuning and Data Pipeline): The method description does not specify how temporal and spatial alignment is enforced between the independent avatar renderer’s flat-lit outputs and the light-stage relit targets during fine-tuning, nor whether any auxiliary losses (e.g., optical-flow consistency or normal-map guidance) are used beyond the base video diffusion objective. Without these details, it is unclear how the model avoids drift or artifacts when the driving avatar renderings deviate from the exact capture distribution.
  3. [Table 1] Table 1 or quantitative comparison subsection: The paper asserts superiority over physically-based relightable avatar baselines, yet the abstract and results summary contain no numerical values or statistical significance tests. A direct side-by-side table with error metrics on both relighting fidelity and temporal consistency for the same set of novel poses and illuminations is required to substantiate the claim.
minor comments (3)
  1. [§2] The related-work section should include recent citations on diffusion-based relighting and video consistency techniques to better contextualize the LoRA adaptation choice.
  2. [Figures 4-6] Figure captions for qualitative results should explicitly state the source of the driving avatar renderings (white-light vs. relit) and the exact illumination conditions used for each row to improve reproducibility.
  3. [§3.1] The notation for the conditioning signals (e.g., how target HDR illumination is encoded and injected into the diffusion U-Net) is introduced without a clear equation or diagram; a small diagram or pseudocode block would clarify the forward pass.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough and constructive review. The comments have prompted us to enhance the quantitative evaluation and provide more details on the training process. We address each point below and have updated the manuscript with the requested additions.

read point-by-point responses

  1. Referee: §4 (Experiments and Results): The central claim of outperformance and artifact-free generalization to arbitrary illuminations and novel expressive poses rests on the diffusion prior correctly inferring physics without explicit geometry or light transport. However, the evaluation provides only qualitative examples on seen or lightly varied conditions; no quantitative metrics (e.g., PSNR, SSIM, or perceptual scores) or explicit out-of-distribution tests on held-out lighting environments and avatar-driven novel poses are reported. This leaves the domain-gap risk between real light-stage flat-lit captures and synthetic avatar renders unquantified and undermines the assertion that the method works for arbitrary new inputs.

    Authors: We agree with the referee that quantitative metrics and explicit OOD tests are necessary to fully support our claims. In the revised version, we have included quantitative evaluations in §4 using PSNR, SSIM, and LPIPS on held-out light-stage captures with novel illuminations and poses. Additionally, we report results on synthetic flat-lit renders from the avatar model under unseen HDR environments to quantify the domain gap. These new results demonstrate the method's generalization and are summarized in an updated Table 1. revision: yes

  2. Referee: §3.2 (LoRA Fine-Tuning and Data Pipeline): The method description does not specify how temporal and spatial alignment is enforced between the independent avatar renderer’s flat-lit outputs and the light-stage relit targets during fine-tuning, nor whether any auxiliary losses (e.g., optical-flow consistency or normal-map guidance) are used beyond the base video diffusion objective. Without these details, it is unclear how the model avoids drift or artifacts when the driving avatar renderings deviate from the exact capture distribution.

    Authors: Thank you for pointing this out. The alignment is achieved by using the same motion capture sequences to drive both the light-stage actor and the avatar renderer, ensuring pixel-level correspondence in the paired data. No auxiliary losses are added; we rely on the pre-trained video diffusion model's temporal modeling capabilities and the LoRA fine-tuning on precisely paired frames. We have revised §3.2 to explicitly describe the data pairing process and alignment strategy. revision: yes

  3. Referee: Table 1 or quantitative comparison subsection: The paper asserts superiority over physically-based relightable avatar baselines, yet the abstract and results summary contain no numerical values or statistical significance tests. A direct side-by-side table with error metrics on both relighting fidelity and temporal consistency for the same set of novel poses and illuminations is required to substantiate the claim.

