pith. sign in

arxiv: 2605.19613 · v1 · pith:XFSRYIGSnew · submitted 2026-05-19 · 💻 cs.CV

White-Balance First, Adjust Later: Cross-Camera Color Constancy via Vision-Language Evaluation

Shuwei Li , Lei Tan , Robby T. Tan This is my paper

Pith reviewed 2026-05-20 06:52 UTC · model grok-4.3

classification 💻 cs.CV
keywords color constancycross-camera generalizationvision-language modeliterative refinementwhite balanceilluminant estimationperceptual feedback
0
0 comments X

The pith

VLM-CC reframes color constancy as iterative VLM feedback to correct residual casts after white balancing for cross-camera robustness.

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

The paper introduces VLM-CC as a feedback-guided framework that treats color constancy as an iterative refinement process rather than direct illuminant regression from raw images. At each step the current estimate white-balances the input into a pseudo-sRGB image; a lightweight LoRA-tuned vision-language model then identifies the dominant remaining color cast and supplies qualitative feedback that is mapped to a residual illumination direction to update the estimate. The loop repeats until the image appears neutral. This design is motivated by the observation that learned models overfit to the color response of the training camera and therefore degrade on images from unseen sensors. The authors report that the resulting method delivers state-of-the-art cross-camera performance on multiple standard datasets.

Core claim

VLM-CC formulates color constancy as an iterative perceptual feedback problem: after each white-balance step the image is converted to pseudo-sRGB, a LoRA-tuned VLM evaluates the dominant residual color cast, and the qualitative assessment is converted into a residual illumination direction that updates the illuminant estimate until convergence.

What carries the argument

The iterative loop that uses VLM perceptual evaluation of residual color cast in pseudo-sRGB images to generate qualitative feedback mapped to red/green/blue residual illumination directions for successive illuminant updates.

If this is right

  • Direct RGB regression can be replaced by perceptual feedback without loss of accuracy when cross-camera generalization is required.
  • Illuminant estimation accuracy improves on unseen cameras because the VLM operates on perceptual appearance rather than camera-specific raw statistics.
  • The same iterative correction strategy applies to any downstream task that needs consistent object colors across varying sensors.

Where Pith is reading between the lines

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

  • The method could reduce reliance on large camera-specific training sets if the VLM feedback proves stable across a wide range of sensors.
  • Extending the feedback vocabulary beyond red/green/blue casts might allow finer control over more complex illumination spectra.
  • The approach suggests a broader pattern in which vision-language models serve as perceptual critics inside classical computer-vision pipelines.

Load-bearing premise

A lightweight LoRA-tuned VLM can reliably detect the dominant residual color cast in a pseudo-sRGB image and translate that qualitative judgment into the correct residual illumination direction for the next update.

What would settle it

A controlled experiment in which the VLM is shown to misclassify the dominant residual cast on images from a held-out camera, causing the iterative updates to increase rather than reduce error relative to a non-iterative baseline.

Figures

Figures reproduced from arXiv: 2605.19613 by Lei Tan, Robby T. Tan, Shuwei Li.

Figure 1
Figure 1. Figure 1: Rather than directly predicting a light color, our method first white-balances the image then later updates the light estimate via [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our proposed framework. Given a raw input image, we first apply white balance using the current illuminant estimate and convert the result to sRGB for VLM processing. A pretrained VLM extracts semantic color priors from the pseudo-sRGB image, identifying objects whose inherent colors are reliable under neutral light. A LoRA-finetuned VLM then predicts the dominant residual light color label (re… view at source ↗
Figure 3
Figure 3. Figure 3: Finetuning pipeline of VLM. Given a raw image, we first apply light-color augmentation in camera color space and con￾vert the results to sRGB. These images are processed by a LoRA￾finetuned [36] VLM, using the same color-prior prompting strat￾egy as in the inference pipeline. The model predicts the domi￾nant residual light color (red, green, or blue), supervised by the ground-truth illuminant direction. A … view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative example of our iterative correction process. The scene contains large wooden surfaces, which leads CCMNet [41] toward an over-red illuminant estimate. As a result, its white-balanced result appears blue. Our method starts from Gray-World [15] initialization that is also biased by the wood, but iteratively refines the estimate through feedback and converges to 0.57◦ . The right plot shows the tr… view at source ↗
read the original abstract

