Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

HVI-CIDNet+: Beyond Extreme Darkness for Low-Light Image Enhancement

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The HVI color space removes the red and black noise artifacts that HSV introduces, and the HVI-CIDNet+ network built on it outperforms prior low-light enhancement methods on ten benchmark datasets.

desk verdict Plausible color-space idea with a decent LOL ablation, but Eq. (4) contradicts its own collapse mechanism and the SOTA claims overreach the evidence. read the letter →

arxiv 2507.06814 v1 pith:FOECTRDI submitted 2025-07-09 cs.CV

Qingsen Yan , Kangbiao Shi , Yixu Feng , Tao Hu , Peng Wu , Guansong Pang , Yanning Zhang This is my paper
classification cs.CV
keywords low-lightimageenhancementHVIcolorspacenoisevision-languagepriorscross-attentionregionrefinementextremedarknessbenchmarkevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that the color spaces commonly used for low-light image enhancement are themselves a source of artifacts, and that a purpose-built color space can remove those artifacts and push enhancement quality higher. The proposed Horizontal/Vertical-Intensity (HVI) space polarizes the hue axis so that the two ends of red become close, and applies a learnable intensity collapse function that squeezes near-black pixels together, eliminating the red discontinuity and black-plane noise that appear when brightness and color are decoupled in HSV space. On top of the HVI space, the paper builds HVI-CIDNet+, a dual-branch network that uses latent semantic priors and degraded representations from a pre-trained vision-language model, plus a region refinement block, to restore content and correct color in extremely dark regions. The paper reports that HVI-CIDNet+ outperforms existing methods on ten low-light benchmarks and that the HVI transform, used as a plug-in, improves several sRGB-based methods. A sympathetic reader would care because the core claim is a reusable color representation, not just a bigger network.

What carries the argument

The load-bearing object is the HVI color space transform plus its inverse. Forward: $\hat H = C_k \odot S \odot \cos(\pi H/3)$, $\hat V = C_k \odot S \odot \sin(\pi H/3)$, with intensity $I_{\max}=\max_c I_c$ and $C_k(x)=k/\sqrt{\sin(\pi I_{\max}(x)/2)+\varepsilon}$. The cosine/sine polarization removes the hue-axis wrap that puts the same red at $h=0$ and $h=6$; the trainable scalar $k$ controls how strongly low-intensity chroma is collapsed toward zero, absorbing black noise. The inverse PHVIT recovers HSV by $\hat h = \hat H/(C_k+\varepsilon)$, $\hat v = \hat V/(C_k+\varepsilon)$, then converts back to sRGB, with user parameters $\alpha_S,\alpha_I$ for saturation and brightness. In the network, the Prior-guided Attention Block (PAB) uses cross-attention so the I-branch receives latent semantic priors $\mathbf{e}_s$ and the HV-branch receives degraded representations $\mathbf{e}_d$, and the Region Refinement Block (RRB) splits feature maps into information-scarce and information-rich regions, applying window attention to the former and multi-scale contextual convolution to the latter.

What would settle it

Take a method that currently benefits from the HVI plug-in and run it on a test set of very dark images whose maximum intensity is below 0.05 in most pixels and whose ground truth has saturated red objects. If $C_k$ has been trained on a dataset with milder darkness, the inverse PHVIT will divide predicted chroma by a value close to $\varepsilon$ in those regions; observing large false red or black blobs in the output, or a PSNR drop below the original sRGB method, would falsify the claim that HVI removes red and black artifacts in extreme darkness.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the two characteristic artifacts of HSV-based low-light enhancement—red discontinuity noise, where the same red hue sits at both ends of the hue axis, and black-plane noise, where near-black pixels spread into a noisy plane—can be removed by a coordinate change. The HVI transform maps hue $h$ to $\hat{h}=\cos(\pi h/3)$ and $\hat{v}=\sin(\pi h/3)$, multiplies both by saturation $S$ and by a learnable intensity collapse $C_k(x)=k/\sqrt{\sin(\pi I_{\max}(x)/2)+\varepsilon}$, and keeps $I_{\max}$ as the intensity channel. The result is a color space in which similar colors have small Euclidean distances: red is continuous across the polar angle, and dark pixels are clustered at the origin. The companion network, HVI-CIDNet+, then learns enhancement in this space with an HV-branch for color and an I-branch for brightness, fused through cross-attention guided by semantic and degradation priors and refined per region. The paper claims this combination, measured by PSNR, SSIM, and LPIPS, beats state-of-the-art methods on ten datasets, and that the HVI transform itself transfers as a plug-in to other sRGB-based LLIE models.

Load-bearing premise

The load-bearing premise is that one learned number $k$ per dataset can collapse all near-black chroma toward zero without erasing the color information the network needs, and that dividing predicted chroma by $C_k+\varepsilon$ during the inverse transform remains numerically safe in exactly the darkest regions.