    Authors: We have added a comprehensive Table 1 in the revised manuscript that provides side-by-side quantitative comparisons with the baselines. The table includes error metrics for relighting fidelity (PSNR, SSIM, LPIPS) and temporal consistency (e.g., average optical flow magnitude between consecutive frames) evaluated on identical sets of novel poses and illuminations. We also include statistical analysis where relevant. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper presents a practical engineering method that decouples relighting from avatar modeling by applying a LoRA-fine-tuned video diffusion model as an image-space post-process. The driving signals come from an independent white-light avatar renderer and the training pairs are newly captured light-stage data; neither the model architecture nor the claimed performance reduces by construction to a reparameterization of the input renderings or fitted parameters. No equations are offered that would make the relit output definitionally equivalent to the flat-lit input, and no load-bearing uniqueness theorem or ansatz is imported via self-citation. The central claim therefore rests on empirical generalization of the learned translation rather than on any self-referential identity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the generative prior of a pre-trained video diffusion model being sufficient to learn a faithful flat-to-relit mapping from limited light-stage pairs, plus standard assumptions about LoRA fine-tuning preserving temporal and view consistency.

free parameters (1)
  • LoRA rank and scaling factors
    Chosen during fine-tuning to adapt the diffusion model; not derived from first principles.
axioms (1)
  • domain assumption A pre-trained video diffusion model can be adapted via LoRA to translate flat-lit avatar renderings into consistent relit versions under new HDR illumination.
    Invoked in the description of the image-space post-process and the use of light-stage paired data.

pith-pipeline@v0.9.0 · 5577 in / 1552 out tokens · 98886 ms · 2026-05-07T07:43:06.396946+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

59 extracted references · 19 canonical work pages · 4 internal anchors

  1. [1]

    Github (2021),https://github

    Easymocap - make human motion capture easier. Github (2021),https://github. com/zju3dv/EasyMocap

    2021

  2. [2]

    https://treedys.com/(2024)

    3D Body Scanning Technology.: Treedys. .https://treedys.com/(2024)

    2024

  3. [3]

    Agisoft LLC: Metashape.https://www.agisoft.com/downloads/installer/ (2025), version retrieved July 2025

    2025

  4. [4]

    In: International Conference on 3D Vision (3DV)

    Alldieck, T., Magnor, M., Xu, W., Theobalt, C., Pons-Moll, G.: Detailed human avatars from monocular video. In: International Conference on 3D Vision (3DV). pp. 98–109 (Sep 2018).https://doi.org/10.1109/3{DV}.2018.00022

    work page doi:10.1109/3 2018

  5. [5]

    In: International ConferenceonComputerVision(ICCV).p.1021–1030.IEEE(Oct2017).https:// doi.org/10.1109/iccv.2017.116,http://dx.doi.org/10.1109/ICCV.2017.116

    Bulat, A., Tzimiropoulos, G.: How far are we from solving the 2d & 3d face align- ment problem? (and a dataset of 230,000 3d facial landmarks). In: International ConferenceonComputerVision(ICCV).p.1021–1030.IEEE(Oct2017).https:// doi.org/10.1109/iccv.2017.116,http://dx.doi.org/10.1109/ICCV.2017.116

    work page doi:10.1109/iccv.2017.116 2017

  6. [6]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2019)

    Cao, Z., Hidalgo, G., Simon, T., Wei, S.E., Sheikh, Y.: Openpose: Realtime multi- person 2d pose estimation using part affinity fields. IEEE Transactions on Pattern Analysis and Machine Intelligence (2019)

    2019

  7. [7]

    In: European Conference on Computer Vision (ECCV) (2024)

    Chen, Y., Zheng, Z., Li, Z., Xu, C., Liu, Y.: Meshavatar: Learning high-quality triangular human avatars from multi-view videos. In: European Conference on Computer Vision (ECCV) (2024)

    2024

  8. [8]

    In: SIGGRAPH (2000)

    Debevec, P., Hawkins, T., Tchou, C., Sarokin, W., Sagar, M.: Acquiring the re- flectance field of a human face. In: SIGGRAPH (2000)

    2000

  9. [9]

    Fang, Y., Wu, T., Deschaintre, V., Ceylan, D., Georgiev, I., Huang, C.H.P., Hu, Y., Chen, X., Wang, T.Y.: V-rgbx: Video editing with accurate controls over intrinsic properties (2025),https://arxiv.org/abs/2512.11799

    work page arXiv 2025

  10. [10]

    Habermann, M., Liu, L., Xu, W., Pons-Moll, G., Zollhoefer, M., Theobalt, C.: Hdhumans: A hybrid approach for high-fidelity digital humans. Proc. ACM Com- put. Graph. Interact. Tech.6(3) (aug 2023).https://doi.org/10.1145/3606927, https://doi.org/10.1145/3606927

    work page doi:10.1145/3606927 2023

  11. [11]

