Pith. sign in

REVIEW 5 major objections 4 minor 79 references

The paper introduces MSFT, a supervised U-shaped network for low-light image enhancement that fuses Fourier-domain amplitude information from a Retinex-based illumination estimate into multi-scale self-attention at every stage. It claims th

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 →

T0 review · deepseek-v4-flash

2026-07-31 23:18 UTC pith:SRKM3U6Q

load-bearing objection The architecture is a plausible Fourier-attention combination, but the benchmark claims are unravelled by the paper's own table inconsistencies. the 5 major comments →

arxiv 2607.24002 v1 pith:SRKM3U6Q submitted 2026-07-27 cs.CV

Low-light Image Enhancement via Multi-scale Attention combined with Fourier Transform

classification cs.CV
keywords low-light image enhancementFourier transformmulti-scale attentionRetinex priortransformeramplitude fusionimage restoration
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 paper introduces MSFT, a supervised U-shaped network for low-light image enhancement that fuses Fourier-domain amplitude information from a Retinex-based illumination estimate into multi-scale self-attention at every stage. It claims this design achieves state-of-the-art results on all seven low-light benchmarks tested, most strikingly on SDSD-outdoor where it reports 41.76 dB PSNR and 0.988 SSIM, surpassing the Retinexformer baseline by 11.92 dB PSNR. The architecture is compact at 1.25 million parameters and one-stage, with ablations attributing most of the gain to the Fourier amplitude guidance and the multi-shape attention module. If the reported comparisons hold, the method offers a practical recipe for simultaneously restoring brightness and texture in difficult low-light footage.

Core claim

On its own terms, the paper's discovery is that brightness information can be transported across the network in the frequency domain. Instead of only processing the low-light RGB image, MSFT computes a Retinex illumination prior from the max channel, concatenates it as a fourth channel, applies a Fourier transform to both the feature map and this prior, and replaces the feature map's amplitude with a weighted sum that carries the prior's brightness while preserving the original phase for structure. This Fourier-guided self-attention is repeated at three scales in a U-Net, and a Multi-Shape Synergistic Attention block further re-weights high-dimensional feature channels. The paper reports tha

What carries the argument

The central mechanism is the Fourier Transform-Guided Multi-Scale Attention (FTG-MSA) module: it runs an FFT on the low-light feature and on a four-channel Retinex-derived brightness map, sums their amplitudes, applies an inverse FFT, and uses the result to modulate value vectors in self-attention via a Hadamard product. This is supported by the Multi-Shape Synergistic Attention (MSSA) block, which combines SCSA, Dilated Square Attention, Dilated Rectangle Attention, and CMUNeXt at the highest channel scale to re-weight sparse high-dimensional features. The Retinex max-channel illumination map serves as the fourth input channel, providing the brightness prior that the frequency fusion carrie

Load-bearing premise

The comparative results assume that the published numbers for existing methods were produced under the same training and evaluation protocol as MSFT; the manuscript's own tables disagree on at least one baseline (Retinexformer is given a PSNR of 25.16 dB on LOL-v1 in one table and 22.43 dB in another), so if the baselines were not run comparably, the claimed margins could collapse.

What would settle it

Recompute every baseline metric from scratch on the same seven dataset splits with the same crop size, number of iterations, and evaluation metrics, then re-run MSFT; the central claim fails if the 11.92 dB gap on SDSD-outdoor does not persist under a unified protocol.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the results hold, MSFT provides a parameter-light (1.25M) supervised alternative to diffusion-based low-light enhancers, suitable for paired-data settings.
  • The amplitude-fusion attention mechanism could be adopted as a drop-in module in other U-shaped restoration networks for tasks where global brightness priors matter.
  • The reported SSIM of 0.988 on SDSD-outdoor suggests strong structural fidelity, not just brightness recovery, on dynamic video frames.
  • Ablations attribute most of the SDSD-outdoor gain (6.78 dB) to the Fourier module, implying frequency-domain guidance is the primary performance driver rather than the transformer backbone alone.
  • The paper's statistical analysis shows high stability on SDSD (std ≤ 0.03 dB) but weaker stability on SMID and LOL-v2-real (std ≥ 1.38 dB), indicating margins may vary across runs.

