Pith. sign in

REVIEW 2 major objections 5 minor 75 references

Lume-Palette decouples multi-view indoor relighting into text-aligned illumination palettes and geometry-driven casting so diffusion models keep photorealism while accepting exact 3D light placement.

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 ↗

T0 review · grok-4.5

2026-07-13 06:03 UTC pith:YHTVO7DP

load-bearing objection Clean two-stage split that keeps diffusion lighting priors intact while adding real 3D spatial control; synthetic casting is the soft spot, but the paper is honest and the engineering is usable. the 2 major comments →

arxiv 2607.08879 v1 pith:YHTVO7DP submitted 2026-07-09 cs.CV

Decoupled Illumination Priors for Spatially Controllable Multi-View Indoor Scene Relighting

Chenjian Gao , Linning Xu , Tianfan Xue This is my paper
classification cs.CV
keywords indoor scene relightingillumination distillationillumination palettesmulti-view consistencydiffusion modelsspatial lighting controlasymmetric multi-view conditioning
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.

Indoor scene relighting needs three things at once: photorealistic material response, user-placed 3D lights, and consistent appearance across many viewpoints. Directly forcing a pretrained diffusion editor to accept spatial light maps tends to erase the lighting knowledge the model already has. Lume-Palette therefore splits the job. First it distills a small set of canonical "illumination palettes" (left, right, front, top) from real multi-illumination photographs using only text prompts the model already understands; these palettes capture how each surface actually reacts to light. Second, it casts those palettes onto new target lighting by rendering simple white-shaded maps from a coarse reconstructed mesh and feeding them, together with the palettes, through an asymmetric multi-view network that keeps dense conditions only on the active view. The result is feed-forward, spatially controllable, multi-view-consistent relighting that stays visually natural on both synthetic and real rooms.

Core claim

The paper shows that the generative lighting prior of a large diffusion model can be preserved for indoor multi-view relighting by first distilling it into view-specific canonical illumination palettes via text-conditioned LoRAs, then casting those palettes under receiver-centric spatial lighting maps derived from coarse 3D geometry, using asymmetric multi-view conditioning to keep the computation tractable. This two-stage decoupling yields higher photorealism and stricter spatial and multi-view fidelity than methods that inject spatial maps directly into the diffusion backbone.

What carries the argument

Illumination palettes: four canonical directional relightings (left/right/front/top) distilled per view from a pretrained diffusion model; they serve as material-response references that the casting stage then remaps under user-defined 3D lights via receiver-centric shading maps and asymmetric multi-view conditioning.

Load-bearing premise

That a coarse mesh reconstructed from multi-view images, when rendered with uniform white material under the user’s lights, already supplies accurate enough shading maps for the network to produce correct shadows and material response on real scenes.

What would settle it

On a real ScanNet++ scene containing thin furniture or glossy surfaces, replace the Depth-Anything-3 mesh with a high-quality laser scan, re-render the white lighting maps, and check whether measured shadow boundaries, highlight placement, and multi-view LPIPS remain statistically unchanged; a large drop would falsify the claim that coarse geometry is sufficient.

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

If this is right

  • Users can place arbitrary point, spot or area lights inside a reconstructed room and obtain multi-view consistent, photorealistic results without per-scene radiance-field optimization.
  • Because the palettes are distilled from real photographs, the method transfers material response (specularities, soft fall-off) that pure synthetic training tends to lose.
  • Asymmetric conditioning lets the same feed-forward network scale to denser viewpoint sets without an explosion of conditioning tokens.
  • Receiver-centric maps naturally handle off-screen lights, removing the need to keep light sources inside the camera frustum.

Where Pith is reading between the lines

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

  • The same palette-then-cast pattern could be applied to outdoor or object-centric relighting where environment maps replace the four directional prompts.
  • If the casting network were conditioned on a small set of learned material tokens instead of full palette images, memory and latency could drop further while still protecting the generative prior.
  • Failures on transparent or highly specular surfaces suggest that an explicit residual path for view-dependent effects may be needed before the method can support interior-design applications that rely on glass and metal.

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. Lume-Palette is a two-stage diffusion framework for multi-view indoor scene relighting that aims for photorealism, explicit 3D spatial light control, and cross-view consistency. Illumination distillation fine-tunes LoRA modules on MIT Multi-Illumination so a pretrained Flux.2 model produces canonical directional palettes (left/right/front/top) from text prompts, preserving material–light priors without invasive spatial conditioning. Illumination casting then combines those palettes with multi-view source images and receiver-centric lighting maps L (white-mesh renders of user-placed lights on a Depth Anything 3 reconstruction) under an asymmetric multi-view scheme: only the active view receives dense conditions while inactive views supply noisy latents as spatial anchors. The casting stage is trained on 480 synthetic multi-view pairs. On held-out synthetic scenes the method reports large gains over RGB↔X, IC-Light, LumiNet and ScribbleLight (PSNR 24.45 vs next-best 16.59; Table 1) plus better multi-view warp metrics; ablations (Table 3, Fig. 6) support both components. Real ScanNet++ results are qualitative plus a 25-person Bradley–Terry user study (150 judgments) favoring the method on realism, lighting adherence and multi-view consistency.