    ACM Trans

    Habermann, M., Liu, L., Xu, W., Zollhoefer, M., Pons-Moll, G., Theobalt, C.: Real-time deep dynamic characters. ACM Trans. Graph.40(4) (aug 2021)

    2021

  12. [12]

    In: Neural Information Processing Systems (NeurIPS) (2025)

    He, K., Liang, R., Munkberg, J., Hasselgren, J., Vijaykumar, N., Keller, A., Fidler, S., Gilitschenski, I., Gojcic, Z., Wang, Z.: Unirelight: Learning joint decomposi- tion and synthesis for video relighting. In: Neural Information Processing Systems (NeurIPS) (2025)

    2025

  13. [13]

    In: SIGGRAPH Asia

    He, M., Clausen, P., Taşel, A.L., Ma, L., Pilarski, O., Xian, W., Rikker, L., Yu, X., Burgert, R., Yu, N., Debevec, P.: Diffrelight: Diffusion-based facial performance relighting. In: SIGGRAPH Asia. SA ’24 (2024)

    2024

  14. [14]

    In: International Conference on Computer Vision (ICCV) (2023)

    Iqbal, U., Caliskan, A., Nagano, K., Khamis, S., Molchanov, P., Kautz, J.: Rana: Relightable articulated neural avatars. In: International Conference on Computer Vision (ICCV) (2023)

    2023

  15. [15]

    In: International Conference on Computer Vision (ICCV) Findings (2025)

    Jiang, Z., Wang, S., Tang, S.: Dnf-avatar: Distilling neural fields for real-time animatable avatar relighting. In: International Conference on Computer Vision (ICCV) Findings (2025)

    2025

  16. [16]

    In: Neural Information Pro- cessing Systems (NeurIPS) (2024)

    Jin, H., Li, Y., Luan, F., Xiangli, Y., Bi, S., Zhang, K., Xu, Z., Sun, J., Snavely, N.: Neural gaffer: Relighting any object via diffusion. In: Neural Information Pro- cessing Systems (NeurIPS) (2024)

    2024

  17. [17]

    In: SIGGRAPH (2025) 16 Teufel et al

    Junkawitsch, H., Sun, G., Zhu, H., Theobalt, C., Habermann, M.: Eva: Expressive virtual avatars from multi-view videos. In: SIGGRAPH (2025) 16 Teufel et al

    2025

  18. [18]

    In: Proceedings of the 2007 Symposium on Interactive 3D Graphics and Games

    Kavan, L., Collins, S., Žára, J., O’Sullivan, C.: Skinning with dual quaternions. In: Proceedings of the 2007 Symposium on Interactive 3D Graphics and Games. p. 39–46. I3D ’07, Association for Computing Machinery, New York, NY, USA (2007).https://doi.org/10.1145/1230100.1230107,https://doi.org/10. 1145/1230100.1230107

    work page doi:10.1145/1230100.1230107 2007

  19. [19]

    ACM Trans

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph.42(4), 139–1 (2023)

    2023

  20. [20]

    In: European Conference on Computer Vision (ECCV) (2024)

    Khirodkar, R., Bagautdinov, T., Martinez, J., Zhaoen, S., James, A., Selednik, P., Anderson, S., Saito, S.: Sapiens: Foundation for human vision models. In: European Conference on Computer Vision (ECCV) (2024)

    2024

  21. [21]

    In: Computer Vision and Pattern Recognition (CVPR) (2024)

    Kim, H., Jang, M., Yoon, W., Lee, J., Na, D., Woo, S.: Switchlight: Co-design of physics-driven architecture and pre-training framework for human portrait relight- ing. In: Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  22. [22]

    Kocsis, P., Sitzmann, V., Nießner, M.: Intrinsic image diffusion for indoor single- view material estimation (2024),https://arxiv.org/abs/2312.12274

    work page arXiv 2024

  23. [23]

    ACM Trans

    Li, T., Bolkart, T., Black, M.J., Li, H., Romero, J.: Learning a model of facial shape and expression from 4D scans. ACM Trans. Graph.36(6), 194:1–194:17 (2017),https://doi.org/10.1145/3130800.3130813

    work page doi:10.1145/3130800.3130813 2017

  24. [24]

    In: Computer Vision and Pattern Recognition (CVPR) (2024)