Color constancy aims to keep object colors consistent under varying illumination. Cross-camera generalization in color constancy remains challenging because learning-based models often overfit to the color response characteristics of the training camera, resulting in degraded performance on images captured by other cameras. We propose VLM-CC, a feedback-guided framework that formulates color constancy as an iterative refinement process. Instead of directly estimating the illuminant from raw input, VLM-CC performs iterative correction driven by vision-language model (VLM)-based evaluation. At each iteration, the image is white-balanced using the current estimate and converted to pseudo-sRGB. A lightweight LoRA-tuned VLM then assesses the corrected image, identifying the dominant residual color cast and providing qualitative feedback. This feedback is mapped to a residual illumination direction (red, green, or blue) and used to update the illuminant estimate until convergence. Our key idea is to reframe color constancy as an iterative perceptual feedback problem, leveraging VLM evaluation instead of direct RGB regression. By replacing direct RGB estimation with VLM-guided perceptual feedback, VLM-CC achieves state-of-the-art robustness in cross-camera color constancy across multiple datasets. Code will be available at https://github.com/NothingIknow/VLM-CC.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces VLM-CC, a feedback-guided iterative framework for cross-camera color constancy. Instead of direct illuminant regression from raw input, the method white-balances the image with the current estimate, converts to pseudo-sRGB, queries a lightweight LoRA-tuned VLM to identify the dominant residual color cast, maps the qualitative VLM output to a residual RGB illumination direction, and updates the estimate until convergence. The central claim is that reframing color constancy as VLM-driven perceptual feedback yields state-of-the-art robustness across unseen cameras and multiple datasets.

Significance. If the VLM feedback step proves reliable, the approach could meaningfully improve generalization in color constancy by sidestepping camera-specific overfitting that plagues direct regression methods. The iterative perceptual-correction paradigm is a fresh direction that may transfer to other low-level vision tasks where qualitative visual assessment is easier than precise numeric regression. Code release is noted as a reproducibility strength.

major comments (2)
  1. [§3] §3 (Iterative Refinement): The mapping from VLM qualitative feedback (e.g., 'residual red cast') to a residual illumination direction is load-bearing for convergence and the claimed robustness gain, yet no quantitative VLM accuracy metrics, confusion matrix, or ablation on mapping errors are supplied. Systematic misidentification due to VLM color biases or pseudo-sRGB artifacts would invalidate the iterative updates.
  2. [§4] §4 (Experiments): The abstract asserts SOTA cross-camera performance, but the provided description supplies no specific quantitative tables, baseline comparisons, dataset details, or failure-case analysis isolating the VLM component. Without these, the robustness advantage over direct RGB regression remains unverified.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'Code will be available at https://github.com/NothingIknow/VLM-CC' should include a commit hash or release tag for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential of the iterative VLM-driven perceptual feedback paradigm for improving cross-camera generalization in color constancy. We address each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [§3] §3 (Iterative Refinement): The mapping from VLM qualitative feedback (e.g., 'residual red cast') to a residual illumination direction is load-bearing for convergence and the claimed robustness gain, yet no quantitative VLM accuracy metrics, confusion matrix, or ablation on mapping errors are supplied. Systematic misidentification due to VLM color biases or pseudo-sRGB artifacts would invalidate the iterative updates.

    Authors: We agree that validating the VLM feedback step is essential. Section 3 of the manuscript defines the mapping explicitly: the VLM's qualitative output (e.g., dominant residual red, green, or blue cast) is converted to a unit vector adjustment in RGB space that is added to the current illuminant estimate, with a fixed step size until the VLM reports no residual cast. To strengthen this, the revised manuscript will add a quantitative evaluation of VLM color-cast identification accuracy on a held-out set of pseudo-sRGB images with synthetically introduced residual casts. This will include precision/recall per cast category and a confusion matrix. We will also report an ablation measuring the effect of controlled mapping errors (e.g., 10-20% misidentification rate) on final angular error. These additions directly address the concern about systematic biases or pseudo-sRGB artifacts. revision: yes

  2. Referee: [§4] §4 (Experiments): The abstract asserts SOTA cross-camera performance, but the provided description supplies no specific quantitative tables, baseline comparisons, dataset details, or failure-case analysis isolating the VLM component. Without these, the robustness advantage over direct RGB regression remains unverified.