Editorial extensions

If this is right

  • HVI-CIDNet+ achieves the best PSNR/SSIM/LPIPS on LOLv1 and LOLv2 among compared methods, so the combination of the color space and the architecture is claimed to be the current state of the art on those paired benchmarks.
  • On the extreme-darkness SICE and Sony-Total-Dark sets, the reported gains over LLFlow are 1.713 dB and 7.256 dB PSNR, respectively.
  • The HVI transform is claimed to be portable: wrapping sRGB-based methods in HVI (forward transform and PHVIT) improves their PSNR, e.g., GSAD gains 3.562 dB on LOLv2-Real, without retraining those methods' color assumptions.
  • The ablation results identify the separate contributions: polarization alone or $C_k$ alone each improve HSV, and combining them yields the full HVI gain; adding degraded representations, semantic priors, and the region refinement block each raises the baseline.
  • Supervising in both sRGB and HVI spaces yields better results than either loss alone, indicating that the HVI domain carries distributional information the sRGB loss misses.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A single trainable scalar $k$ per dataset is a strong simplification; one testable extension is to make $k$ spatially adaptive (e.g., predicted per region or conditioned on the estimated noise level), which could extend HVI to scenes with mixed extreme-dark and well-lit regions.
  • The inverse PHVIT divides by $C_k+\varepsilon$, so numerical stability near $I_{\max}\approx 0$ is the hidden failure mode; an explicit stability check or a regularized inverse would reveal whether the reported gains persist at even lower sensor exposures than the benchmarks.
  • Because the HVI space is defined independently of the network, it may transfer to other low-level tasks that suffer from red/black artifacts, such as dehazing, white-balancing, or underexposed video enhancement; this is an inference, not a claim the paper makes.
  • The plug-in results suggest that part of the improvement comes from the color space itself rather than the architecture; an editorial reading is that HVI could be combined with lightweight or zero-reference methods to make color-space gains available in low-compute settings, which the paper does not test.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a new color space, Horizontal/Vertical-Intensity (HVI), derived from HSV by polarizing the Hue–Saturation plane and applying a learnable intensity collapse function C_k, together with a dual-branch enhancement network HVI-CIDNet+ that incorporates pre-trained vision-language priors. The authors claim that HVI removes red and black noise artifacts of HSV and that HVI-CIDNet+ outperforms state-of-the-art methods on 10 datasets, with ablations on LOLv1 and plug-in experiments on other LLIE models.

Significance. If the HVI color space and the reported results are correct, the paper would offer a practically useful color-space transformation for LLIE that can be grafted onto existing sRGB-based methods, which is a distinctive and potentially valuable contribution. Additional strengths are the availability of code, the breadth of evaluation across 10 datasets, and the explicit ablation isolating the color-space components from the network components. However, the central formula defining C_k appears to be internally inconsistent with the claimed collapse behavior, and several benchmark comparisons may be unfair because baseline training protocols are not stated; these issues must be resolved before the contribution can be assessed.

major comments (4)
  1. [§III-B, Eq. (4) and §IV-D] The intensity collapse function is printed as C_k(x) = k / (sin(π I_max(x)/2) + ε). This function is strictly decreasing in I_max on [0,1]: for a dark pixel with I_max ≈ 0, C_k ≈ k/ε ≈ 10^8, the largest possible value, while for a bright pixel with I_max = 1, C_k ≈ k. The text states that 'smaller C_k correspond[s] to smaller radius or lower intensity values' and that 'black points are clustered as C_k decreases', which is the opposite of what Eq. (4) computes. Consequently, the forward HVI transform in Eq. (5) expands the chroma radius of dark pixels rather than collapsing it, and the inverse PHVIT in Eq. (10) divides network outputs by C_k+ε, creating large numerical amplification in the darkest regions. If the intended formula is monotone increasing (e.g., C_k = k sin(π I_max/2)), then Eq. (4) is a typesetting error and the color-space ablations in Table V and the plug-in results in Table III must be re-run with the actually implemented transform; if the code implements Eq. (4) as printed, the HVI color space does not provide the advertised black-noise compression and the central claim of the paper is unsupported.
  2. [§V-B, Table IV] The SICE and Sony-Total-Dark results are used to support the claim that HVI-CIDNet+ outperforms prior methods under extreme darkness, but the manuscript does not specify whether the baseline methods (RetinexNet, ZeroDCE, URetinexNet, RUAS, LLFlow, CIDNet) were retrained on the SICE and Sony-Total-Dark training splits or evaluated with their original pretrained weights. Since HVI-CIDNet+ is trained on those datasets (as described in the Experiment Settings), comparing it against baselines that are not retrained would conflate the effect of training data with the effect of the proposed method. The PSNR gain of 7.256 dB over LLFlow on Sony-Total-Dark and the 1.713 dB gain on SICE are only meaningful under identical training protocols; please state the training setup for every baseline and, if they were not retrained, either retrain them or clearly qualify the comparison as a generalization test rather than a SOTA comparison.
  3. [§V-B, Table III] The text claims that applying the HVI transform as a plug-in to six SOTA methods gives 'consistent improvements in the PSNR, SSIM, and LPIPS metrics', but Table III shows several regressions: SNR-Aware loses 0.009 in SSIM and worsens by 0.011 in LPIPS, and LEDNet worsens by 0.005 in LPIPS. Other methods, such as FourLLIE, also show mixed changes. This contradicts the stated consistency and weakens the generalizability claim. Please correct the assertion, or provide an explanation (e.g., significance testing or per-image analysis) for why the plug-in degrades some metrics on some methods.
  4. [§III-B, Eq. (4) and §IV-B] The parameter k is described as a trainable 'darkness density' that adaptively controls the intensity collapse, and the paper motivates the entire HVI color space on this mechanism. However, the manuscript does not report the learned value(s) of k, how it varies across datasets, or its sensitivity to initialization. Since every downstream result depends on the behavior of C_k, the paper should include an analysis of the optimized k values and a stability study (e.g., perturbing k and measuring PSNR/SSIM on the validation set). Without this, the claim that k is 'learnable' and that the collapse is 'adaptive' is not empirically supported.
minor comments (5)
  1. [§III-B, Eq. (4)] The text says ε = 1×10⁻⁸ is used to 'avoid gradient explosion', but in Eq. (4) the role of ε is to prevent division by zero when I_max = 0; in Eq. (10) it also prevents division by zero in PHVIT. Please rephrase to accurately describe the purpose of ε.
  2. [§V-B, LOL results paragraph] The sentence 'it effectively balances enhancement across both information-rich regions and information-rich regions' appears to contain a typo; the second occurrence should likely be 'information-scarce regions'.
  3. [References, [12]] Reference [12] is cited as an arXiv preprint (arXiv:2502.20272) while the text states the work was published at CVPR 2025; please update the reference with the CVPR details if applicable.
  4. [Table I and Table III] GSAD is labeled with color model 'RGB' in Table I and as 'Diffusion' in Table III; the category labels should be harmonized to avoid confusion about the type of baseline.
  5. [Abstract and Introduction, code link] The provided code URL 'https://github.com/shikangbiao/CIDNet extension' contains a space and does not appear to be a valid repository address; please verify and provide the correct link.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation found; HVI transform and network are empirically evaluated on held-out benchmarks, with only disclosed, non-load-bearing self-citations.