Significance. If the claims hold, the paper offers a practical route to spatially controllable multi-view indoor relighting that keeps pretrained generative lighting priors rather than overwriting them with synthetic spatial maps. The decoupling into text-aligned palettes plus receiver-centric casting, together with the asymmetric conditioning that reduces DiT input sequences from 6V to V+5, is a concrete engineering contribution for dense multi-modal multi-view diffusion. Quantitative synthetic superiority and clean ablations are solid; the real-scene user study and continuous light-control demos (Figs. 8–9) increase practical interest for AR/interior-design applications. Strengths that should be credited: explicit ablations isolating palette and multi-view awareness, a well-motivated asymmetric design with a clear complexity argument, and honest limitation discussion of coarse geometry.

major comments (2)
  1. §4.1 and §4.2: The casting network that actually enforces receiver-centric maps L and multi-view consistency is trained exclusively on 480 synthetic multi-view pairs, while real-scene evaluation (ScanNet++ Fig. 5, Table 2) has no ground-truth target illumination. The user study scores preference for “lighting adherence to the reference diffuse lighting condition,” which is itself a white-mesh render from Depth Anything 3. Preference can therefore be satisfied by plausible shading that merely correlates with L rather than by correct material–light transport under the intended 3D lights. This synthetic-to-real gap is load-bearing for the central claim of spatially controllable real-scene relighting; either a controlled real multi-illumination subset, a proxy metric that isolates spatial fidelity, or a clearer statement of the claim’s scope is needed.
  2. §3.3 and §5: Receiver-centric conditions L are rendered from coarse Depth Anything 3 meshes with uniform white material. The paper correctly flags failures on thin structures, reflections and complex indirect light, yet the quantitative multi-view metrics (Table 1) and the user-study “lighting adherence” scores both depend on the fidelity of these maps. Without a controlled ablation that injects known geometric error or compares against higher-quality meshes, it remains unclear how much of the reported spatial control is an artifact of the synthetic training distribution versus genuine robustness on real geometry.
minor comments (5)
  1. §3.2: The four canonical directions are fixed a priori; a short justification or ablation on the number/choice of directions would strengthen the design claim.
  2. §3.3, inference paragraph: The procedure of cycling every view as active and aggregating velocities is clear, but the precise sampling of the V−1 inactive anchors when N > V is underspecified (random, nearest, fixed window?).
  3. Table 1 / Fig. 4: Baseline adaptation details (how spatial maps or multi-view inputs were supplied to methods not originally designed for them) should be stated so the comparison is reproducible.
  4. Fig. 2 caption and §3.3: “asymmetric multi-view conditioning” is introduced with a concrete sequence-length reduction (24→9 for V=4); placing that calculation in the main text rather than only the prose would aid readers.
  5. Minor typography: occasional missing spaces after commas in the abstract and early sections; “Multi-Illumination Dataset” capitalization is inconsistent.

Circularity Check

0 steps flagged

No circularity: empirical two-stage diffusion pipeline with independent training data, held-out metrics, and user study; no derivation reduces a claimed prediction to its own inputs by construction.

full rationale

Lume-Palette is a standard computer-vision engineering paper. Illumination distillation fine-tunes LoRAs on the external MIT Multi-Illumination dataset under fixed text prompts; illumination casting is trained on 480 synthetic multi-view pairs and evaluated on 10 held-out synthetic scenes (PSNR/SSIM/LPIPS/MV metrics) plus a real-scene user study. Receiver-centric lighting maps L are rendered from Depth Anything 3 meshes under user lights; they are conditioning inputs, not fitted parameters that are later “predicted.” Asymmetric multi-view conditioning is an architectural choice, not a uniqueness theorem. No equation equates a claimed result to a fitted quantity by construction, no load-bearing self-citation of an unverified uniqueness result appears, and no known empirical pattern is merely renamed. The geometry-limitation note in §5 is an honest caveat, not circularity. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

As an empirical computer-vision method the paper rests on standard diffusion and multi-view assumptions plus a handful of design choices (four canonical directions, LoRA ranks, V=4 window, white-mesh lighting maps). No new physical entities are postulated; the free parameters are ordinary hyper-parameters and the invented entities are named intermediate representations whose utility is measured by ablation.