    Li, Z., Zheng, Z., Wang, L., Liu, Y.: Animatable gaussians: Learning pose- dependent gaussian maps for high-fidelity human avatar modeling. In: Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  25. [25]

    In: Computer Vision and Pat- tern Recognition (CVPR) (June 2025)

    Liang, R., Gojcic, Z., Ling, H., Munkberg, J., Hasselgren, J., Lin, Z.H., Gao, J., Keller, A., Vijaykumar, N., Fidler, S., Wang, Z.: Diffusionrenderer: Neural inverse and forward rendering with video diffusion models. In: Computer Vision and Pat- tern Recognition (CVPR) (June 2025)

    2025

  26. [26]

    In: Association for the Advancement of Artificial Intelligence (AAAI) (2024)

    Lin, W., Zheng, C., Yong, J.H., Xu, F.: Relightable and animatable neural avatars from videos. In: Association for the Advancement of Artificial Intelligence (AAAI) (2024)

    2024

  27. [27]

    ACM Trans

    Loper, M., Mahmood, N., Romero, J., Pons-Moll, G., Black, M.J.: SMPL: A skinned multi-person linear model. ACM Trans. Graph.34(6), 248:1–248:16 (Oct 2015)

    2015

  28. [28]

    Lugaresi, C., Tang, J., Nash, H., McClanahan, C., Uboweja, E., Hays, M., Zhang, F., Chang, C.L., Yong, M.G., Lee, J., Chang, W.T., Hua, W., Georg, M., Grund- mann, M.: Mediapipe: A framework for building perception pipelines (2019), https://arxiv.org/abs/1906.08172

    work page internal anchor Pith review arXiv 2019

  29. [29]

    In: European Conference on Computer Vision (ECCV) (2024)

    Luvizon, D., Golyanik, V., Kortylewski, A., Habermann, M., Theobalt, C.: Re- lightable neural actor with intrinsic decomposition and pose control. In: European Conference on Computer Vision (ECCV) (2024)

    2024

  30. [30]

    Markerless motion capture technology.: Captury .https://captury.com(2021)

    2021

  31. [31]

    In: Computer Vision and Pattern Recognition (CVPR) (2025)

    Mei, Y., He, M., Ma, L., Philip, J., Xian, W., George, D.M., Yu, X., Dedic, G., Taşel, A.L., Yu, N., Patel, V.M., Debevec, P.: Lux post facto: Learning portrait performance relighting with conditional video diffusion and a hybrid dataset. In: Computer Vision and Pattern Recognition (CVPR) (2025)

    2025

  32. [32]

    In: Eu- ropean Conference on Computer Vision (ECCV) (2020)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: Eu- ropean Conference on Computer Vision (ECCV) (2020)

    2020

  33. [33]

    ACM Trans

    Pandey, R., Orts-Escolano, S., Legendre, C., Haene, C., Bouaziz, S., Rhemann, C., Debevec, P.E., Fanello, S.R.: Total relighting: learning to relight portraits for background replacement. ACM Trans. Graph.40(4), 43–1 (2021) D-Rex: Diffusion Rendering for Relightable Expressive Avatars 17

    2021

  34. [34]

    In: Computer Vision and Pattern Recognition (CVPR)

    Pang,H.,Zhu,H.,Kortylewski,A.,Theobalt,C.,Habermann,M.:Ash:Animatable gaussian splats for efficient and photoreal human rendering. In: Computer Vision and Pattern Recognition (CVPR). pp. 1165–1175 (June 2024)

    2024

  35. [35]

    In: Computer Vision and Pattern Recognition (CVPR)

    Pavlakos, G., Choutas, V., Ghorbani, N., Bolkart, T., Osman, A.A.A., Tzionas, D., Black, M.J.: Expressive body capture: 3D hands, face, and body from a single image. In: Computer Vision and Pattern Recognition (CVPR)

  36. [36]

    Peebles, W., Xie, S.: Scalable diffusion models with transformers (2023),https: //arxiv.org/abs/2212.09748

    work page internal anchor Pith review arXiv 2023

  37. [37]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Peng, S., Dong, J., Wang, Q., Zhang, S., Shuai, Q., Zhou, X., Bao, H.: An- imatable Neural Radiance Fields for Modeling Dynamic Human Bodies . In: International Conference on Computer Vision (ICCV). pp. 14294–14303. IEEE Computer Society, Los Alamitos, CA, USA (Oct 2021).https : / / doi . org / 10.1109/ICCV48922.2021.01405,https://doi.ieeecomputersociety....

