Pith. sign in

REVIEW 1 major objections 5 minor 58 references

Color-name tone curves make learned image enhancement both better and editable by hand.

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 ↗

NamedCurves+ conditions Bezier tone curves on color-naming probability maps and fuses them with a transposed-attention transformer, yielding SOTA interpretable enhancement on MIT-5K, PPR10K, MSEC and SICE.

T0 review reviewed 2026-07-10 challenge →

load-bearing objection Solid systems extension of the authors’ own NamedCurves work: real SOTA gains on three tasks plus genuine interactivity, with no load-bearing flaws. the 1 major comments →

arxiv 2607.08185 v1 pith:AOFLICN7 submitted 2026-07-09 cs.CV cs.AI

Leveraging Color Naming for Image Enhancement

David Serrano-Lozano , Luis Herranz , Michael S. Brown , Javier Vazquez-Corral This is my paper
classification cs.CV cs.AI
keywords image enhancementcolor namingtone curvesimage retouchingtone mappingexposure correctionuser interactiontransformer fusion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Most deep image enhancers copy an expert's style but leave users with an opaque black box they cannot tweak. NamedCurves+ claims the remedy is to first standardize the photo, then split it into six universal color-name maps (red, green, blue, orange-brown-yellow, pink-purple, achromatic), learn a smooth Bezier tone curve for each map, and fuse the six globally adjusted versions with a lightweight transformer that restores local context. Because every edit is still a familiar color curve, the model is both stronger than prior LUT- and curve-based systems on retouching, tone-mapping and exposure-correction benchmarks and open to live user adjustment: drag a curve and only the matching color regions change, without retraining. The paper therefore argues that color naming is the missing inductive bias that reconciles automatic quality with human control.

Core claim

Conditioning learned Bezier tone curves on a fixed, perceptually grounded color-naming decomposition, then fusing the resulting globally adjusted images with a transposed-attention transformer, yields state-of-the-art enhancement while remaining fully interpretable and user-editable.

What carries the argument

NamedCurves+: a pipeline that maps an input through a standardization backbone, six color-name probability maps, per-name Bezier tone curves, and a single multi-head transposed-attention fusion block.

Load-bearing premise

The fixed color-naming model (and the authors' grouping of eleven names into six) must produce stable probability maps even under unusual cameras and lighting, or the later curves and fusion will mis-edit the image.

What would settle it

Replace the color-naming stage with random RGB partitions or pure intensity curves on the same training data; if the quantitative gains and the absence of boundary halos disappear, the color-naming claim fails.

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

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 5 minor

Summary. NamedCurves+ is a supervised image-enhancement framework that first standardizes an input with a UNet-like backbone, decomposes the result into six color-name probability maps (Van de Weijer et al. with an ad-hoc grouping of the classic 11 terms), estimates a set of Bezier-parametrized global tone curves conditioned on each map, and fuses the six globally adjusted images with a transposed-attention transformer block. The design is intended to be both competitive with SOTA and interactive: users can edit the learned control points of individual color-name curves at inference time without retraining. The method is evaluated on image retouching (MIT-Adobe-5K, PPR10K), tone mapping (MIT-Adobe-5K XYZ o sRGB), and exposure correction (MSEC, SICE), with systematic ablations of backbone, naming scheme, fusion module, control-point count and loss weight α, plus a small 2AFC user study.

Significance. If the reported gains hold, the paper supplies a practical middle ground between black-box image-to-image networks and classical parametric editors: competitive quantitative performance on four standard benchmarks together with an explicit, user-editable representation (per-color-name Bezier curves) that matches the color-naming vocabulary already used in commercial software. The extension from the conference NamedCurves paper—transformer fusion that reduces halo artifacts, multi-task evaluation, and clearer interactivity demos—is incremental but useful. Strengths that should be credited include the breadth of the experimental suite (Tables I–IV), the module and naming ablations (Tables V–VII), the controlled 2AFC study, and the explicit Limitations section that already flags boundary leakage of the fixed color-naming prior.