    Authors: We apologize if the experimental presentation was insufficiently detailed in the reviewed version. The full manuscript's Section 4 reports results on standard cross-camera protocols using the NUS 8-camera dataset and the Gehler-Shi dataset, with tables comparing mean and median angular error against direct-regression baselines (e.g., FC4, C5, and recent learning-based methods). Dataset splits and camera-holdout details are described in Section 4.1. In the revision we will expand this section with (i) an explicit failure-case analysis that isolates VLM misclassifications, (ii) an ablation that removes the iterative VLM feedback and compares against a single-pass version, and (iii) clearer side-by-side tables highlighting the cross-camera robustness gain. These changes will make the quantitative support for the SOTA claim fully verifiable. revision: yes

Circularity Check

0 steps flagged

No significant circularity; iterative VLM feedback is an external mechanism, not a self-referential fit

full rationale

The paper proposes an iterative refinement loop in which a LoRA-tuned VLM evaluates a white-balanced pseudo-sRGB image and supplies qualitative feedback that is then mapped to a residual illumination direction. This mapping and the convergence criterion are presented as design choices rather than quantities derived from or fitted to the target illuminant estimate itself. No equations, self-citations, or uniqueness theorems are invoked in the provided abstract that would reduce the claimed cross-camera robustness to a re-labeling of the input data or to a parameter tuned on the evaluation set. The central performance claim is therefore an empirical statement about the external VLM's reliability, not a result forced by construction from the paper's own inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that VLMs can provide accurate perceptual color-cast feedback and on the design choice of LoRA tuning, which introduces tunable parameters whose values are not detailed in the abstract.

free parameters (1)
  • LoRA adaptation parameters
    Lightweight LoRA tuning of the VLM is mentioned; these parameters are fitted or chosen to enable the evaluation step.
axioms (1)
  • domain assumption VLM can reliably detect and describe residual color casts in pseudo-sRGB images
    The feedback loop depends on this capability of the vision-language model.

pith-pipeline@v0.9.0 · 5753 in / 1341 out tokens · 36177 ms · 2026-05-20T06:52:53.964709+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

71 extracted references · 71 canonical work pages · 2 internal anchors

  1. [1]

    Adobe photoshop

    Adobe Inc. Adobe photoshop. 2

    work page

  2. [2]

    Sensor-independent illumination estimation for dnn models

    Mahmoud Afifi and Michael S Brown. Sensor-independent illumination estimation for dnn models. InBritish Machine Vision Conference (BMVC), 2019. 2, 3, 6, 7

    work page 2019

  3. [3]

    Cross-camera convolutional color constancy

    Mahmoud Afifi, Jonathan T Barron, Chloe LeGendre, Yun- Ta Tsai, and Francois Bleibel. Cross-camera convolutional color constancy. InProceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pages 1981–1990,

    work page 1981

  4. [4]

    Auto white-balance correction for mixed-illuminant scenes

    Mahmoud Afifi, Marcus A Brubaker, and Michael S Brown. Auto white-balance correction for mixed-illuminant scenes. InProceedings of the IEEE/CVF Winter Conference on Ap- plications of Computer Vision, pages 1210–1219, 2022. 2

    work page 2022

  5. [5]

    Optimiz- ing illuminant estimation in dual-exposure hdr imaging

    Mahmoud Afifi, Zhenhua Hu, and Liang Liang. Optimiz- ing illuminant estimation in dual-exposure hdr imaging. In Computer Vision – ECCV 2024, 2024. 2

    work page 2024

  6. [6]

    Alejandro Parraga

    Arash Akbarinia and C. Alejandro Parraga. Colour con- stancy beyond the classical receptive field.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 40(9): 2081–2094, 2018. 6

    work page 2081

  7. [7]

    Qwen2.5-VL Technical Report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2.5-vl technical report.arXiv preprint arXiv:2502.13923, 2025. 3, 5, 8

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Computational color constancy: taking the- ory into practice