full rationale

The paper's central claims are empirical: HVI-CIDNet+ is trained on standard training splits and evaluated on test splits of ten datasets; the trainable density parameter k in Eq. (4) is a fitted parameter of the proposed transform, not a quantity whose value is later reported as a prediction. The HVI transformation is fully specified by Eqs. (1)-(5) and the inverse PHVIT by Eq. (10), so the color-space properties (red continuity via cos/sin polarization, dark-region collapse via C_k) are design choices, not results derived from the fitted k. The plug-in experiments in Table III retrain each baseline with the HVI transform and compare held-out test metrics, so no improvement reduces to a fitted value by construction. The self-citations ([12] CIDNet, [84] DiffLight) are disclosed prior works used as baselines or architectural starting points; the equations and ablations in this manuscript are self-contained, and no load-bearing argument rests on an unverified self-citation or uniqueness theorem. The apparent sign/behavior inconsistency in Eq. (4) (C_k as printed grows for dark pixels, opposite to the stated collapse) is a correctness/consistency issue, not a circularity: no claimed output is equivalent to a fitted input by construction. Under the hard rules requiring a specific reduction to flag circularity, no circular step is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central method rests on one trainable scalar k that shapes the color space, plus standard network hyperparameters. No new physical or conceptual entities beyond the transform are introduced.

free parameters (3)
  • k (darkness density) = trained per dataset, value not reported
    Trainable parameter in Eq. 4 that scales the chroma of low-intensity pixels. Learned end-to-end on each benchmark, so the color space itself is partly fit to the training data.
  • lambda (loss weight) = not reported
    Hyperparameter in Eq. 11 balancing HVI-space and sRGB losses. Chosen by hand, no sweep reported.
  • alpha_S, alpha_I = not reported
    Saturation and brightness scaling parameters in the inverse PHVIT, Eq. 10. Described as 'customizing linear parameters'; unclear whether fixed or tuned.
assumptions (4)
  • domain assumption Max-RGB intensity estimation (Eq. 1)
    Assumes scene illuminant is white and intensity is the per-pixel channel maximum. Standard Retinex/color-constancy assumption, not verified on the test datasets.
  • domain assumption Noise primarily shifts hue and saturation with minimal effect on intensity (Sec. III-A)
    Basis for decoupling brightness and color. Cited from prior work [19,26,43] but not validated for extremely dark regions in this paper.
  • domain assumption DA-CLIP image and distortion encoders provide useful latent semantic and degraded representations for LLIE
    The PAB injects es and ed as guidance. No analysis of failure modes or dependence on the pre-training distribution is provided.
  • standard math C_k remains positive so the HVI transform is invertible
    epsilon=1e-8 avoids division by zero in PHVIT. The inverse exists by construction, but stability at tiny C_k is unexamined.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HVI-CIDNet+: Beyond Extreme Darkness for Low-Light Image Enhancement." pith.science (2026). https://pith.science/paper/FOECTRDI

@misc{pith2026250706814,
  author       = {Pith},
  title        = {Pith review of: HVI-CIDNet+: Beyond Extreme Darkness for Low-Light Image Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FOECTRDI}},
  note         = {Machine review of arXiv:2507.06814}
}
read the original abstract