major comments (1)
  1. The central empirical claim is adequately supported; no load-bearing technical error was found. The only material caveat is reproducibility: source code and models are promised (Abstract) but not yet released, so independent verification of the reported numbers and of the interactive editing interface is currently impossible. This should be remedied before final acceptance.
minor comments (5)
  1. Tables I–IV report point estimates only; even a single-run standard deviation or a note that results are from one seed would strengthen confidence.
  2. Inference-time figures (Table I, Table VIII) are given for a single hardware pair; a short note on batch size and whether the color-naming step is included would help readers compare fairly with LUT-based methods.
  3. Section III-B and Figure 5 motivate the 11 o6 grouping by intensity-dependent hue overlap; a one-sentence quantitative check (e.g., average probability mass retained after grouping) would make the design choice more transparent.
  4. Figure 2 caption and the surrounding text correctly highlight halo reduction, but the visual difference is subtle; a zoomed inset or residual map would make the improvement clearer.
  5. A few typographical issues remain (e.g., “F orum” in the references, inconsistent spacing around “ΔE”). A final proof-reading pass is recommended.

Circularity Check

0 steps flagged

No significant circularity: supervised empirical pipeline against external expert ground truth; color-naming prior is off-the-shelf, not defined by the enhancement objective.

full rationale

NamedCurves+ is a standard supervised learning architecture (backbone standardization + fixed Van de Weijer color-naming maps grouped into six channels + Bezier curve estimation + transformer fusion) trained end-to-end against external expert-retouched targets on public datasets (MIT-Adobe-5K, PPR10K, MSEC, SICE). The loss (Eq. 5) is ordinary L2 + SSIM fidelity to those targets; the Bezier control points are free parameters estimated by the network, not fitted constants later re-presented as predictions. Color naming is taken from an independent prior (Van de Weijer et al. 2009) and is ablated against alternatives (Table VII); the grouping of 11 names into 6 is an engineering choice justified by intensity-range coverage, not a uniqueness theorem or self-citation chain. No equation reduces by construction to its own input, no fitted parameter is renamed a prediction, and the self-citation to the authors’ prior conference paper is merely the baseline being improved upon. The central performance claims rest on external benchmarks and ablations, not on circular derivation.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 1 invented entities

The method rests on standard supervised-learning assumptions plus a handful of architectural and hyper-parameter choices that are either taken from prior literature or selected by validation. No new physical entities are postulated; the free parameters are ordinary ML knobs.

free parameters (4)
  • loss weight α = 0.5
    Balances backbone standardization loss against final reconstruction; set to 0.5 after ablation (Table X).
  • number of Bezier control points N = 11
    Chosen by ablation among {7,11,16}; 11 yields best metrics (Table IX).
  • transformer channel dimension C = 18
    Set equal to the 18 input channels of the six RGB images; not further tuned.
  • learning-rate schedule = 1e-4 / 50-epoch half
    Adam 1e-4 decayed 50 % every 50 epochs for 200 epochs; standard but still a free choice.
axioms (4)
  • domain assumption Van de Weijer et al. (2009) 11-class color-naming probabilities are a faithful perceptual decomposition of sRGB pixels.
    Invoked in Section III-B as the sole source of the six probability maps that condition every subsequent curve.
  • ad hoc to paper Grouping orange-brown-yellow, pink-purple and white-grey-black into three composite maps preserves all intensity-dependent hue information needed for tone curves.
    Authors’ design choice justified by intensity plots (Fig. 5) but not derived from first principles.
  • domain assumption Expert-retouched images (Expert C of MIT-5K, three experts of PPR10K, etc.) constitute the correct target distribution for ‘visually appealing’ enhancement.
    Standard supervised-retouching assumption used for all quantitative tables.
  • domain assumption Monotonic Bezier curves with fixed input abscissae and normalized positive increments are sufficiently expressive for global color-name edits.
    Section III-C; common in the curve-estimation literature the paper cites.
invented entities (1)
  • NamedCurves+ end-to-end architecture (backbone + 6 color-name Bezier heads + transposed-attention fusion) no independent evidence
    purpose: To produce an interpretable, interactive, multi-task image enhancer.
    The specific modular composition is new to this paper; independent evidence is the public benchmarks and ablations, not an external physical prediction.