    Kobus Barnard. Computational color constancy: taking the- ory into practice. 1995. 2

    work page 1995

  9. [9]

    A comparison of computational color constancy algorithms

    Kobus Barnard, Vlad Cardei, and Brian Funt. A comparison of computational color constancy algorithms. i: Methodol- ogy and experiments with synthesized data.IEEE transac- tions on Image Processing, 11(9):972–984, 2002. 7

    work page 2002

  10. [10]

    Convolutional color constancy

    Jonathan T Barron. Convolutional color constancy. InPro- ceedings of the IEEE International Conference on Computer Vision, pages 379–387, 2015. 2, 3

    work page 2015

  11. [11]

    Fast fourier color con- stancy

    Jonathan T Barron and Yun-Ta Tsai. Fast fourier color con- stancy. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 886– 894, 2017. 2, 6, 7

    work page 2017

  12. [12]

    Quasi-unsupervised color constancy

    Simone Bianco and Claudio Cusano. Quasi-unsupervised color constancy. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pages 12212–12221, 2019. 2, 6

    work page 2019

  13. [13]

    Color constancy using cnns

    Simone Bianco, Claudio Cusano, and Raimondo Schettini. Color constancy using cnns. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2015. 2, 6

    work page 2015

  14. [14]

    Analysis of the retinex theory of color vision.JOSA A, 3(10):1651–1661,

    David H Brainard and Brian A Wandell. Analysis of the retinex theory of color vision.JOSA A, 3(10):1651–1661,

    work page

  15. [15]

    A spatial processor model for object colour perception.Journal of the Franklin institute, 310(1): 1–26, 1980

    Gershon Buchsbaum. A spatial processor model for object colour perception.Journal of the Franklin institute, 310(1): 1–26, 1980. 2, 4, 5, 6, 7, 8

    work page 1980

  16. [16]

    Color constancy by learning to predict chromaticity from luminance.Advances in Neural Informa- tion Processing Systems, 28, 2015

    Ayan Chakrabarti. Color constancy by learning to predict chromaticity from luminance.Advances in Neural Informa- tion Processing Systems, 28, 2015. 7

    work page 2015

  17. [17]

    Color constancy with spatio-spectral statistics.IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(8):1509– 1519, 2012

    Ayan Chakrabarti, Keigo Hirakawa, and Todd Zickler. Color constancy with spatio-spectral statistics.IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(8):1509– 1519, 2012. 2

    work page 2012

  18. [18]

    Gcc: Generative color constancy via diffusing a color checker

    Chen-Wei Chang, Cheng-De Fan, Chia-Che Chang, Yi- Chen Lo, Yu-Chee Tseng, Jiun-Long Huang, and Yu-Lun Liu. Gcc: Generative color constancy via diffusing a color checker. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 10868–10878, 2025. 2, 3, 6, 7

    work page 2025

  19. [19]

    Internvl: Scaling up vision foundation mod- els and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, et al. Internvl: Scaling up vision foundation mod- els and aligning for generic visual-linguistic tasks. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 24185–24198, 2024. 3, 8

    work page 2024

  20. [20]

    Il- luminant estimation for color constancy: why spatial-domain methods work and the role of the color distribution.JOSA A, 31(5):1049–1058, 2014

    Dongliang Cheng, Dilip K Prasad, and Michael S Brown. Il- luminant estimation for color constancy: why spatial-domain methods work and the role of the color distribution.JOSA A, 31(5):1049–1058, 2014. 2, 6, 7

    work page 2014

  21. [21]

    Generative models for multi-illumination color constancy

    Partha Das, Yang Liu, Sezer Karaoglu, and Theo Gevers. Generative models for multi-illumination color constancy. InProceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 1194–1203,

    work page

  22. [22]

    The cube++ illumination estimation dataset.IEEE Access, 8:227511–227527, 2020

    Egor Ershov, Alexey Savchik, Illya Semenkov, Nikola Bani´c, Alexander Belokopytov, Daria Senshina, Karlo Ko ˇsˇcevi´c, Marko Subaˇsi´c, and Sven Lonˇcari´c. The cube++ illumination estimation dataset.IEEE Access, 8:227511–227527, 2020. 6

    work page 2020

  23. [23]