Low-Light Image Enhancement (LLIE) aims to restore vivid content and details from corrupted low-light images. However, existing standard RGB (sRGB) color space-based LLIE methods often produce color bias and brightness artifacts due to the inherent high color sensitivity. While Hue, Saturation, and Value (HSV) color space can decouple brightness and color, it introduces significant red and black noise artifacts. To address this problem, we propose a new color space for LLIE, namely Horizontal/Vertical-Intensity (HVI), defined by the HV color map and learnable intensity. The HV color map enforces small distances for the red coordinates to remove red noise artifacts, while the learnable intensity compresses the low-light regions to remove black noise artifacts. Additionally, we introduce the Color and Intensity Decoupling Network+ (HVI-CIDNet+), built upon the HVI color space, to restore damaged content and mitigate color distortion in extremely dark regions. Specifically, HVI-CIDNet+ leverages abundant contextual and degraded knowledge extracted from low-light images using pre-trained vision-language models, integrated via a novel Prior-guided Attention Block (PAB). Within the PAB, latent semantic priors can promote content restoration, while degraded representations guide precise color correction, both particularly in extremely dark regions through the meticulously designed cross-attention fusion mechanism. Furthermore, we construct a Region Refinement Block that employs convolution for information-rich regions and self-attention for information-scarce regions, ensuring accurate brightness adjustments. Comprehensive results from benchmark experiments demonstrate that the proposed HVI-CIDNet+ outperforms the state-of-the-art methods on 10 datasets.

Figures

Figures reproduced from arXiv: 2507.06814 by the authors.

Figure 1
Figure 1. The top row illustrates the process of transforming images from the sRGB color space, via HSV, to the HVI color space. The bottom row presents [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overall of the proposed HVI-CIDNet+. In Stage 1 (left), the input low-light sRGB image is converted to the HVI color space, yielding HV color [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The architecture of Prior-guided Attention Block (PAB). The PAB [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The architecture of Region Refinement Block (RRB). The I-branch [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visual comparison of the enhanced images yielded by different SOTA methods on LOLv1 (top two rows) and LOLv2 (bottom two rows). Below each [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison on the five unpaired datasets. Follow RetinexFormer [15], we select one image in each dataset to compare our method with the [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Abaltion results on LOLv1 dataset for five different color spaces used by HVI-CIDNet+. The area outlined by the [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Ablation results of HVI-CIDNet+ by using different structures on LOLv1. Below each image patch is labeled the PSNR [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Learning Dexterous Grasping from Sparse Taxonomy Guidance

    cs.RO 2026-04 unverdicted novelty 6.0 of 10

    GRIT learns dexterous grasping from sparse taxonomy guidance, achieving 87.9% success and better generalization to novel objects via a two-stage prediction-plus-policy approach.

Reference graph

Works this paper leans on

83 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    Low-light image and video enhancement using deep learning: A sur- vey,

    C. Li, C. Guo, L. Han, J. Jiang, M.-M. Cheng, J. Gu, and C. C. Loy, “Low-light image and video enhancement using deep learning: A sur- vey,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 44, no. 12, pp. 9396–9416, 2022

    2022

  2. [2]

    Snr-aware low-light image enhancement,

    X. Xu, R. Wang, C.-W. Fu, and J. Jia, “Snr-aware low-light image enhancement,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 17 693–17 703

    2022

  3. [3]

    Low-light image enhancement via structure modeling and guidance,

    X. Xu, R. Wang, and J. Lu, “Low-light image enhancement via structure modeling and guidance,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 9893–9903

    2023

  4. [4]

    Kindling the darkness: A practical low-light image enhancer,

    Y . Zhang, J. Zhang, and X. Guo, “Kindling the darkness: A practical low-light image enhancer,” in Proceedings of the 27th ACM International Conference on Multimedia , ser. MM ’19. New York, NY , USA: ACM, 2019, pp. 1632–1640. [Online]. Available: http://doi.acm.org/10.1145/3343031.3350926

    arXiv 2019

  5. [5]

    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

  6. [6]

    Zero-reference deep curve estimation for low-light image enhance- ment,

    C. G. Guo, C. Li, J. Guo, C. C. Loy, J. Hou, S. Kwong, and R. Cong, “Zero-reference deep curve estimation for low-light image enhance- ment,” in Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR) , June 2020, pp. 1780–1789

    work page 2020

  7. [7]

    Glare: Low light image enhancement via generative latent feature based codebook retrieval,

    H. Zhou, W. Dong, X. Liu, S. Liu, X. Min, G. Zhai, and J. Chen, “Glare: Low light image enhancement via generative latent feature based codebook retrieval,” in Proceedings of the European Conference on Computer Vision (ECCV) , 2024

    work page 2024

  8. [8]

    Mpc-net: Multi- prior collaborative network for low-light image enhancement,

    C. She, F. Han, L. Wang, S. Duan, and T. Huang, “Mpc-net: Multi- prior collaborative network for low-light image enhancement,” IEEE Transactions on Circuits and Systems for Video Technology , pp. 1–1, 2024. 13

    work page 2024

Show all 83 references
  1. [9]