reviewed 2026-07-10 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Leveraging Color Naming for Image Enhancement." pith.science (2026). https://pith.science/paper/AOFLICN7

@misc{pith2026260708185,
  author       = {Pith},
  title        = {Pith review of: Leveraging Color Naming for Image Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AOFLICN7}},
  note         = {Machine review of arXiv:2607.08185}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Enhancing images to make them visually appealing is a persistent challenge in computer vision. Many deep-learning methods train models on paired datasets to replicate expert editing styles. However, these approaches struggle with two key issues: (1) interpretability and (2) a parametrization suitable for user adjustments. To address these challenges, we present NamedCurves+, an approach inspired by the concept of Color Naming, a universal set of familiar colors widely used in software tools for intuitive editing. Our method integrates color names into a learning-based framework, enabling global adjustments for each named color through tone curves. To address local image variations, we incorporate a transformer block that captures spatial dependencies, enabling context-aware edits across the image. NamedCurves+ enhances the retouching process's interpretability and supports user interaction, allowing flexible modifications of individual tone curves to refine the retouched image according to personal preferences. Extensive experiments on tasks such as image retouching, tone mapping, and exposure correction demonstrate that NamedCurves+ outperforms state-of-the-art methods. Notably, our approach is both explainable, as the tone curves explicitly represent how each color name contributes to the enhancement, and interactive, allowing users to customize the retouching process and achieve results tailored to their liking.

Figures

Figures reproduced from arXiv: 2607.08185 by David Serrano-Lozano, Javier Vazquez-Corral, Luis Herranz, Michael S. Brown.

Figure 1
Figure 1. Figure 1: Our framework (a) enhances images to achieve visually appealing results. Starting with the input image [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples where NamedCurves struggled with. It generates halo artifacts in color boundaries and brittleness when color naming probabilities change drastically. composite categories to a small number of those. The 11 color names found that most societies and cultures share are: orange, brown, yellow, white, grey, black, pink, purple, red, green and blue. Following Berlin and Kay’s research, different studies… view at source ↗
Figure 3
Figure 3. Figure 3: Van de Weijer et al. [20] color names grouped in the Munsell color [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Van de Weijer et al. [20] color naming method applied pixel-wise to the top-left image. The other 11 images show the 11 probability color names [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Grouped color names with respect to the intensity value. Pixels with [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Estimation of the Bezier tone curves. First, we extract 64-D con [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the transformer block. The globally-adjusted images, each one conditioned by a color term, are concatenated and passed through a [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results on image-retouching: (a) MIT5K, and (b) PPR10K datasets. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative results on exposure correction: (a) ME, and (b) SICE (d) datasets. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Examples demonstrating the interactivity of our method. The first column displays the input image followed by the output image given by our [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Progressive interactive editing of a single image. From left to right: input, NamedCurves+ automatic enhancement, and successive edits increasing all blue tone curve control points of the color name Red by 10% per step (when possible). Input image Standardized image NamedCurves+ Ground Truth [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Examples of NamedCurves+ and the intermediate standardized image. The standardized image compresses the dynamic range and corrects exposure before color-specific curves are applied. H. Fusion Mechanism Computational Cost Table VIII presents a detailed computational analysis com￾paring our new fusion module with the previous approach. We report: (i) average inference time over 100 images from the MIT5K dat… view at source ↗
Figure 14
Figure 14. Figure 14: Two-alternative forced choice (2AFC) on 25 images from MIT5K and PPR10K datasets. 15 observers took part. -0.3793; NamedCurves: -0.6878. Our method is statistically significantly better than the other two —95% confidence interval is 0.33 [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] 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

58 extracted references · 58 canonical work pages · 1 internal anchor

  1. [1]

    Learning image-adaptive 3d lookup tables for high performance photo enhancement in real- time,

    H. Zeng, J. Cai, L. Li, Z. Cao, and L. Zhang, “Learning image-adaptive 3d lookup tables for high performance photo enhancement in real- time,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 4, pp. 2058–2073, 2020

    work page 2058

  2. [2]

    Adaint: Learning adaptive intervals for 3d lookup tables on real-time image enhancement,