Where Pith is reading between the lines

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

  • The same amplitude-fusion trick might generalize to other domains with global illumination priors, such as underwater or nighttime dehazing, though the paper does not test this.
  • Because the paper notes the model fails on strong-light pollution and requires paired clean data, a natural extension is to replace the paired Retinex prior with a self-supervised or zero-shot estimate and measure how much of the gain survives.
  • A simpler test than building the full architecture: compute the amplitude-swapped Fourier reconstruction from Figure 1 and check whether the perceptual gain matches what the network achieves; if a closed-form amplitude swap already explains most of the improvement, the attention module may be secondary.
  • The internal inconsistency in baseline tables suggests a reproducibility check: if Retinexformer's true LOL-v1 score is 22.43 rather than 25.16, MSFT's lead shrinks from 2.32 dB to over 5 dB, changing the headline comparison.

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

5 major / 4 minor

Summary. The paper proposes MSFT, a supervised low-light image enhancement network with a U-shaped architecture combining Fourier-transform-guided multi-scale attention (FTG-MSA), a Multi-Shape Synergistic Attention (MSSA) module built from SCSA, MSA (DSA/DRA), and CMUNeXt, plus multi-scale feature alignment. The central claim is empirical: MSFT outperforms state-of-the-art methods on seven benchmarks (LOL-v1, LOL-v2-real/syn, SID, SMID, SDSD-indoor/outdoor), with highlighted gains of 11.92 dB PSNR over Retinexformer on SDSD-outdoor (41.76 dB, SSIM 0.988). The paper also reports ablations, complexity statistics, and repeated-run statistics.

Significance. If the reported results are accurate, the claimed margins—particularly the 9–16 dB gains on SDSD datasets—would be a substantial advance in low-light image enhancement. The architecture description is fairly detailed, and the attempt to report run-to-run variability is a good practice. However, the paper provides no code, trained models, or baseline reproduction protocol, and several core tables contain mechanical inconsistencies. Because the headline contribution is an empirical benchmark comparison, these issues directly affect the validity of the stated results. The strengths are the clear modular design and the inclusion of repeated-run statistics; the weakness is that the numerical evidence is not internally reliable.

major comments (5)
  1. [Table 4 vs Table 5, §4.2] In Table 4, Retinexformer's LOL-v1 PSNR is 25.16 dB; in Table 5, the same model is reported as 22.43 dB on the same dataset. 22.43 dB is exactly Restormer's LOL-v1 value in Table 4. This collision indicates that at least one baseline number is wrong. Since the abstract's 11.92 dB margin on SDSD-outdoor and the 'outperforms Retinexformer' claim rest on the Table 4 baseline values, this is a load-bearing error.
  2. [Tables 7 and 8, §4.3] The 'W/O MSA' row in Table 7 is numerically identical in PSNR on all seven datasets (26.94, 25.95, 25.55, 26.54, 29.80, 39.04, 41.75) to the 'W/O DRA' row in Table 8, with SSIM equal except for one digit on SID. Since MSA is described as a parallel combination of DSA and DRA, removing the whole MSA should differ from removing only the DRA branch. The duplication suggests one row was copied from the other, so the ablation study does not support the claimed attribution of gains to DRA or MSA.
  3. [§4.1, Tables 3 and 4] The MSFT values in Table 4 are exactly the 12-run means reported in Table 3 on all seven datasets (e.g., 27.48, 25.78, 41.76). Yet §4.1 states that the model with the highest validation PSNR was chosen from the 12 runs. With reported standard deviations of 0.44–1.41 dB, the best-of-12 value cannot equal the mean to two decimals on every dataset. Table 4 therefore appears to report mean performance, which contradicts the stated selection protocol and makes the comparison against baselines potentially unfair.
  4. [Table 4, §4.2] The categories in Table 4 are mislabeled: SNR-Net and Retinexformer are listed in the 'UL' (unsupervised) group, and the text explicitly calls them 'unsupervised methods' in §4.2. Both are supervised methods trained on paired data. This misclassification obscures the comparison structure and raises doubt about the care with which the baseline numbers were assembled.
  5. [Overall empirical evidence] No code, trained weights, or baseline reproduction details are provided, and the baseline numbers are not sourced beyond the table. Given the internal inconsistencies listed above, the central claim that MSFT outperforms all SOTA methods is not supported by the submitted evidence. The authors should make the evaluation protocol public and rerun or verify all baselines under a common training setting.
minor comments (4)
  1. [Abstract; Figure 1; §4.2] There are several typos: 'priori channels' should be 'prior channels'; the Figure 1 caption contains 'of of'; and 'superivised' should be 'supervised' in §4.2.
  2. [Table 6] The table formatting is difficult to parse: check marks are used as row labels, and entries such as '31.78✓ SSIM 0.843' mix values and labels in a single cell. Please reformat so each row is clearly labeled.
  3. [Abstract; §3; §3.2] The terminology is inconsistent: the abstract describes MSFT as 'one-stage', while the contributions state a 'two-stage integrated network', and §3.2 describes a 'three-scale U-shaped architecture'. Please clarify the stage/scale terminology.
  4. [Eq. (9)] The attention formula is nonstandard and ambiguous. As written, the softmax argument is K_i^T Q_i (dimension d_k × d_k) rather than the usual Q_i K_i^T (dimension HW × HW), and the modulated value (bY_i ⊙ V_i) appears before the softmax. Please clarify the intended computation and verify the shapes.