    Beyond brightening low-light images,

    Y . Zhang, X. Guo, J. Ma, W. Liu, and J. Zhang, “Beyond brightening low-light images,” International Journal of Computer Vision , vol. 129, pp. 1013–1037, 2021

    2021

  2. [10]

    Learning to enhance low-light image via zero-reference deep curve estimation,

    C. Li, C. Guo, and C. C. Loy, “Learning to enhance low-light image via zero-reference deep curve estimation,” IEEE transactions on pattern analysis and machine intelligence , vol. 44, no. 8, pp. 4225–4238, 2021

    2021

  3. [11]

    Inter- pretable optimization-inspired unfolding network for low-light image enhancement,

    W. Wu, J. Weng, P. Zhang, X. Wang, W. Yang, and J. Jiang, “Inter- pretable optimization-inspired unfolding network for low-light image enhancement,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  4. [12]

    Hvi: A new color space for low-light image enhancement,

    Q. Yan, Y . Feng, C. Zhang, G. Pang, K. Shi, P. Wu, W. Dong, J. Sun, and Y . Zhang, “Hvi: A new color space for low-light image enhancement,” arXiv preprint arXiv:2502.20272 , 2025

    2025 arXiv

  5. [13]

    Gevers, A

    T. Gevers, A. Gijsenij, J. Van de Weijer, and J.-M. Geusebroek, Color in computer vision: fundamentals and applications . John Wiley & Sons, 2012

    2012

  6. [14]

    Rethinking RGB color representation for image restoration models,

    J. Lee, J. Park, S. Baik, and K. M. Lee, “Rethinking RGB color representation for image restoration models,” 2024. [Online]. Available: https://openreview.net/forum?id=zhZXk5Ctz2

    2024

  7. [15]

    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,” in Proceedings of ICCV, October 2023, pp. 12 504–12 513

    2023

  8. [16]

    Low illumination video image enhancement,

    Z. Li, Z. Jia, J. Yang, and N. Kasabov, “Low illumination video image enhancement,” IEEE photonics journal , vol. 12, no. 4, pp. 1–13, 2020

    2020

  9. [17]

    A low-light image enhancement method based on hsv space,

    L. Zhou, X. Chen, B. Ye, X. Jiang, S. Zou, L. Ji, Z. Yu, J. Wei, Y . Zhao, and T. Wang, “A low-light image enhancement method based on hsv space,” The Imaging Science Journal , pp. 1–14, 2023

    2023

  10. [18]

    Better than reference in low-light image enhancement: Conditional re-enhancement network,

    Y . Zhang, X. Di, B. Zhang, R. Ji, and C. Wang, “Better than reference in low-light image enhancement: Conditional re-enhancement network,” IEEE Transactions on Image Processing , vol. 31, pp. 759–772, 2021

    2021

  11. [19]

    Zero-reference low-light enhance- ment via physical quadruple priors,

    W. Wang, H. Yang, J. Fu, and J. Liu, “Zero-reference low-light enhance- ment via physical quadruple priors,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 26 057–26 066

    2024

  12. [20]

    Region- based unsupervised low-light image enhancement in the wild with explicit domain supervision,

    Y . Ma, S. Xie, W. Xu, X. Chen, X. Huang, Y . Sun, and W. Liu, “Region- based unsupervised low-light image enhancement in the wild with explicit domain supervision,” IEEE Transactions on Instrumentation and Measurement, 2024

    2024

  13. [21]

    Learning semantic-aware knowledge guidance for low-light image en- hancement,

    Y . Wu, C. Pan, G. Wang, Y . Yang, J. Wei, C. Li, and H. T. Shen, “Learning semantic-aware knowledge guidance for low-light image en- hancement,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 1662–1671

    2023

  14. [23]

    J. D. Foley and A. Van Dam, Fundamentals of interactive computer graphics. Addison-Wesley Longman Publishing Co., Inc., 1982

    1982

  15. [24]

    Low-light image enhancement via breaking down the darkness,

    X. Guo and Q. Hu, “Low-light image enhancement via breaking down the darkness,” International Journal of Computer Vision , vol. 131, no. 1, pp. 48–66, Jan 2023. [Online]. Available: https: //doi.org/10.1007/s11263-022-01667-9

    2023 doi

  16. [25]

    Lyt-net: Lightweight yuv transformer-based network for low-light image enhancement,

    A. Brateanu, R. Balmez, A. Avram, and C. Orhei, “Lyt-net: Lightweight yuv transformer-based network for low-light image enhancement,” 2024

    2024

  17. [26]

    Deep retinex decomposition for low-light enhancement,

    C. Wei, W. Wang, W. Yang, and J. Liu, “Deep retinex decomposition for low-light enhancement,” arXiv preprint arXiv:1808.04560 , 2018

    2018 arXiv

  18. [27]

    Uretinex- net: Retinex-based deep unfolding network for low-light image enhance- ment,

    W. Wu, J. Weng, P. Zhang, X. Wang, W. Yang, and J. Jiang, “Uretinex- net: Retinex-based deep unfolding network for low-light image enhance- ment,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , June 2022, pp. 5901–5910

    2022

  19. [28]