    C. Yang, M. Jin, X. Jia, Y . Xu, and Y . Chen, “Adaint: Learning adaptive intervals for 3d lookup tables on real-time image enhancement,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17 522–17 531

    work page 2022

  3. [3]

    Seplut: Sep- arable image-adaptive lookup tables for real-time image enhancement,

    C. Yang, M. Jin, Y . Xu, R. Zhang, Y . Chen, and H. Liu, “Seplut: Sep- arable image-adaptive lookup tables for real-time image enhancement,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 201– 217

    work page 2022

  4. [4]

    Image-adaptive 3d lookup tables for real-time image enhancement with bilateral grids,

    W. Kim and N. I. Cho, “Image-adaptive 3d lookup tables for real-time image enhancement with bilateral grids,” inEuropean Conference on Computer Vision. Springer, 2025, pp. 91–108

    work page 2025

  5. [5]

    Deep photo enhancer: Unpaired learning for image enhancement from photographs with gans,

    Y .-S. Chen, Y .-C. Wang, M.-H. Kao, and Y .-Y . Chuang, “Deep photo enhancer: Unpaired learning for image enhancement from photographs with gans,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018

    work page 2018

  6. [6]

    Towards unsupervised deep image enhancement with generative adversarial network,

    Z. Ni, W. Yang, S. Wang, L. Ma, and S. Kwong, “Towards unsupervised deep image enhancement with generative adversarial network,”IEEE Transactions on Image Processing, vol. 29, pp. 9140–9151, 2020

    work page 2020

  7. [7]

    Enlightengan: Deep light enhancement without paired supervision,

    Y . Jiang, X. Gong, D. Liu, Y . Cheng, C. Fang, X. Shen, J. Yang, P. Zhou, and Z. Wang, “Enlightengan: Deep light enhancement without paired supervision,”IEEE Transactions on Image Processing, vol. 30, pp. 2340–2349, 2021

    work page 2021

  8. [8]

    Adobe photoshop,

    Adobe Systems, “Adobe photoshop,” Computer software. [Online]. Available: photoshop.adobe.com

    work page

  9. [9]

    Berlin and P

    B. Berlin and P. Kay,Basic color terms: Their universality and evolution. Univ of California Press, 1991

    work page 1991

  10. [10]

    Learning photo- graphic global tonal adjustment with a database of input/output image pairs,

    V . Bychkovsky, S. Paris, E. Chan, and F. Durand, “Learning photo- graphic global tonal adjustment with a database of input/output image pairs,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2011, pp. 97–104

    work page 2011

  11. [11]

    Ppr10k: A large- scale portrait photo retouching dataset with human-region mask and group-level consistency,

    J. Liang, H. Zeng, M. Cui, X. Xie, and L. Zhang, “Ppr10k: A large- scale portrait photo retouching dataset with human-region mask and group-level consistency,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 653–661

    work page 2021

  12. [12]

    Learning multi-scale photo exposure correction,

    M. Afifi, K. G. Derpanis, B. Ommer, and M. S. Brown, “Learning multi-scale photo exposure correction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 9157–9167

    work page 2021

  13. [13]

    Learning a deep single image contrast enhancer from multi-exposure images,

    J. Cai, S. Gu, and L. Zhang, “Learning a deep single image contrast enhancer from multi-exposure images,”IEEE Transactions on Image Processing, vol. 27, no. 4, pp. 2049–2062, 2018

    work page 2049

  14. [14]

    Namedcurves: Learned image enhancement via color naming,

    D. Serrano-Lozano, L. Herranz, M. S. Brown, and J. Vazquez-Corral, “Namedcurves: Learned image enhancement via color naming,” in European Conference on Computer Vision, 2025, pp. 92–108

    work page 2025

  15. [15]

    Integrating high-level features for consistent palette-based multi-image recoloring,

    D. Xue, J. V . Corral, L. Herranz, Y . Zhang, and M. S. Brown, “Integrating high-level features for consistent palette-based multi-image recoloring,” inComputer Graphics F orum, vol. 42, no. 7. Wiley Online Library, 2023, p. e14964

    work page 2023

  16. [16]