free parameters (5)
  • LoRA rank (distillation) = 32
    Set to 32; controls capacity of the four directional adapters and is not derived from first principles.
  • LoRA rank (casting) = 128
    Set to 128; chosen to handle denser spatial conditions.
  • context window size V = 4
    Fixed at 4 views during training; trades multi-view context against memory.
  • learning rate = 1e-4
    1e-4 for both stages with AdamW; standard but free choice.
  • number of canonical directions = 4
    Four directions (left/right/front/top) define the palette; arbitrary discrete set.
axioms (4)
  • domain assumption A pretrained text-to-image diffusion model already encodes realistic material-light interaction priors that can be elicited by simple directional text prompts after light LoRA fine-tuning.
    Stated in §3.2 and used to justify distillation on MIT Multi-Illumination without invasive spatial conditioning.
  • domain assumption Coarse geometry from Depth Anything 3 plus uniform white material is a sufficient receiver-centric proxy for target illumination distribution across views.
    Core of the lighting-condition construction in §3.3; the paper later lists its failure modes as a limitation.
  • ad hoc to paper Asymmetric conditioning (dense signals only on the active view, noisy latents on inactive views) preserves enough cross-view information for multi-view consistency.
    Introduced in §3.3 to solve the dimensionality explosion; validated only by the paper’s own ablation.
  • standard math Flow-matching velocity prediction with VAE latents is an adequate training objective for both stages.
    Standard for the chosen Flux.2 backbone; equations (1) and the distillation objective follow the usual formulation.
invented entities (2)
  • illumination palette no independent evidence
    purpose: Four view-specific images under canonical lighting directions that serve as material-response references for the casting stage.
    Defined in §3.2; utility shown by ablation removing the palette, but no independent physical measurement outside the pipeline.
  • asymmetric multi-view conditioning no independent evidence
    purpose: Reduce conditional sequence length from 6V to V+5 by giving dense conditions only to the active view while using inactive noisy latents as spatial anchors.
    Introduced in §3.3; computational necessity rather than a discovered natural phenomenon.

pith-pipeline@v1.1.0-grok45 · 20441 in / 3272 out tokens · 39692 ms · 2026-07-13T06:03:47.158400+00:00 · methodology

0 comments
read the original abstract

Indoor scene relighting demands photorealism, precise spatial control, and strict multi-view consistency. While diffusion-based image editing models enable semantic lighting manipulation via text prompts, enforcing exact 3D light placement often disrupts their generative priors. We propose Lume-Palette, a progressive framework that leverages semantic lighting priors for spatially controllable multi-view indoor relighting. The approach decouples relighting into two stages: (1) illumination distillation, which extracts canonical illumination palettes from a pretrained diffusion model to preserve realistic material-light interactions, and (2) illumination casting, which explicitly maps target spatial lighting conditions defined from coarse 3D geometry. To efficiently handle dense multi-view and multi-modal inputs, we introduce an asymmetric multi-view conditioning strategy that selectively injects essential spatial context. Experiments on diverse synthetic scenes and real-world scenes demonstrate that Lume-Palette produces photorealistic, spatially controllable, and multi-view consistent relighting results. Project Page: https://cjeen.github.io/lumepalette

Figures

Figures reproduced from arXiv: 2607.08879 by Chenjian Gao, Linning Xu, Tianfan Xue.

Figure 1
Figure 1. Figure 1: Multi-view indoor scene relighting with Lume-Palette. Our progressive frame￾work first extracts canonical "illumination palettes" from multi-view source images to capture material-light interactions. These palettes subsequently guide the synthesis of photorealistic, multi-view consistent relit images under new target illuminations. Recently, diffusion-based image editing models [3, 26, 62] have excelled at… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of Lume-Palette. Our framework decouples indoor relighting into two sequential stages. Illumination Distillation fine-tunes a diffusion model to extract canonical illumination palettes, capturing intrinsic material responses. Illumination Casting synthesizes the target illumination by combining these palettes with source images and user-defined spatial lighting conditions. To ensure scalable multi… view at source ↗
Figure 3
Figure 3. Figure 3: Generation of spatial lighting conditions. First, a coarse 3D mesh is recon￾structed from multi-view input images [32]. Users can then explicitly place custom light sources within this 3D space. Finally, the scene is rendered to produce view￾specific lighting conditions for the illumination casting stage. objective. Let z0 = E(Id) be the latent representation of the target image en￾coded by a pretrained VA… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison against baselines on synthetic scenes. Our method achieves precise spatial control, realistic appearance, and multi-view consistency [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison against baselines on real scenes. Lume-Palette robustly synthesizes realistic shading that adheres to imposed spatial lighting conditions. We further evaluate Lume-Palette on real scenes from ScanNet++ [65]. As shown in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative ablation of the illumination palette and multi-view awareness. Re￾moving the illumination palette or multi-view awareness degrades performance. Input Illumination Palette Input Illumination Palette [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Robustness of the illumination palette to varying input illumination. Changes in input light sources and shadows have almost no effect on the extracted palette. a worse LPIPS score compared to the full model. Qualitatively, [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Continuous spatial and dynamic controllability of light intensity and direction. The generated illumination, including specular highlights and cast shadows, updates naturally and remains firmly attached to the underlying geometry. 4.4 Spatial and Dynamic Controllability A core advantage of Lume-Palette is its fine-grained lighting control. As illus￾trated in [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Lume-Palette supports flexible lighting configurations, including point lights at different positions, multiple simultaneous lights, and rectangular area lights. 5 Conclusion and Limitations We presented Lume-Palette, a progressive framework for spatially controllable multi-view indoor scene relighting. By decoupling relighting into illumination dis￾tillation and illumination casting, our method leverages … view at source ↗