    Unsu- pervised low-light image enhancement with self-paced learning,

    Y . Luo, X. Chen, J. Ling, C. Huang, W. Zhou, and G. Yue, “Unsu- pervised low-light image enhancement with self-paced learning,” IEEE Transactions on Multimedia , 2024

    2024

  20. [29]

    Naturalness preserved enhancement algorithm for non-uniform illumination images,

    S. Wang, J. Zheng, H.-M. Hu, and B. Li, “Naturalness preserved enhancement algorithm for non-uniform illumination images,” IEEE transactions on image processing , vol. 22, no. 9, pp. 3538–3548, 2013

    2013

  21. [30]

    A prob- abilistic method for image enhancement with simultaneous illumination and reflectance estimation,

    X. Fu, Y . Liao, D. Zeng, Y . Huang, X.-P. Zhang, and X. Ding, “A prob- abilistic method for image enhancement with simultaneous illumination and reflectance estimation,” IEEE Transactions on Image Processing , vol. 24, no. 12, pp. 4965–4977, 2015

    2015

  22. [31]

    Llnet: A deep autoencoder approach to natural low-light image enhancement,

    K. G. Lore, A. Akintayo, and S. Sarkar, “Llnet: A deep autoencoder approach to natural low-light image enhancement,” 2016

    2016

  23. [32]

    Low-light image enhancement with multi-scale attention and frequency-domain optimization,

    Z. He, W. Ran, S. Liu, K. Li, J. Lu, C. Xie, Y . Liu, and H. Lu, “Low-light image enhancement with multi-scale attention and frequency-domain optimization,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 4, pp. 2861–2875, 2024

    2024

  24. [33]

    Multiscale low-light image enhancement network with illumination constraint,

    G.-D. Fan, B. Fan, M. Gan, G.-Y . Chen, and C. L. P. Chen, “Multiscale low-light image enhancement network with illumination constraint,” IEEE Transactions on Circuits and Systems for Video Technology , vol. 32, no. 11, pp. 7403–7417, 2022

    2022

  25. [34]

    Low-light image enhancement with normalizing flow,

    Y . Wang, R. Wan, W. Yang, H. Li, L.-P. Chau, and A. C. Kot, “Low-light image enhancement with normalizing flow,” arXiv preprint arXiv:2109.05923, 2021

    2021 arXiv

  26. [35]

    Flow learning based dual networks for low-light image enhancement,

    S. Wang, C. Hu, W. Yi, Z. Cai, M. Zhai, and W. Yang, “Flow learning based dual networks for low-light image enhancement,” Neural Processing Letters, vol. 55, no. 6, pp. 8115–8130, 2023

    2023

  27. [36]

    Residual-unet generative flow model for low- light image enhancement,

    S. Wang and C. Hu, “Residual-unet generative flow model for low- light image enhancement,” in 2023 China Automation Congress (CAC) . IEEE, 2023, pp. 2046–2051

    2023

  28. [37]

    Fourllie: Boosting low-light image enhancement by fourier frequency information,

    C. Wang, H. Wu, and Z. Jin, “Fourllie: Boosting low-light image enhancement by fourier frequency information,” in Proceedings of the 31st ACM International Conference on Multimedia , 2023, pp. 7459– 7469

    2023

  29. [38]

    Iterative prompt learning for unsupervised backlit image enhancement,

    Z. Liang, C. Li, S. Zhou, R. Feng, and C. C. Loy, “Iterative prompt learning for unsupervised backlit image enhancement,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 8094–8103

    2023

  30. [39]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning . PmLR, 2021, pp. 8748–8763

    2021

  31. [40]

    Deep high-resolution representation learning for visual recognition,

    J. Wang, K. Sun, T. Cheng, B. Jiang, C. Deng, Y . Zhao, D. Liu, Y . Mu, M. Tan, X. Wang et al. , “Deep high-resolution representation learning for visual recognition,” IEEE transactions on pattern analysis and machine intelligence , vol. 43, no. 10, pp. 3349–3364, 2020

    2020

  32. [41]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Advances in Neural Information Processing Systems , H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 6840–6851

    2020

  33. [42]

    Lightendiffusion: Unsupervised low-light image enhancement with latent-retinex diffusion models,

    H. Jiang, A. Luo, X. Liu, S. Han, and S. Liu, “Lightendiffusion: Unsupervised low-light image enhancement with latent-retinex diffusion models,” in European Conference on Computer Vision. Springer, 2024, pp. 161–179

    2024

  34. [43]

    Diff-retinex: Rethinking low-light image enhancement with a generative diffusion model,

    X. Yi, H. Xu, H. Zhang, L. Tang, and J. Ma, “Diff-retinex: Rethinking low-light image enhancement with a generative diffusion model,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 12 302–12 311

    2023

  35. [44]

    Pyramid diffusion models for low-light image enhancement,

    D. Zhou, Z. Yang, and Y . Yang, “Pyramid diffusion models for low-light image enhancement,” arXiv preprint arXiv:2305.10028 , 2023

    2023 arXiv

  36. [45]

    Anligh- tendiff: Anchoring diffusion probabilistic model on low light image enhancement,