    Palette-based color harmonization via color naming,

    D. Xue, J. Vazquez-Corral, L. Herranz, Y . Zhang, and M. S. Brown, “Palette-based color harmonization via color naming,”IEEE Signal Processing Letters, vol. 31, pp. 1474–1478, 2024

    work page 2024

  17. [17]

    Color naming reflects optimal partitions of color space,

    T. Regier, P. Kay, and N. Khetarpal, “Color naming reflects optimal partitions of color space,”Proceedings of the National Academy of Sciences, vol. 104, no. 4, pp. 1436–1441, 2007

    work page 2007

  18. [18]

    Modeling color difference for visualization design,

    D. A. Szafir, “Modeling color difference for visualization design,”IEEE Transactions on Visualization and Computer Graphics, vol. 24, no. 1, pp. 392–401, 2017

    work page 2017

  19. [19]

    Coloring with words: Guiding image colorization through text-based palette generation,

    H. Bahng, S. Yoo, W. Cho, D. K. Park, Z. Wu, X. Ma, and J. Choo, “Coloring with words: Guiding image colorization through text-based palette generation,” inEuropean Conference on Computer Vision, 2018, pp. 431–447

    work page 2018

  20. [20]

    Learning color names for real-world applications,

    J. Van De Weijer, C. Schmid, J. Verbeek, and D. Larlus, “Learning color names for real-world applications,”IEEE Transactions on Image Processing, vol. 18, no. 7, pp. 1512–1523, 2009

    work page 2009

  21. [21]

    Parametric fuzzy sets for automatic color naming,

    R. Benavente, M. Vanrell, and R. Baldrich, “Parametric fuzzy sets for automatic color naming,”Journal of the Optical Society of America A, vol. 25, no. 10, pp. 2582–2593, 2008

    work page 2008

  22. [22]

    Weakly supervised domain- specific color naming based on attention,

    L. Yu, Y . Cheng, and J. van de Weijer, “Weakly supervised domain- specific color naming based on attention,” inInternational Conference on Pattern Recognition. IEEE, 2018. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    work page 2018

  23. [23]

    Nice: A computational solution to close the gap from colour perception to colour categorization,

    C. A. Parraga and A. Akbarinia, “Nice: A computational solution to close the gap from colour perception to colour categorization,”PloS one, vol. 11, no. 3, p. e0149538, 2016

    work page 2016

  24. [24]

    Adaptive image contrast enhancement using generalizations of histogram equalization,

    J. A. Stark, “Adaptive image contrast enhancement using generalizations of histogram equalization,”IEEE Transactions on Image Processing, vol. 9, no. 5, pp. 889–896, 2000

    work page 2000

  25. [25]

    Realization of the contrast limited adaptive histogram equalization (clahe) for real-time image enhancement,

    A. M. Reza, “Realization of the contrast limited adaptive histogram equalization (clahe) for real-time image enhancement,”Journal of VLSI Signal Processing-Systems for Signal, Image, and Video Technology, vol. 38, pp. 35–44, 2004

    work page 2004

  26. [26]

    Fast local laplacian filters: Theory and applications,

    M. Aubry, S. Paris, S. W. Hasinoff, J. Kautz, and F. Durand, “Fast local laplacian filters: Theory and applications,”ACM Transactions on Graphics, vol. 33, no. 5, pp. 1–14, 2014

    work page 2014

  27. [27]

    Fast bilateral filtering for the display of high-dynamic-range images,

    F. Durand and J. Dorsey, “Fast bilateral filtering for the display of high-dynamic-range images,” inConference on Computer Graphics and Interactive Techniques, 2002

    work page 2002

  28. [28]

    Edge-based color constancy,

    J. Van De Weijer, T. Gevers, and A. Gijsenij, “Edge-based color constancy,”IEEE Transactions on Image Processing, vol. 16, no. 9, pp. 2207–2214, 2007

    work page 2007

  29. [29]

    Real- time image enhancer via learnable spatial-aware 3d lookup tables,

    T. Wang, Y . Li, J. Peng, Y . Ma, X. Wang, F. Song, and Y . Yan, “Real- time image enhancer via learnable spatial-aware 3d lookup tables,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021

    work page 2021

  30. [30]