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

75 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Azinovic, D., Li, T.M., Kaplanyan, A., Nießner, M.: Inverse path tracing for joint material and lighting estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2447–2456 (2019)

    2019

  2. [2]

    IEEE Trans

    Barron, J.T., Malik, J.: Shape, illumination, and reflectance from shading. IEEE Trans. Pattern Anal. Mach. Intell.37(8), 1670–1687 (2015).https://doi.org/ 10.1109/TPAMI.2014.2377712,https://doi.org/10.1109/TPAMI.2014.2377712

    doi:10.1109/tpami.2014.2377712 2015

  3. [3]

    1 kontext: Flow matching for in-context image generation and editing in latent space

    Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dockhorn, T., En- glish, J., English, Z., Esser, P., Kulal, S., et al.: Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space. arXiv e-prints pp. arXiv–2506 (2025)

    2025

  4. [4]

    In: Proceedings of the Eighth International Conference On Computer Vision (ICCV- 01), Vancouver, British Columbia, Canada, July 7-14, 2001 - Volume 1

    Bell, M., Freeman, W.T.: Learning local evidence for shading and reflectance. In: Proceedings of the Eighth International Conference On Computer Vision (ICCV- 01), Vancouver, British Columbia, Canada, July 7-14, 2001 - Volume 1. pp. 670–

    2001

  5. [5]

    Bohr Hamiltonian with Hulth\'en plus ring-shaped potential for triaxial nuclei with deformation-dependent mass term

    IEEE Computer Society (2001).https://doi.org/10.1109/ICCV.2001. 10095,https://doi.ieeecomputersociety.org/10.1109/ICCV.2001.10095

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1109/iccv.2001 2001

  6. [6]

    In: Proceedings of the SIGGRAPH Asia 2025 Conference Papers

    Bharadwaj, S., Feng, H., Becherini, G., Fernandez Abrevaya, V., Black, M.J.: Gen- lit: Reformulating single-image relighting as video generation. In: Proceedings of the SIGGRAPH Asia 2025 Conference Papers. pp. 1–12 (2025)

    2025

  7. [7]

    In: European Conference on Computer Vision

    Bi, S., Xu, Z., Sunkavalli, K., Hašan, M., Hold-Geoffroy, Y., Kriegman, D., Ra- mamoorthi, R.: Deep reflectance volumes: Relightable reconstructions from multi- view photometric images. In: European Conference on Computer Vision. pp. 294–

  8. [8]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Boss, M., Braun, R., Jampani, V., Barron, J.T., Liu, C., Lensch, H.: Nerd: Neural reflectance decomposition from image collections. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12684–12694 (2021)

    2021

  9. [9]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Brooks, T., Holynski, A., Efros, A.A.: Instructpix2pix: Learning to follow image editing instructions. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 18392–18402 (2023)

    2023

  10. [10]

    In: Pro- ceedings of the Special Interest Group on Computer Graphics and Interactive Tech- niques Conference Conference Papers

    Careaga, C., Aksoy, Y.: Physically controllable relighting of photographs. In: Pro- ceedings of the Special Interest Group on Computer Graphics and Interactive Tech- niques Conference Conference Papers. pp. 1–10 (2025)

    2025

  11. [11]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Choi, J.M., Wang, A., Peers, P., Bhattad, A., Sengupta, R.: Scribblelight: Single image indoor relighting with scribbles. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5720–5731 (2025)

    2025

  12. [12]

    In: Proceedings of the 27th annual conference on Computer graphics and interactive techniques