    C.-Y . Chan, W.-C. Siu, Y .-H. Chan, and H. Anthony Chan, “Anligh- tendiff: Anchoring diffusion probabilistic model on low light image enhancement,” IEEE Transactions on Image Processing , vol. 33, pp. 6324–6339, 2024

    2024

  37. [46]

    Multi-domain multi-scale diffusion model for low-light image enhancement,

    K. Shang, M. Shao, C. Wang, Y . Cheng, and S. Wang, “Multi-domain multi-scale diffusion model for low-light image enhancement,” in Pro- ceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 5, 2024, pp. 4722–4730

    2024

  38. [47]

    Low-light image enhancement with wavelet-based diffusion models,

    H. Jiang, A. Luo, H. Fan, S. Han, and S. Liu, “Low-light image enhancement with wavelet-based diffusion models,” ACM Transactions on Graphics (TOG) , vol. 42, no. 6, pp. 1–14, 2023

    2023

  39. [48]

    Scaling up visual and vision-language representation learning with noisy text supervision,

    C. Jia, Y . Yang, Y . Xia, Y .-T. Chen, Z. Parekh, H. Pham, Q. Le, Y .-H. Sung, Z. Li, and T. Duerig, “Scaling up visual and vision-language representation learning with noisy text supervision,” in International conference on machine learning . PMLR, 2021, pp. 4904–4916

    2021

  40. [49]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,

    J. Li, D. Li, C. Xiong, and S. Hoi, “Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,” in International conference on machine learning . PMLR, 2022, pp. 12 888–12 900

    2022

  41. [50]

    Lora-ir: Taming low-rank experts for efficient all-in-one image restoration,

    Y . Ai, H. Huang, and R. He, “Lora-ir: Taming low-rank experts for efficient all-in-one image restoration,” arXiv preprint arXiv:2410.15385, 2024

    2024 arXiv

  42. [51]

    A hard-to- beat baseline for training-free clip-based adaptation,

    Z. Wang, J. Liang, L. Sheng, R. He, Z. Wang, and T. Tan, “A hard-to- beat baseline for training-free clip-based adaptation,” in ICLR, 2024

    2024

  43. [52]

    Zegclip: Towards adapting clip for zero-shot semantic segmentation,

    Z. Zhou, Y . Lei, B. Zhang, L. Liu, and Y . Liu, “Zegclip: Towards adapting clip for zero-shot semantic segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2023, pp. 11 175–11 185

    2023

  44. [53]

    Improving zero-shot generalization for clip with synthesized prompts,

    Z. Wang, J. Liang, R. He, N. Xu, Z. Wang, and T. Tan, “Improving zero-shot generalization for clip with synthesized prompts,” in ICCV, 2023, pp. 3032–3042

    2023

  45. [54]

    Controlling vision-language models for universal image restoration,

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sj ¨olund, and T. B. Sch ¨on, “Controlling vision-language models for universal image restoration,” arXiv preprint arXiv:2310.01018 , 2023. 14

    2023 arXiv

  46. [55]

    Learning a simple low-light image enhancer from paired low-light instances,

    Z. Fu, Y . Yang, X. Tu, Y . Huang, X. Ding, and K.-K. Ma, “Learning a simple low-light image enhancer from paired low-light instances,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 252–22 261

    2023

  47. [56]

    The retinex theory of color vision,

    E. H. Land, “The retinex theory of color vision,” Scientific american , vol. 237, no. 6, pp. 108–129, 1977

    1977

  48. [57]

    Using color to separate reflection components,

    S. A. Shafer, “Using color to separate reflection components,” Color Research & Application , vol. 10, no. 4, pp. 210–218, 1985

    1985

  49. [58]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 10 012–10 022

    2021

  50. [59]

    Camixersr: Only details need more

    Y . Wang, Y . Liu, S. Zhao, J. Li, and L. Zhang, “Camixersr: Only details need more” attention”,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 25 837–25 846

    2024

  51. [60]

    Lyt- net: Lightweight yuv transformer-based network for low-light image enhancement,

    A. Brateanu, R. Balmez, A. Avram, C. Orhei, and C. Ancuti, “Lyt- net: Lightweight yuv transformer-based network for low-light image enhancement,” IEEE Signal Processing Letters , 2025

    2025

  52. [61]

    Ultra-high- definition low-light image enhancement: A benchmark and transformer- based method,

    T. Wang, K. Zhang, T. Shen, W. Luo, B. Stenger, and T. Lu, “Ultra-high- definition low-light image enhancement: A benchmark and transformer- based method,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 3, 2023, pp. 2654–2662

    2023

  53. [63]

    Fast context-based low-light image enhancement via neural implicit repre- sentations,

    T. Chobola, Y . Liu, H. Zhang, J. A. Schnabel, and T. Peng, “Fast context-based low-light image enhancement via neural implicit repre- sentations,” in Proceedings of the European Conference on Computer Vision (ECCV), 2024

    2024

  54. [64]

    Zero-ig: Zero-shot illumination-guided joint denoising and adaptive enhancement for low- light images,

    Y . Shi, D. Liu, L. Zhang, Y . Tian, X. Xia, and X. Fu, “Zero-ig: Zero-shot illumination-guided joint denoising and adaptive enhancement for low- light images,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 3015–3024