Circularity Check

0 steps flagged

No significant circularity: the central claim is an empirical benchmark result, not a derivation from its own inputs.

full rationale

The paper's central claim is empirical: a supervised model (MSFT) is trained on paired low-/normal-light images and evaluated on held-out test splits, reporting PSNR/SSIM. The Fourier amplitude/phase prior and the Retinex max-channel illumination initialization (Eqs. 1-5) are external observations and architectural inductive biases, not conclusions derived from the model's own outputs. No equation defines the reported metrics in terms of the fitted parameters or the baseline numbers, and no fitted quantity is renamed as a prediction. The only identified self-citation is the 3DLUT baseline (Ref. [72], co-authored by one of the current authors), which is used merely as a comparison method and is not load-bearing for the claimed superiority of MSFT. The ablations test known modules (SCSA, MSA, CMUNeXt) and measure their individual contributions, which is standard empirical validation rather than circular justification. The table inconsistencies noted by the skeptic, if real, are experimental-validity or reporting concerns, not circularity of the derivation chain.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central claim is an empirical benchmark comparison, not a derivation, so the free parameters are standard supervised-learning hyperparameters plus the learned network weights. No scientific constants are fitted. The main load-bearing assumptions are the Retinex max-channel illumination prior, the Fourier amplitude/brightness interpretation, the sufficiency of PSNR/SSIM, and the validity of borrowed attention modules; the invented entities are two new architectural modules whose independent evidence is limited to the paper's own inconsistent ablations.

free parameters (6)
  • Base channel width C = 32
    Hand-chosen model width; all hierarchical feature channels are 2^i*C; affects capacity and FLOPs (1.25M params, 18.07 GFLOPs).
  • Number of U-Net scales = 3
    Architecture choice; the 'multi-scale guidance' claim depends on this specific depth.
  • Training patch size = 128×128
    Random crops of 128×128; affects receptive field and statistics seen by the model.
  • Learning rate schedule = 2e-6 → 1e-6 (cosine)
    Hand-tuned; no sensitivity analysis is provided.
  • Training iterations and batch size = 1.5e5 iterations, batch 8
    Training budget; chosen by hand.
  • Number of attention heads k = unspecified (Figure 2b shows k=1 case)
    A hyperparameter in Eqs. 6-8 but not stated explicitly for the final model.
axioms (5)
  • domain assumption Retinex decomposition I_low = L ⊙ R and illumination estimate L̃ = max_c I_low (Eqs. 1-2).
    The four-channel guidance input is built from this prior; if the max-channel map is not a valid illumination estimate, the guidance signal is unprincipled.
  • domain assumption Fourier amplitude encodes brightness while phase encodes structure/noise (Figure 1).
    This observation justifies fusing amplitude spectra in FTG-MSA; Figure 1 provides anecdotal examples, not a general proof.
  • domain assumption PSNR and SSIM are sufficient measures of enhancement quality for the central claim.
    The paper evaluates only PSNR/SSIM; no perceptual, task-based, or user-study metrics are reported.
  • domain assumption Paired low/normal-light training data capture the target deployment distribution.
    The method is supervised and the paper's own Limitations section admits it fails for unpaired, noisy, or strong-light scenarios.
  • domain assumption The cited modules SCSA [32], MSA [33], and CMUNeXt [66] behave as described when embedded in MSFT.
    The paper does not re-derive these modules; their effectiveness inside the new architecture is only asserted via ablations.
invented entities (2)
  • FTG-MSA (Fourier Transform-Guided Multi-Scale Attention) module no independent evidence
    purpose: Core attention block that fuses the amplitude spectrum of a four-channel guidance map with the low-light feature amplitude before computing self-attention (Eqs. 4-9).
    No code or third-party replication; only the paper's own ablations support its contribution, and the FFT ablation in Table 9 is the only direct evidence.
  • MSSA (Multi-Shape Synergistic Attention) module no independent evidence
    purpose: Processes the highest-dimensional guidance features using SCSA, MSA (DSA+DRA), and CMUNeXt to extract texture information (Eqs. 10-17).
    Assembled from published modules; its claimed effect is internally inconsistent (W/O MSA and W/O DRA ablation rows are identical), and no external validation exists.