    Debevec, P., Hawkins, T., Tchou, C., Duiker, H.P., Sarokin, W., Sagar, M.: Ac- quiring the reflectance field of a human face. In: Proceedings of the 27th annual conference on Computer graphics and interactive techniques. pp. 145–156 (2000)

    2000

  13. [13]

    ACM Transactions on Graph- ics (ToG)37(4), 1–15 (2018)

    Deschaintre, V., Aittala, M., Durand, F., Drettakis, G., Bousseau, A.: Single-image svbrdf capture with a rendering-aware deep network. ACM Transactions on Graph- ics (ToG)37(4), 1–15 (2018)

    2018

  14. [14]

    Bermano, A., Theobalt, C.: Practilight: Practical light control using foundational diffusion models

    Erel, Y., Dabral, R., Golyanik, V., H. Bermano, A., Theobalt, C.: Practilight: Practical light control using foundational diffusion models. ACM Transactions on Graphics (TOG)44(6), 1–11 (2025)

    2025

  15. [15]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Gao, C., Jiang, B., Li, X., Zhang, Y., Yu, Q.: Genesistex: Adapting image denois- ing diffusion to texture space. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4620–4629 (2024)

    2024

  16. [16]

    In: European Conference on Computer Vision

    Gao, J., Gu, C., Lin, Y., Li, Z., Zhu, H., Cao, X., Zhang, L., Yao, Y.: Relightable 3d gaussians: Realistic point cloud relighting with brdf decomposition and ray tracing. In: European Conference on Computer Vision. pp. 73–89. Springer (2024) Lume-Palette 17

    2024

  17. [17]

    IEEE Trans

    Goldman, D.B., Curless, B., Hertzmann, A., Seitz, S.M.: Shape and spatially- varying brdfs from photometric stereo. IEEE Trans. Pattern Anal. Mach. Intell. 32(6), 1060–1071 (2010).https://doi.org/10.1109/TPAMI.2009.102,https: //doi.org/10.1109/TPAMI.2009.102

    work page doi:10.1109/tpami.2009.102 2010

  18. [18]

    In: 2009 IEEE 12th Inter- national Conference on Computer Vision

    Grosse, R., Johnson, M.K., Adelson, E.H., Freeman, W.T.: Ground truth dataset and baseline evaluations for intrinsic image algorithms. In: 2009 IEEE 12th Inter- national Conference on Computer Vision. pp. 2335–2342. Ieee (2009)

    2009

  19. [19]

    arXiv preprint arXiv:2506.15673 (2025)

    He, K., Liang, R., Munkberg, J., Hasselgren, J., Vijaykumar, N., Keller, A., Fidler, S., Gilitschenski, I., Gojcic, Z., Wang, Z.: Unirelight: Learning joint decomposition and synthesis for video relighting. arXiv preprint arXiv:2506.15673 (2025)

    Pith/arXiv arXiv 2025

  20. [20]

    In: The Eleventh In- ternational Conference on Learning Representations (2023),https://openreview

    Hertz, A., Mokady, R., Tenenbaum, J., Aberman, K., Pritch, Y., Cohen-Or, D.: Prompt-to-prompt image editing with cross-attention control. In: The Eleventh In- ternational Conference on Learning Representations (2023),https://openreview. net/forum?id=_CDixzkzeyb

    2023

  21. [21]

    In: International Con- ference on Learning Representations (2022),https://openreview.net/forum?id= nZeVKeeFYf9

    Hu, E.J., yelong shen, Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W.: LoRA: Low-rank adaptation of large language models. In: International Con- ference on Learning Representations (2022),https://openreview.net/forum?id= nZeVKeeFYf9

    2022

  22. [22]

    arXiv preprint arXiv:2504.16086 (2025)

    Ji, G., Sawyer, A.O., Narasimhan, S.G.: Digital kitchen remodeling: editing and relighting intricate indoor scenes from a single panorama. arXiv preprint arXiv:2504.16086 (2025)

    Pith/arXiv arXiv 2025

  23. [23]

    Advances in Neural Information Processing Systems37, 141129–141152 (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. Advances in Neural Information Processing Systems37, 141129–141152 (2024)

    2024

  24. [24]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    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 re- lighting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 25096–25106 (2024)

    2024

  25. [25]

    In: Bengio, Y., Le- Cun, Y

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. In: Bengio, Y., Le- Cun, Y. (eds.) 2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14-16, 2014, Conference Track Proceedings (2014), http://arxiv.org/abs/1312.6114

    Pith/arXiv arXiv 2014

  26. [26]

    Kocsis,P.,Philip,J.,Sunkavalli,K.,Nießner,M.,Hold-Geoffroy,Y.:Lightit:Illumi- nationmodelingandcontrolfordiffusionmodels.In:ProceedingsoftheIEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9359–9369 (2024)