    Deeplpf: Deep local parametric filters for image enhancement,

    S. Moran, P. Marza, S. McDonagh, S. Parisot, and G. Slabaugh, “Deeplpf: Deep local parametric filters for image enhancement,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 12 826–12 835

    work page 2020

  31. [31]

    Rsfnet: A white-box image retouching approach using region-specific color filters,

    W. Ouyang, Y . Dong, X. Kang, P. Ren, X. Xu, and X. Xie, “Rsfnet: A white-box image retouching approach using region-specific color filters,” inICCV, 2023

    work page 2023

  32. [32]

    Zero- reference deep curve estimation for low-light image enhancement,

    C. Guo, C. Li, J. Guo, C. C. Loy, J. Hou, S. Kwong, and R. Cong, “Zero- reference deep curve estimation for low-light image enhancement,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020, pp. 1780–1789

    work page 2020

  33. [33]

    Curl: Neural curve layers for global image enhancement,

    S. Moran, S. McDonagh, and G. Slabaugh, “Curl: Neural curve layers for global image enhancement,” inInternational Conference on Pattern Recognition, 2021, pp. 9796–9803

    work page 2021

  34. [34]

    Flexicurve: Flex- ible piecewise curves estimation for photo retouching,

    C. Li, C. Guo, S. Zhou, Q. Ai, R. Feng, and C. C. Loy, “Flexicurve: Flex- ible piecewise curves estimation for photo retouching,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1092–1101

    work page 2023

  35. [35]

    Personalized image enhancement using neural spline color transforms,

    S. Bianco, C. Cusano, F. Piccoli, and R. Schettini, “Personalized image enhancement using neural spline color transforms,”IEEE Transactions on Image Processing, vol. 29, pp. 6223–6236, 2020

    work page 2020

  36. [36]

    Learning tone curves for local image enhancement,

    L. Zhao, A. Abdelhamed, and M. S. Brown, “Learning tone curves for local image enhancement,”IEEE Access, vol. 10, pp. 60 099–60 113, 2022

    work page 2022

  37. [37]

    Learning exposure correction via consistency modeling,

    N. E. Nsampi, Z. Hu, and Q. Wang, “Learning exposure correction via consistency modeling,” inBritish Machine Vision Conference, 2021, p. 12

    work page 2021

  38. [38]

    Deep fourier-based exposure correction network with spatial- frequency interaction,

    J. Huang, Y . Liu, F. Zhao, K. Yan, J. Zhang, Y . Huang, M. Zhou, and Z. Xiong, “Deep fourier-based exposure correction network with spatial- frequency interaction,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 163–180

    work page 2022

  39. [39]

    Per- ceptual image enhancement for smartphone real-time applications,

    M. V . Conde, F. Vasluianu, J. Vazquez-Corral, and R. Timofte, “Per- ceptual image enhancement for smartphone real-time applications,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 1848–1858

    work page 2023

  40. [40]

    MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, “Mobilenets: Efficient convo- lutional neural networks for mobile vision applications,”arXiv preprint arXiv:1704.04861, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  41. [41]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” inEuropean Conference on Computer Vision, 2018, pp. 3–19

    work page 2018

  42. [42]

    Ex- pandnet: A deep convolutional neural network for high dynamic range expansion from low dynamic range content,

    D. Marnerides, T. Bashford-Rogers, J. Hatchett, and K. Debattista, “Ex- pandnet: A deep convolutional neural network for high dynamic range expansion from low dynamic range content,” inComputer Graphics F orum, vol. 37, no. 2. Wiley Online Library, 2018, pp. 37–49

    work page 2018

  43. [43]

    Deep bilateral learning for real-time image enhancement,

    M. Gharbi, J. Chen, J. T. Barron, S. W. Hasinoff, and F. Durand, “Deep bilateral learning for real-time image enhancement,”ACM Transactions on Graphics, vol. 36, no. 4, pp. 1–12, 2017

    work page 2017

  44. [44]

    Influences of psychological factors on image color preferences evaluation,

    G. Shing-Sheng and H. Po-Sung, “Influences of psychological factors on image color preferences evaluation,”Color Research & Application, vol. 35, no. 3, pp. 213–232, 2010

    work page 2010

  45. [45]