pith-pipeline@v1.3.0-alltime-deepseek · 21643 in / 20407 out tokens · 168556 ms · 2026-07-31T23:18:11.918014+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Low-light Image Enhancement via Multi-scale Attention combined with Fourier Transform." pith.science (2026). https://pith.science/paper/SRKM3U6Q

@misc{pith2026260724002,
  author       = {Pith},
  title        = {Pith review of: Low-light Image Enhancement via Multi-scale Attention combined with Fourier Transform},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SRKM3U6Q}},
  note         = {Machine review of arXiv:2607.24002}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Low-light image enhancement (LLIE) aims to improve image quality and clarity in diverse and demanding low-illumination environments. However, existing deep learning-based LLIE methods struggle to accurately capture real-world illumination and restore texture details, largely because their algorithmic strengths remain underutilized. To address these issues, we present a supervised frequency domain deep learning network for LLIE, named multi-scale attention combined with the Fourier transform (MSFT) which adopts a U-shaped, one-stage architecture that infuses guidance from low-light images into the network by channeling it through multi-scale attention. We further fuse the amplitude information from priori channels with that of the low-light image in MSFT's self-created module, and carry out multi-scale guidance along with the network. Subsequently, to better enhance the faint feature, such as fine content and textures, and to better fuse global context confidence in the decoding stage, we separately introduce a multi-shape synergistic attention and a lightweight network that effectively integrate information in high-dimensional space to embed into the superlative feature space channel containing rich texture information. Extensive experiments conducted on LOL, SID, SMID, and SDSD datasets demonstrate that MSFT significantly outperforms state-of-the-art competitors. For example, compared with Retinexformer, our method achieves a peak signal-to-noise ratio of up to 41.76 decibels on the SDSD-outdoor dataset with an increase of 11.92 decibels and a structural similarity index of 0.988 with a 13.80% improvement.

Figures

Figures reproduced from arXiv: 2607.24002 by Jian Long, Wenbin Du, Zhu Cao.

Figure 1
Figure 1. Figure 1: (a) Fourier transforms of of amplitude addition. (b) Fourier transforms of phase addition. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Overall structure of MSFT. (b) Schematic diagram of FTG-MSA. (c) Module diagram of FTGT. (d) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: visualization results of repeated runs for each dataset. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison charts on four datasets. SMID, SDSD-indoor, and SDSD-outdoor, respectively. Especially on the SDSD indoor and outdoor datasets, the SSIM of MIRNet and IPI are increased by 12.7 % and 16.24%, respectively. Compared with SOTA supervised deep learning methods, MSFT achieves 3.34, 7.03, 3.73, 2.63, 5.09, 12.93, and 16.48 dB improvements on the seven benchmarks in [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of qualitative comparisons with SNR-Net, and Retinexformer on the LOLv1 dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of qualitative comparisons with SNR-Net, and Retinexformer on the LOLv2 synthetic dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of qualitative comparisons with SNR-Net, and Retinexformer on the LOLv2 real dataset. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of qualitative comparisons with SNR-Net, and Retinexformer on the SDSD-indoor (top row), [PITH_FULL_IMAGE:figures/full_fig_p017_8.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

79 extracted references · 7 canonical work pages

  1. [1]

    J. Tan, L. Huang, Z. Chen, R. Qu, C. Li, Darksegnet: Low-light semantic segmentation network based on image pyramid, Signal Processing: Image Communication 135 (2025) 117265.doi: https://doi.org/10.1016/j.image.2025.117265. URLhttps://www.sciencedirect.com/science/article/pii/S0923596525000128

  2. [2]

    M. M. Memon, M. A. Hashmani, A. Z. Junejo, S. S. Rizvi, K. Raza, Unified deeplabv3+ for semi-dark image semantic segmentation, Sensors 22 (14) (2022).doi:10.3390/s22145312. URLhttps://www.mdpi.com/1424-8220/22/14/5312