    Advanc- ing sequential numerical prediction in autoregressive mod- els

    Xiang Fei, Jinghui Lu, Qi Sun, Hao Feng, Yanjie Wang, Wei Shi, An-Lan Wang, Jingqun Tang, and Can Huang. Advanc- ing sequential numerical prediction in autoregressive mod- els. InProceedings of the 63rd Annual Meeting of the Asso- ciation for Computational Linguistics (Volume 2: Short Pa- pers), 2025. 4

    work page 2025

  24. [24]

    Finlayson and Gerald Schaefer

    Graham D. Finlayson and Gerald Schaefer. Convex and non- convex illuminant constraints for dichromatic colour con- stancy. InProceedings of the 2001 IEEE Computer Soci- ety Conference on Computer Vision and Pattern Recognition (CVPR), pages 598–604. IEEE, 2001. 1

    work page 2001

  25. [25]

    Solving for colour constancy using a constrained dichromatic reflection model.International Journal of Computer Vision, 42:127– 144, 2001

    Graham D Finlayson and Gerald Schaefer. Solving for colour constancy using a constrained dichromatic reflection model.International Journal of Computer Vision, 42:127– 144, 2001. 1

    work page 2001

  26. [26]

    Shades of gray and colour constancy

    Graham D Finlayson and Elisabetta Trezzi. Shades of gray and colour constancy. InColor and Imaging Conference, pages 37–41. Society for Imaging Science and Technology,

    work page

  27. [27]

    Efficient color constancy with local surface re- flectance statistics

    Shaobing Gao, Wangwang Han, Kaifu Yang, Chaoyi Li, and Yongjie Li. Efficient color constancy with local surface re- flectance statistics. InEuropean Conference on Computer Vision, pages 158–173. Springer, 2014. 6, 7

    work page 2014

  28. [28]

    Bayesian color constancy revisited

    Peter Vincent Gehler, Carsten Rother, Andrew Blake, Tom Minka, and Toby Sharp. Bayesian color constancy revisited. In2008 IEEE Conference on Computer Vision and Pattern Recognition, pages 1–8. IEEE, 2008. 2, 6, 7

    work page 2008

  29. [29]

    Exploring the frontier of vision- language models: A survey of current methodologies and future directions

    Akash Ghosh, Arkadeep Acharya, Sriparna Saha, Vinija Jain, and Aman Chadha. Exploring the frontier of vision- language models: A survey of current methodologies and future directions.arXiv preprint arXiv:2404.07214, 2024. 3

    work page arXiv 2024

  30. [30]

    Color constancy using nat- ural image statistics and scene semantics.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 33(4): 687–698, 2011

    Arjan Gijsenij and Theo Gevers. Color constancy using nat- ural image statistics and scene semantics.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 33(4): 687–698, 2011. 2

    work page 2011

  31. [31]

    Im- proving color constancy by photometric edge weighting

    Arjan Gijsenij, Theo Gevers, and Joost Van De Weijer. Im- proving color constancy by photometric edge weighting. IEEE Transactions on Pattern Analysis and Machine Intel- ligence, 34(5):918–929, 2012. 2

    work page 2012

  32. [32]

    Color names in vision-language models

    Alexandra Gomez-Villa, Pablo Hern ´andez-C´amara, Muham- mad Atif Butt, Valero Laparra, Jesus Malo, and Javier Vazquez-Corral. Color names in vision-language models. arXiv preprint arXiv:2509.22524, 2025. 3

    work page arXiv 2025

  33. [33]

    Memory modulates color appearance

    Thorsten Hansen, Maria Olkkonen, Sebastian Walter, and Karl R Gegenfurtner. Memory modulates color appearance. Nature neuroscience, 9(11):1367–1368, 2006. 2

    work page 2006

  34. [34]

    Finlayson, Arjan Gijsenij, Peter Gehler, Simone Bianco, Brian Funt, Mark Drew, and Lilong Shi

    Ghalia Hemrit, Graham D. Finlayson, Arjan Gijsenij, Peter Gehler, Simone Bianco, Brian Funt, Mark Drew, and Lilong Shi. Rehabilitating the colorchecker dataset for illuminant estimation. InColor and Imaging Conference, pages 350– 353, 2018. 6

    work page 2018

  35. [35]