    2024

  27. [27]

    Labs, B.F.: FLUX.2: Frontier Visual Intelligence.https://bfl.ai/blog/flux-2 (2025)

    2025

  28. [28]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Li, Z., Shafiei, M., Ramamoorthi, R., Sunkavalli, K., Chandraker, M.: Inverse ren- dering for complex indoor scenes: Shape, spatially-varying lighting and svbrdf from a single image. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2475–2484 (2020)

    2020

  29. [29]

    In: European Conference on Computer Vision

    Li, Z., Shi, J., Bi, S., Zhu, R., Sunkavalli, K., Hašan, M., Xu, Z., Ramamoorthi, R., Chandraker, M.: Physically-based editing of indoor scene lighting from a single image. In: European Conference on Computer Vision. pp. 555–572. Springer (2022)

    2022

  30. [30]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Liang, R., Gojcic, Z., Ling, H., Munkberg, J., Hasselgren, J., Lin, C.H., Gao, J., Keller, A., Vijaykumar, N., Fidler, S., et al.: Diffusion renderer: Neural inverse and forward rendering with video diffusion models. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 26069–26080 (2025)

    2025

  31. [31]

    arXiv preprint arXiv:2601.15283 (2026) 18 C

    Liang, R., Müller, N., Weber, E., Zauss, D., Vijaykumar, N., Kontschieder, P., Richardt, C.: Luxremix: Lighting decomposition and remixing for indoor scenes. arXiv preprint arXiv:2601.15283 (2026) 18 C. Gao et al

    Pith/arXiv arXiv 2026

  32. [32]

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

    Lin,C.H.,Huang,J.B.,Li,Z.,Dong,Z.,Richardt,C.,Li,T.,Zollhöfer,M.,Kopf,J., Wang,S.,Kim,C.:IRIS:Inverserenderingofindoorscenesfromlowdynamicrange images. In: Conference on Computer Vision and Pattern Recognition (CVPR) (2025)

    2025

  33. [33]

    arXiv preprint arXiv:2511.10647 (2025)

    Lin, H., Chen, S., Liew, J.H., Chen, D.Y., Li, Z., Shi, G., Feng, J., Kang, B.: Depth anything 3: Recovering the visual space from any views. arXiv preprint arXiv:2511.10647 (2025)

    Pith/arXiv arXiv 2025

  34. [34]

    Advances in Neural Information Processing Systems38, 27798–27829 (2026)

    Lin, Y., Chen, Y.W., Tsai, Y.H., Clark, R., Yang, M.H.: Illumicraft: Unified ge- ometry and illumination diffusion for controllable video generation. Advances in Neural Information Processing Systems38, 27798–27829 (2026)

    2026

  35. [35]

    arXiv preprint arXiv:2210.02747 (2022)

    Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

    Pith/arXiv arXiv 2022

  36. [36]

    In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision

    Litman, Y., De la Torre, F., Tulsiani, S.: Lightswitch: Multi-view relighting with material-guided diffusion. In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision. pp. 27750–27759 (2025)

    2025

  37. [37]

    In: The Twelfth International Conference on Learning Representations (2024),https: //openreview.net/forum?id=MN3yH2ovHb

    Liu, Y., Lin, C., Zeng, Z., Long, X., Liu, L., Komura, T., Wang, W.: Sync- dreamer: Generating multiview-consistent images from a single-view image. In: The Twelfth International Conference on Learning Representations (2024),https: //openreview.net/forum?id=MN3yH2ovHb

    2024

  38. [38]

    arXiv preprint arXiv:1906.07751 (2019)

    Lombardi, S., Simon, T., Saragih, J., Schwartz, G., Lehrmann, A., Sheikh, Y.: Neural volumes: Learning dynamic renderable volumes from images. arXiv preprint arXiv:1906.07751 (2019)

    Pith/arXiv arXiv 1906

  39. [39]

    In: International Conferenceon LearningRepresentations(2019),https://openreview.net/forum? id=Bkg6RiCqY7

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: International Conferenceon LearningRepresentations(2019),https://openreview.net/forum? id=Bkg6RiCqY7

    2019

  40. [40]

    ACM Transactions on Graphics (TOG)44(4), 1–13 (2025)

    Lyu, L., Deschaintre, V., Hold-Geoffroy, Y., Hašan, M., Yoon, J.S., Leimkühler, T., Theobalt, C., Georgiev, I.: Intrinsicedit: Precise generative image manipulation in intrinsic space. ACM Transactions on Graphics (TOG)44(4), 1–13 (2025)

    2025

  41. [41]

    In: Proceed- ings of the Special Interest Group on Computer Graphics and Interactive Tech- niques Conference Conference Papers