    The quantitative aspects of color rendering for memory colors,

    K. Topfer and R. Cookingham, “The quantitative aspects of color rendering for memory colors,” inIS&T’s PICS, 2000

    work page 2000

  46. [46]

    Simplifying tone curves for image enhancement,

    J. Bennett and G. D. Finlayson, “Simplifying tone curves for image enhancement,”Color and Imaging Conference, vol. 31, no. 1, pp. 108– 108, 2023

    work page 2023

  47. [47]

    Restormer: Efficient transformer for high-resolution image restoration,

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang, “Restormer: Efficient transformer for high-resolution image restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 5728–5739

    work page 2022

  48. [48]

    Retinex- former: One-stage retinex-based transformer for low-light image en- hancement,

    Y . Cai, H. Bian, J. Lin, H. Wang, R. Timofte, and Y . Zhang, “Retinex- former: One-stage retinex-based transformer for low-light image en- hancement,” inICCV, 2023

    work page 2023

  49. [49]

    A comparative study of image restoration networks for general backbone network design,

    X. Chen, Z. Li, Y . Pu, Y . Liu, J. Zhou, Y . Qiao, and C. Dong, “A comparative study of image restoration networks for general backbone network design,” inECCV, 2024

    work page 2024

  50. [50]

    Revisiting image fusion for multi-illuminant white-balance correction,

    D. Serrano-Lozano, A. Arora, L. Herranz, K. G. Derpanis, M. S. Brown, and J. Vazquez-Corral, “Revisiting image fusion for multi-illuminant white-balance correction,” inICCV, 2025

    work page 2025

  51. [51]

    Self-attention generative adversarial networks,

    H. Zhang, I. Goodfellow, D. Metaxas, and A. Odena, “Self-attention generative adversarial networks,” inInternational Conference on Ma- chine Learning. PMLR, 2019, pp. 7354–7363

    work page 2019

  52. [52]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” inInternational Conference on Learning Representations, 2015

    work page 2015

  53. [53]

    Underexposed photo enhancement using deep illumination estimation,

    R. Wang, Q. Zhang, C.-W. Fu, X. Shen, W.-S. Zheng, and J. Jia, “Underexposed photo enhancement using deep illumination estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019

    work page 2019

  54. [54]

    Hdrnet: Single-image-based hdr reconstruction using channel attention cnn,

    J. Li and P. Fang, “Hdrnet: Single-image-based hdr reconstruction using channel attention cnn,” inInternational Conference on Multimedia Systems and Signal Processing, 2019, pp. 119–124

    work page 2019

  55. [55]

    Adaptive histogram equalization and its variations,

    S. M. Pizer, E. P. Amburn, J. D. Austin, R. Cromartie, A. Geselowitz, T. Greer, B. ter Haar Romeny, J. B. Zimmerman, and K. Zuiderveld, “Adaptive histogram equalization and its variations,”Computer Vision, Graphics, and Image Processing, vol. 39, no. 3, pp. 355–368, 1987

    work page 1987

  56. [56]

    Deep retinex decomposition for low-light enhancement,

    C. Wei, W. Wang, W. Yang, and J. Liu, “Deep retinex decomposition for low-light enhancement,” inBritish Machine Vision Conference, 2018, p. 155

    work page 2018

  57. [57]

    Exposure normalization and compensation for multiple-exposure cor- rection,

    J. Huang, Y . Liu, X. Fu, M. Zhou, Y . Wang, F. Zhao, and Z. Xiong, “Exposure normalization and compensation for multiple-exposure cor- rection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 6043–6052. VII. ACKNOWLEDGEMENTS This work was supported by Grant PID2021- 128178OB-I00, , PID2024-162555OB-I00 fund...

    work page 2022

  58. [58]

    His research interests include deep learning applied to various topics in computer vision and multimedia

    He has worked at the Computer Vision Center (Barcelona), the Institute of Computing Technology (Chinese Academy of Sciences), and Mitsubishi Electric Research and Development. His research interests include deep learning applied to various topics in computer vision and multimedia. Michael S. Brownis a Professor and a Canada Research Chair of Computer Visi...

    work page 2011

This paper was first reviewed by grok-4.5 on July 10, 2026.