    A multi-hypothesis approach to color constancy

    Daniel Hernandez-Juarez, Sarah Parisot, Benjamin Busam, Ales Leonardis, Gregory Slabaugh, and Steven McDonagh. A multi-hypothesis approach to color constancy. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2270–2280, 2020. 2, 3

    work page 2020

  36. [36]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. InIn- ternational Conference on Learning Representations (ICLR),

    work page

  37. [37]

    Fc4: Fully convolutional color constancy with confidence-weighted pooling

    Yuanming Hu, Baoyuan Wang, and Stephen Lin. Fc4: Fully convolutional color constancy with confidence-weighted pooling. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 4085– 4094, 2017. 2, 6, 7

    work page 2017

  38. [38]

    Wavediff-cc: Wavelet-based color constancy in diffusion models

    Linkai Huang, Hongbo Huang, Hui Wang, Longfei Xu, Lishan Wu, Shasha Wang, and Hui Wang. Wavediff-cc: Wavelet-based color constancy in diffusion models. InPattern Recognition and Computer Vision: 8th Chinese Conference, PRCV 2025, Shanghai, China, October 15–18, 2025, Proceedings, Part IX, page 284–298, Berlin, Heidel- berg, 2026. Springer-Verlag. 2

    work page 2025

  39. [39]

    Unsupervised night image enhancement: When layer decomposition meets light-effects suppression

    Yeying Jin, Wenhan Yang, and Robby T Tan. Unsupervised night image enhancement: When layer decomposition meets light-effects suppression. InEuropean Conference on Com- puter Vision, pages 404–421. Springer, 2022. 2

    work page 2022

  40. [40]

    Enhancing visibility in nighttime haze images using guided apsf and gradient adaptive convolution

    Yeying Jin, Beibei Lin, Wending Yan, Yuan Yuan, Wei Ye, and Robby T Tan. Enhancing visibility in nighttime haze images using guided apsf and gradient adaptive convolution. InProceedings of the 31st ACM International Conference on Multimedia, pages 2446–2457, 2023. 2

    work page 2023

  41. [41]

    Brown, and Seon Joo Kim

    Dongyoung Kim, Mahmoud Afifi, Dongyun Kim, Michael S. Brown, and Seon Joo Kim. Ccmnet: Leveraging calibrated color correction matrices for cross-camera color constancy. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 6198–6208,

    work page

  42. [42]

    Cross-dataset color constancy revisited using sensor-to- sensor transfer

    Samu Koskinen, Dan Yang, and Joni-Kristian K ¨am¨ar¨ainen. Cross-dataset color constancy revisited using sensor-to- sensor transfer. InProceedings of the British Machine Vision Conference (BMVC), 2020. 6

    work page 2020

  43. [43]

    Intel-tau: A color con- stancy dataset.IEEE access, 9:39560–39567, 2021

    Firas Laakom, Jenni Raitoharju, Jarno Nikkanen, Alexan- dros Iosifidis, and Moncef Gabbouj. Intel-tau: A color con- stancy dataset.IEEE access, 9:39560–39567, 2021. 6

    work page 2021

  44. [44]

    Edwin H. Land. The retinex theory of color vision.Scientific American, 237(6):108–128, 1977. 2, 6, 7

    work page 1977

  45. [45]

    Lightness and retinex theory.Josa, 61(1):1–11, 1971

    Edwin H Land and John J McCann. Lightness and retinex theory.Josa, 61(1):1–11, 1971. 2

    work page 1971

  46. [46]

    Exposing numeracy gaps: A benchmark to eval- uate fundamental numerical abilities in large language mod- els

    Haoyang Li, Xuejia Chen, Zhanchao Xu, Darian Li, Nicole Hu, Fei Teng, Yiming Li, Luyu Qiu, Chen Jason Zhang, Li Qing, et al. Exposing numeracy gaps: A benchmark to eval- uate fundamental numerical abilities in large language mod- els. InFindings of the Association for Computational Lin- guistics: ACL 2025, pages 20004–20026, 2025. 4

    work page 2025

  47. [47]