    Magar, N., Hertz, A., Tabellion, E., Pritch, Y., Rav-Acha, A., Shamir, A., Hoshen, Y.: Lightlab: Controlling light sources in images with diffusion models. In: Proceed- ings of the Special Interest Group on Computer Graphics and Interactive Tech- niques Conference Conference Papers. pp. 1–11 (2025)

    2025

  42. [42]

    In: Proceedings of the AAAI conference on artificial intelligence

    Mou, C., Wang, X., Xie, L., Wu, Y., Zhang, J., Qi, Z., Shan, Y.: T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In: Proceedings of the AAAI conference on artificial intelligence. vol. 38, pp. 4296–4304 (2024)

    2024

  43. [43]

    In: 2019 IEEE International Conference on Computer Vision (ICCV) (Oct 2019)

    Murmann, L., Gharbi, M., Aittala, M., Durand, F.: A multi-illumination dataset of indoor object appearance. In: 2019 IEEE International Conference on Computer Vision (ICCV) (Oct 2019)

    2019

  44. [44]

    ACM Transactions on Graphics (ToG)38(6), 1–17 (2019)

    Nimier-David, M., Vicini, D., Zeltner, T., Jakob, W.: Mitsuba 2: A retargetable forward and inverse renderer. ACM Transactions on Graphics (ToG)38(6), 1–17 (2019)

    2019

  45. [45]

    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)

    2021

  46. [46]

    ACM Trans

    Philip, J., Gharbi, M., Zhou, T., Efros, A.A., Drettakis, G.: Multi-view relighting using a geometry-aware network. ACM Trans. Graph.38(4), 78–1 (2019)

    2019

  47. [47]

    Philip, J., Morgenthaler, S., Gharbi, M., Drettakis, G.: Free-viewpoint indoor neu- ralrelightingfrommulti-viewstereo.ACMTransactionsonGraphics(TOG)40(5), 1–18 (2021) Lume-Palette 19

    2021

  48. [48]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Phongthawee, P., Chinchuthakun, W., Sinsunthithet, N., Jampani, V., Raj, A., Khungurn, P., Suwajanakorn, S.: Diffusionlight: Light probes for free by painting a chrome ball. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 98–108 (2024)

    2024

  49. [49]

    In: Computer Graphics Forum

    Poirier-Ginter, Y., Gauthier, A., Phillip, J., Lalonde, J.F., Drettakis, G.: A dif- fusion approach to radiance field relighting using multi-illumination synthesis. In: Computer Graphics Forum. vol. 43, p. e15147. Wiley Online Library (2024)

    2024

  50. [50]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Ponglertnapakorn, P., Tritrong, N., Suwajanakorn, S.: Difareli: Diffusion face re- lighting. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 22646–22657 (2023)

    2023

  51. [51]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Roberts, M., Ramapuram, J., Ranjan, A., Kumar, A., Bautista, M.A., Paczan, N., Webb, R., Susskind, J.M.: Hypersim: A photorealistic synthetic dataset for holis- tic indoor scene understanding. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 10912–10922 (2021)

    2021

  52. [52]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  53. [53]

    In: European Conference on Computer Vision

    Rudnev, V., Elgharib, M., Smith, W., Liu, L., Golyanik, V., Theobalt, C.: Nerf for outdoor scene relighting. In: European Conference on Computer Vision. pp. 615–631. Springer (2022)

    2022

  54. [54]

    In: ACM SIGGRAPH 2022 confer- ence proceedings

    Saharia, C., Chan, W., Chang, H., Lee, C., Ho, J., Salimans, T., Fleet, D., Norouzi, M.: Palette: Image-to-image diffusion models. In: ACM SIGGRAPH 2022 confer- ence proceedings. pp. 1–10 (2022)

    2022

  55. [55]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Sengupta, S., Gu, J., Kim, K., Liu, G., Jacobs, D.W., Kautz, J.: Neural inverse rendering of an indoor scene from a single image. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 8598–8607 (2019)

    2019

  56. [56]

    In: The Twelfth International Conference on Learning Representations (2024),https://openreview.net/forum?id=FUgrjq2pbB

    Shi, Y., Wang, P., Ye, J., Mai, L., Li, K., Yang, X.: MVDream: Multi-view dif- fusion for 3d generation. In: The Twelfth International Conference on Learning Representations (2024),https://openreview.net/forum?id=FUgrjq2pbB

    2024

  57. [57]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Srinivasan, P.P., Deng, B., Zhang, X., Tancik, M., Mildenhall, B., Barron, J.T.: Nerv: Neural reflectance and visibility fields for relighting and view synthesis. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 7495–7504 (2021)

    2021

  58. [58]