    2024

  55. [65]

    Lightendiffusion: Unsupervised low-light image enhancement with latent-retinex diffusion models,

    H. Jiang, A. Luo, X. Liu, S. Han, and S. Liu, “Lightendiffusion: Unsupervised low-light image enhancement with latent-retinex diffusion models,” in European Conference on Computer Vision , 2024

    2024

  56. [66]

    Sparse gradient reg- ularized deep retinex network for robust low-light image enhancement,

    W. Yang, W. Wang, H. Huang, S. Wang, and J. Liu, “Sparse gradient reg- ularized deep retinex network for robust low-light image enhancement,” IEEE Transactions on Image Processing, vol. 30, pp. 2072–2086, 2021

    2021

  57. [67]

    Contrast enhancement based on layered difference representation of 2d histograms,

    C. Lee, C. Lee, and C.-S. Kim, “Contrast enhancement based on layered difference representation of 2d histograms,”IEEE Transactions on Image Processing, vol. 22, no. 12, pp. 5372–5384, 2013

    2013

  58. [68]

    Lime: Low-light image enhancement via illumination map estimation,

    X. Guo, Y . Li, and H. Ling, “Lime: Low-light image enhancement via illumination map estimation,” IEEE Transactions on Image Processing , vol. 26, no. 2, pp. 982–993, 2017

    2017

  59. [69]

    Perceptual quality assessment for multi-exposure image fusion,

    K. Ma, K. Zeng, and Z. Wang, “Perceptual quality assessment for multi-exposure image fusion,” IEEE Transactions on Image Processing, vol. 24, no. 11, pp. 3345–3356, 2015

    2015

  60. [70]

    Naturalness preserved enhancement algorithm for non-uniform illumination images,

    S. Wang, J. Zheng, H.-M. Hu, and B. Li, “Naturalness preserved enhancement algorithm for non-uniform illumination images,” IEEE transactions on image processing : a publication of the IEEE Signal Processing Society, vol. 22, 05 2013

    2013

  61. [71]

    On the evaluation of illumination compensation algorithms,

    V . V onikakis, R. Kouskouridas, and A. Gasteratos, “On the evaluation of illumination compensation algorithms,” Multimedia Tools and Appli- cations, vol. 77, pp. 1–21, 04 2018

    2018

  62. [72]

    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

    2018

  63. [73]

    Low-light image and video enhancement: A comprehensive survey and beyond,

    S. Zheng, Y . Ma, J. Pan, C. Lu, and G. Gupta, “Low-light image and video enhancement: A comprehensive survey and beyond,” arXiv preprint arXiv:2212.10772, 2022

    2022 arXiv

  64. [74]

    Learning to see in the dark,

    C. Chen, Q. Chen, J. Xu, and V . Koltun, “Learning to see in the dark,” in CVPR, 2018

    2018

  65. [75]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” 2017

    2017

  66. [76]

    Sgdr: Stochastic gradient descent with warm restarts,

    I. Loshchilov and F. Hutter, “Sgdr: Stochastic gradient descent with warm restarts,” 2017

    2017

  67. [77]

    Image quality assess- ment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assess- ment: from error visibility to structural similarity,” IEEE Transactions on Image Processing , vol. 13, no. 4, pp. 600–612, 2004

    2004

  68. [78]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in CVPR, 2018

    2018

  69. [79]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. Hinton, “Imagenet classification with deep convolutional neural networks,” Advances in neural information processing systems, vol. 25, no. 2, 2012

    2012

  70. [80]

    No-reference image quality assessment in the spatial domain,

    A. Mittal, A. K. Moorthy, and A. C. Bovik, “No-reference image quality assessment in the spatial domain,” IEEE Transactions on Image Processing, vol. 21, no. 12, pp. 4695–4708, 2012

    2012

  71. [81]

    Making a “completely blind

    A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,” IEEE Signal Processing Letters, vol. 20, no. 3, pp. 209–212, 2013

    2013

  72. [82]

    Lednet: Joint low-light enhancement and deblurring in the dark,

    S. Zhou, C. Li, and C. C. Loy, “Lednet: Joint low-light enhancement and deblurring in the dark,” in ECCV, 2022

    2022

  73. [83]

    Global structure- aware diffusion process for low-light image enhancement,

    J. Hou, Z. Zhu, J. Hou, H. Liu, H. Zeng, and H. Yuan, “Global structure- aware diffusion process for low-light image enhancement,” Advances in Neural Information Processing Systems , vol. 36, 2024

    2024

  74. [84]

    Difflight: Integrating content and detail for low-light image enhancement,

    Y . Feng, S. Hou, H. Lin, Y . Zhu, P. Wu, W. Dong, J. Sun, Q. Yan, and Y . Zhang, “Difflight: Integrating content and detail for low-light image enhancement,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 6143–6152

    2024

  75. [85]

    Retinex- inspired unrolling with cooperative prior architecture search for low- light image enhancement,

    L. Risheng, M. Long, Z. Jiaao, F. Xin, and L. Zhongxuan, “Retinex- inspired unrolling with cooperative prior architecture search for low- light image enhancement,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , 2021

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.