    work page doi:10.1109/iccv48922.2021.01405 2021

  38. [38]

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis (2023),https://arxiv.org/abs/2307.01952

    work page internal anchor Pith review arXiv 2023

  39. [39]

    Computer Graphics Forum (2024).https://doi.org/10.1111/cgf.15147

    Poirier-Ginter, Y., Gauthier, A., Philip, J., Lalonde, J.F., Drettakis, G.: A Diffu- sion Approach to Radiance Field Relighting using Multi-Illumination Synthesis. Computer Graphics Forum (2024).https://doi.org/10.1111/cgf.15147

    work page doi:10.1111/cgf.15147 2024

  40. [40]

    In: International Conference on 3D Vision (3DV) (2025)

    Shao, Z., Wang, D., Tian, Q.Y., Yang, Y.D., Meng, H., Cai, Z., Dong, B., Zhang, Y., Zhang, K., Wang, Z.: DEGAS: Detailed Expressions on Full-Body Gaussian Avatars. In: International Conference on 3D Vision (3DV) (2025)

    2025

  41. [41]

    In: Computer Vision and Pattern Recognition (CVPR) (2023)

    Shen, K., Guo, C., Kaufmann, M., Zarate, J., Valentin, J., Song, J., Hilliges, O.: X- avatar: Expressive human avatars. In: Computer Vision and Pattern Recognition (CVPR) (2023)

    2023

  42. [42]

    Singh, K.M., Chen, J., Golyanik, V., Garbin, S.J., Beeler, T., Dabral, R., Haber- mann, M., Theobalt, C.: Relightable holoported characters: Capturing and relight- ing dynamic human performance from sparse views (2025)

    2025

  43. [43]

    ACM Trans

    Sun, T., Barron, J.T., Tsai, Y.T., Xu, Z., Yu, X., Fyffe, G., Rhemann, C., Busch, J., Debevec, P., Ramamoorthi, R.: Single image portrait relighting. ACM Trans. Graph.38(4) (Jul 2019).https://doi.org/10.1145/3306346.3323008,https: //doi.org/10.1145/3306346.3323008

    work page doi:10.1145/3306346.3323008 2019

  44. [44]

    In: Neural Information Processing Systems (NeurIPS) (2025)

    Tang, J., Levine, M., Verbin, D., Garbin, S.J., Niessner, M., Martin-Brualla, R., Srinivasan, P.P., Henzler, P.: ROGR: Relightable 3D Objects using Generative Relighting. In: Neural Information Processing Systems (NeurIPS) (2025)

    2025

  45. [45]

    In: International Conference on Computer Vision (ICCV) (2025)

    Teufel, T., Gera, P., Zhou, X., Iqbal, U., Rao, P., Kautz, J., Golyanik, V., Theobalt, C.: Humanolat: A large-scale dataset for full-body human relighting and novel-view synthesis. In: International Conference on Computer Vision (ICCV) (2025)

    2025

  46. [46]

    In: Computer Vision and Pattern Recognition (CVPR)

    Wang, J., Liu, J., Sun, X., Singh, K.K., Shu, Z., Zhang, H., Yang, J., Zhao, N., Wang, T.Y., Chen, S.S., et al.: Comprehensive relighting: Generalizable and con- sistent monocular human relighting and harmonization. In: Computer Vision and Pattern Recognition (CVPR). pp. 380–390 (2025)

    2025

  47. [47]

    In: Computer Vision and Pattern Recognition (CVPR) (2024)

    Wang, S., Antić, B., Geiger, A., Tang, S.: Intrinsicavatar: Physically based inverse rendering of dynamic humans from monocular videos via explicit ray tracing. In: Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  48. [48]

    In: SIGGRAPH (2025) 18 Teufel et al

    Wang, S., Simon, T., Santesteban, I., Bagautdinov, T., Li, J., Agrawal, V., Prada, F., Yu, S.I., Nalbone, P., Gramlich, M., Lubachersky, R., Wu, C., Romero, J., Saragih, J., Zollhoefer, M., Geiger, A., Tang, S., Saito, S.: Relightable full-body gaussian codec avatars. In: SIGGRAPH (2025) 18 Teufel et al