    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. Graph38(4), 1–12 (2019)

    2019

  59. [59]

    In: Advances in Neural Information Processing Systems (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: Advances in Neural Information Processing Systems (2025)

    2025

  60. [60]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Trevithick, A., Paiss, R., Henzler, P., Verbin, D., Wu, R., Alzayer, H., Gao, R., Poole,B.,Barron,J.T.,Holynski,A.,etal.:Simvs:Simulatingworldinconsistencies for robust view synthesis. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 16464–16474 (2025)

    2025

  61. [61]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wang, R., Zhang, Q., Fu, C.W., Shen, X., Zheng, W.S., Jia, J.: Underexposed photo enhancement using deep illumination estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6849–6857 (2019)

    2019

  62. [62]

    In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision

    Wang, Z., Philion, J., Fidler, S., Kautz, J.: Learning indoor inverse rendering with 3d spatially-varying lighting. In: Proceedings of the IEEE/CVF International Con- ference on Computer Vision. pp. 12538–12547 (2021) 20 C. Gao et al

    2021

  63. [63]

    arXiv preprint arXiv:2508.02324 (2025)

    Wu, C., Li, J., Zhou, J., Lin, J., Gao, K., Yan, K., Yin, S.m., Bai, S., Xu, X., Chen, Y., et al.: Qwen-image technical report. arXiv preprint arXiv:2508.02324 (2025)

    Pith/arXiv arXiv 2025

  64. [64]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Xing, X., Groh, K., Karaoglu, S., Gevers, T., Bhattad, A.: Luminet: Latent in- trinsics meets diffusion models for indoor scene relighting. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 442–452 (2025)

    2025

  65. [65]

    arXiv preprint arXiv:2511.13684 (2025)

    Ye, J., Zhuang, J., Mu, L., Zheng, W., Hu, J., Zou, X., Wang, J., Hu, H.: Training- free multi-view extension of ic-light for textual position-aware scene relighting. arXiv preprint arXiv:2511.13684 (2025)

    arXiv 2025

  66. [66]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Yeshwanth, C., Liu, Y.C., Nießner, M., Dai, A.: Scannet++: A high-fidelity dataset of 3d indoor scenes. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 12–22 (2023)

    2023

  67. [67]

    In: ACM SIGGRAPH 2024 Conference Papers

    Zeng, C., Dong, Y., Peers, P., Kong, Y., Wu, H., Tong, X.: Dilightnet: Fine-grained lighting control for diffusion-based image generation. In: ACM SIGGRAPH 2024 Conference Papers. pp. 1–12 (2024)

    2024

  68. [68]

    In: ACM SIGGRAPH 2024 Conference Papers

    Zeng, Z., Deschaintre, V., Georgiev, I., Hold-Geoffroy, Y., Hu, Y., Luan, F., Yan, L.Q., Hašan, M.: Rgb↔x: Image decomposition and synthesis using material- and lighting-aware diffusion models. In: ACM SIGGRAPH 2024 Conference Papers. SIGGRAPH ’24, Association for Computing Machinery, New York, NY, USA (2024).https://doi.org/10.1145/3641519.3657445,https:...

    doi:10.1145/3641519.3657445 2024

  69. [69]

    In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhang, K., Luan, F., Wang, Q., Bala, K., Snavely, N.: Physg: Inverse rendering with spherical gaussians for physics-based material editing and relighting. In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5453–5462 (2021)

    2021

  70. [70]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3836–3847 (2023)

    2023

  71. [71]

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

    2025

  72. [72]

    ACM Transactions on Graphics (ToG)40(6), 1–18 (2021)

    Zhang, X., Srinivasan, P.P., Deng, B., Debevec, P., Freeman, W.T., Barron, J.T.: Nerfactor: Neural factorization of shape and reflectance under an unknown illumi- nation. ACM Transactions on Graphics (ToG)40(6), 1–18 (2021)

    2021

  73. [73]

    arXiv preprint arXiv:2508.14930 (2025)

    Zhao, H., Akers, J., Elmieh, B., Kemelmacher-Shlizerman, I.: Hybrelighter: Com- bining deep anisotropic diffusion and scene reconstruction for on-device real-time relighting in mixed reality. arXiv preprint arXiv:2508.14930 (2025)

    Pith/arXiv arXiv 2025

  74. [74]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhou, H., Hadap, S., Sunkavalli, K., Jacobs, D.W.: Deep single-image portrait relighting. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 7194–7202 (2019)

    2019

  75. [75]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhu, R., Li, Z., Matai, J., Porikli, F., Chandraker, M.: Irisformer: Dense vision transformers for single-image inverse rendering in indoor scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 2822–2831 (2022)

    2022