    Shuwei Li and Robby T. Tan. Nightcc: Nighttime color constancy via adaptive channel masking. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25522–25531, 2024. 2

    work page 2024

  48. [48]

    Brown, and Robby T

    Shuwei Li, Jikai Wang, Michael S. Brown, and Robby T. Tan. Mimt: Multi-illuminant color constancy via multi-task local surface and light color learning, 2023. 6

    work page 2023

  49. [49]

    Colorbench: Can vlms see and under- stand the colorful world? a comprehensive benchmark for color perception, reasoning, and robustness.arXiv preprint arXiv:2504.10514, 2025

    Yijun Liang, Ming Li, Chenrui Fan, Ziyue Li, Dang Nguyen, Kwesi Cobbina, Shweta Bhardwaj, Jiuhai Chen, Fuxiao Liu, and Tianyi Zhou. Colorbench: Can vlms see and under- stand the colorful world? a comprehensive benchmark for color perception, reasoning, and robustness.arXiv preprint arXiv:2504.10514, 2025. 3, 4, 6

    work page arXiv 2025

  50. [50]

    Clcc: Contrastive learning for color constancy

    Yi-Chen Lo, Chia-Che Chang, Hsuan-Chao Chiu, Yu-Hao Huang, Chia-Ping Chen, Yu-Lin Chang, and Kevin Jou. Clcc: Contrastive learning for color constancy. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8053–8063, 2021. 2, 6, 7

    work page 2021

  51. [51]

    Decoupled weight de- cay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight de- cay regularization. InInternational Conference on Learning Representations (ICLR), 2019. 6

    work page 2019

  52. [52]

    Formulating Camera-Adaptive Color Constancy as a Few-shot Meta-Learning Problem

    Steven McDonagh, Sarah Parisot, Fengwei Zhou, Xing Zhang, Ales Leonardis, Zhenguo Li, and Gregory Slabaugh. Formulating camera-adaptive color constancy as a few-shot meta-learning problem.arXiv preprint arXiv:1811.11788,

    work page internal anchor Pith review Pith/arXiv arXiv

  53. [53]

    Approaching the compu- tational color constancy as a classification problem through deep learning.Pattern Recognition, 61:405–416, 2017

    Seonghyeon Oh and Seon Joo Kim. Approaching the compu- tational color constancy as a classification problem through deep learning.Pattern Recognition, 61:405–416, 2017. 2

    work page 2017

  54. [54]

    Color appearance of familiar objects: Effects of object shape, texture, and illumination changes.Journal of vision, 8(5):13–13, 2008

    Maria Olkkonen, Thorsten Hansen, and Karl R Gegenfurt- ner. Color appearance of familiar objects: Effects of object shape, texture, and illumination changes.Journal of vision, 8(5):13–13, 2008. 2

    work page 2008

  55. [55]

    Recurrent color constancy

    Yanlin Qian, Ke Chen, Jarno Nikkanen, Joni-Kristian K¨am¨ar¨ainen, and Jiri Matas. Recurrent color constancy. In Proceedings of the IEEE International Conference on Com- puter Vision, pages 5458–5466, 2017. 2

    work page 2017

  56. [56]

    On finding gray pixels

    Yanlin Qian, Joni-Kristian K¨am¨ar¨ainen, Jarno Nikkanen, and Jiri Matas. On finding gray pixels. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8062–8070, 2019. 2, 6, 7

    work page 2019

  57. [57]

    Revisiting gray pixel for sta- tistical illumination estimation

    Yanlin Qian, Said Pertuz, Jarno Nikkanen, Joni-Kristian K¨am¨ar¨ainen, and Jiri Matas. Revisiting gray pixel for sta- tistical illumination estimation. InProceedings of the Inter- national Conference on Computer Vision Theory and Appli- cations (VISAPP), 2019. 2, 7

    work page 2019

  58. [58]

    Cascading convolutional temporal color constancy.Journal of Electronic Imaging, 32(01), 2023

    Matteo Rizzo, Cristina Conati, Daesik Jang, and Hui Hu. Cascading convolutional temporal color constancy.Journal of Electronic Imaging, 32(01), 2023. 2

    work page 2023

  59. [59]