    2025

  49. [49]

    In: Interna- tional Conference on Computer Vision (ICCV) (2023)

    Wang, Y., Han, Q., Habermann, M., Daniilidis, K., Theobalt, C., Liu, L.: Neus2: Fast learning of neural implicit surfaces for multi-view reconstruction. In: Interna- tional Conference on Computer Vision (ICCV) (2023)

    2023

  50. [50]

    In: Computer Vision and Pattern Recognition (CVPR) (2024)

    Xu, Z., Peng, S., Geng, C., Mou, L., Yan, Z., Sun, J., Bao, H., Zhou, X.: Relightable and animatable neural avatar from sparse-view video. In: Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  51. [51]

    Rechar: Revitalising characters with structure preserved and user-specified aesthetic enhancements

    Yu, X., He, M., George, D., Joshi, A., Debevec, P.: Digital bi-pack: Recording live-action under two near-simultaneous lighting conditions. In: SIGGRAPH Asia. SA Technical Communications ’25, Association for Computing Machinery, New York, NY, USA (2025).https://doi.org/10.1145/3757376.3771405,https: //doi.org/10.1145/3757376.3771405

    work page doi:10.1145/3757376.3771405 2025

  52. [52]

    In: SIGGRAPH (2024)

    Zeng, C., Dong, Y., Peers, P., Kong, Y., Wu, H., Tong, X.: Dilightnet: Fine-grained lighting control for diffusion-based image generation. In: SIGGRAPH (2024)

    2024

  53. [53]

    In: SIGGRAPH

    Zeng, Z., Deschaintre, V., Georgiev, I., Hold-Geoffroy, Y., Hu, Y., Luan, F., Yan, L.Q., Hašan, M.: Rgbx: Image decomposition and synthesis using material- and lighting-aware diffusion models. In: SIGGRAPH. SIGGRAPH, Association for Computing Machinery, New York, NY, USA (2024).https://doi.org/10.1145/ 3641519.3657445,https://doi.org/10.1145/3641519.3657445

    work page doi:10.1145/3641519.3657445 2024

  54. [54]

    In: International Conference on Computer Vision (ICCV) (2021)

    Zhang, L., Zhang, Q., Wu, M., Yu, J., Xu, L.: Neural video portrait relighting in real-time via consistency modeling. In: International Conference on Computer Vision (ICCV) (2021)

    2021

  55. [55]

    In: International Conference on Learning Representations (ICLR) (2025),https: //openreview.net/forum?id=u1cQYxRI1H

    Zhang, L., Rao, A., Agrawala, M.: Scaling in-the-wild training for diffusion-based illumination harmonization and editing by imposing consistent light transport. In: International Conference on Learning Representations (ICLR) (2025),https: //openreview.net/forum?id=u1cQYxRI1H

    2025

  56. [56]

    Zhang, S., Liu, R., Schroers, C., Zhang, Y.: Renderflow: Single-step neural render- ing via flow matching (2026),https://arxiv.org/abs/2601.06928

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  57. [57]

    Zhang, T., Kuang, Z., Jin, H., Xu, Z., Bi, S., Tan, H., Zhang, H., Hu, Y., Hasan, M., Freeman, W.T., Zhang, K., Luan, F.: Relitlrm: Generative relightable radiance for large reconstruction models (2024),https://arxiv.org/abs/2410.06231

    work page arXiv 2024

  58. [58]

    ACM Trans

    Zheng, Z., Zhao, X., Zhang, H., Liu, B., Liu, Y.: Avatarrex: Real-time expressive full-body avatars. ACM Trans. Graph.42(4) (Jul 2023).https://doi.org/10. 1145/3592101,https://doi.org/10.1145/3592101

    work page doi:10.1145/3592101 2023

  59. [59]

    Zhou, T., He, K., Wu, D., Xu, T., Zhang, Q., Shao, K., Chen, W., Xu, L., Yu, J.: Relightable neural human assets from multi-view gradient illuminations. In: Computer Vision and Pattern Recognition (CVPR) (2023) D-Rex: Diffusion Rendering for Relightable Expressive Avatars Supplementary Material Timo Teufel1, Xilong Zhou1, Umar Iqbal2, Jan Kautz2, Marc Hab...

    work page arXiv 2023