    Deep special- ized network for illuminant estimation

    Wu Shi, Chen Change Loy, and Xiaoou Tang. Deep special- ized network for illuminant estimation. InEuropean con- ference on computer vision, pages 371–387. Springer, 2016. 2

    work page 2016

  60. [60]

    Oleksii Sidorov. Conditional gans for multi-illuminant color constancy: Revolution or yet another approach? InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 0–0, 2019. 2

    work page 2019

  61. [61]

    Transfer learning for color constancy via statistic perspective

    Yuxiang Tang, Xuejing Kang, Chunxiao Li, Zhaowen Lin, and Anlong Ming. Transfer learning for color constancy via statistic perspective. InProceedings of the AAAI Conference on Artificial Intelligence, pages 2361–2369, 2022. 2

    work page 2022

  62. [62]

    Multichannel vision system for estimating surface and illumination functions.JOSA A, 13(11):2163– 2173, 1996

    Shoji Tominaga. Multichannel vision system for estimating surface and illumination functions.JOSA A, 13(11):2163– 2173, 1996. 1

    work page 1996

  63. [63]

    Edge- based color constancy.IEEE Transactions on image process- ing, 16(9):2207–2214, 2007

    Joost Van De Weijer, Theo Gevers, and Arjan Gijsenij. Edge- based color constancy.IEEE Transactions on image process- ing, 16(9):2207–2214, 2007. 2, 6, 7, 8

    work page 2007

  64. [64]

    Object knowledge modulates colour appearance.i-Perception, 2(1):13–49, 2011

    Christoph Witzel, Hanna Valkova, Thorsten Hansen, and Karl R Gegenfurtner. Object knowledge modulates colour appearance.i-Perception, 2(1):13–49, 2011. 2

    work page 2011

  65. [65]

    Improving color constancy in an ambient light environ- ment using the phong reflection model.IEEE Transactions on Image Processing, 27(4):1862–1877, 2018

    Sung-Min Woo, Sang-Ho Lee, Jun-Sang Yoo, and Jong-Ok Kim. Improving color constancy in an ambient light environ- ment using the phong reflection model.IEEE Transactions on Image Processing, 27(4):1862–1877, 2018. 1, 6

    work page 2018

  66. [66]

    Incorporating degradation es- timation in light field spatial super-resolution.Computer Vi- sion and Image Understanding, 252:104295, 2025

    Zeyu Xiao and Zhiwei Xiong. Incorporating degradation es- timation in light field spatial super-resolution.Computer Vi- sion and Image Understanding, 252:104295, 2025. 2

    work page 2025

  67. [67]

    Learning dual modality interactions for event- based motion deblurring.IEEE Transactions on Multimedia,

    Zeyu Xiao, Zhuoyuan Li, Yang Zhao, Yu Liu, Zhao Zhang, and Wei Jia. Learning dual modality interactions for event- based motion deblurring.IEEE Transactions on Multimedia,

    work page

  68. [68]

    End-to-end illuminant estimation based on deep met- ric learning

    Bolei Xu, Jingxin Liu, Xianxu Hou, Bozhi Liu, and Guoping Qiu. End-to-end illuminant estimation based on deep met- ric learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3616– 3625, 2020. 2, 6

    work page 2020

  69. [69]

    Cascading convolutional color con- stancy

    Huanglin Yu, Ke Chen, Kaiqi Wang, Yanlin Qian, Zhaoxi- ang Zhang, and Kui Jia. Cascading convolutional color con- stancy. InProceedings of the AAAI Conference on Artificial Intelligence, pages 12725–12732, 2020. 2

    work page 2020

  70. [70]

    Color constancy from a pure color view.JOSA A, 40(3):602–610, 2023

    Shuwei Yue and Minchen Wei. Color constancy from a pure color view.JOSA A, 40(3):602–610, 2023. 2

    work page 2023

  71. [71]

    Effective cross-sensor color constancy using a dual-mapping strategy.Journal of the Op- tical Society of America A, 41(2):329–337, 2024

    Shuwei Yue and Minchen Wei. Effective cross-sensor color constancy using a dual-mapping strategy.Journal of the Op- tical Society of America A, 41(2):329–337, 2024. 2, 3, 6

    work page 2024