  3. [3]

    D. Dai, L. V. Gool, Dark model adaptation: Semantic image segmentation from daytime to nighttime, in: 2018 21st International Conference on Intelligent Transportation Systems (ITSC), 2018, pp. 3819–3824.doi:10.1109/ITSC.2018.8569387

  4. [4]

    X. Yin, Z. Yu, Z. Fei, W. Lv, X. Gao, Pe-yolo: Pyramid enhancement network for dark object detection (2023).arXiv:2307.10953. URLhttps://arxiv.org/abs/2307.10953

  5. [5]

    Sasagawa, H

    Y. Sasagawa, H. Nagahara, Yolo in the dark - domain adaptation method for merging multiple models, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Computer Vision – ECCV 2020, Springer International Publishing, Cham, 2020, pp. 345–359

  6. [6]

    W. Wang, W. Yang, J. Liu, Hla-face: Joint high-low adaptation for low light face detection (2021).arXiv:2104.01984. URLhttps://arxiv.org/abs/2104.01984

  7. [7]

    C. Wang, J. Jiang, K. Jiang, X. Liu, Low-light face super-resolution via illumination, struc- ture, and texture associated representation, in: Proceedings of the Thirty-Eighth AAAI Con- ference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of 21 Artificial Intelligence and Fourteenth Symposium on Educational Advances in...

  8. [8]

    Zhang, D

    C. Zhang, D. Lee, Advancing nighttime object detection through image enhancement and do- main adaptation, Applied Sciences 14 (18) (2024).doi:10.3390/app14188109. URLhttps://www.mdpi.com/2076-3417/14/18/8109

  9. [9]

    M. A. Qadar, C. Nie, Y. Tan, H. Zhang, Sun, Zhifeng, A comparative study of nighttime object detection with datasets from australia and china, in: 2020 Chinese Automation Congress (CAC), 2020, pp. 5407–5412.doi:10.1109/CAC51589.2020.9327278

  10. [10]

    M. Rea, X. Mou, J. Bullough, Scene brightness of illuminated interiors, Lighting Research & Technology 48 (7) (2016) 823–831

  11. [11]

    Park, K.-H

    S.-C. Park, K.-H. Park, J.-H. Chang, Luminance-degradation compensation based on multi- stream self-attention to address thin-film transistor-organic light emitting diode burn-in, Sensors 21 (9) (2021).doi:10.3390/s21093182. URLhttps://www.mdpi.com/1424-8220/21/9/3182

  12. [12]

    Zhang, X

    S. Zhang, X. Shen, Z. Lin, R. Měch, J. P. Costeira, J. M. Moura, Learning to understand image blur, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 6586–6595

  13. [13]

    Patel, A

    P. Patel, A. Bhandari, A review on image contrast enhancement techniques, Int. J. Online Sci 5 (5) (2019) 14–18

  14. [14]

    C. Lee, C. Lee, C.-S. Kim, Contrast enhancement based on layered difference representation of 2d histograms, IEEE transactions on image processing 22 (12) (2013) 5372–5384

  15. [15]

    J. Liu, D. Xu, W. Yang, M. Fan, H. Huang, Benchmarking low-light image enhancement and beyond, International Journal of Computer Vision 129 (2021) 1153–1184

  16. [16]

    Parthasarathy, P

    S. Parthasarathy, P. Sankaran, An automated multi scale retinex with color restoration for image enhancement, in: 2012 National Conference on Communications (NCC), IEEE, 2012, pp. 1–5

  17. [17]

    W. Wu, J. Weng, P. Zhang, X. Wang, W. Yang, J. Jiang, Uretinex-net: Retinex-based deep un- folding network for low-light image enhancement, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5901–5910

  18. [18]

    X. Fu, D. Zeng, Y. Huang, X.-P. Zhang, X. Ding, A weighted variational model for simultaneous reflectance and illumination estimation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2782–2790

  19. [19]

    Z. Wang, A. Bovik, H. Sheikh, E. Simoncelli, Image quality assessment: from error visibility to structural similarity, IEEE Transactions on Image Processing 13 (4) (2004) 600–612.doi: 10.1109/TIP.2003.819861

  20. [20]

    Schmidhuber, Deep learning in neural networks: An overview, Neural Networks 61 (2015) 85–117.doi:https://doi.org/10.1016/j.neunet.2014.09.003

    J. Schmidhuber, Deep learning in neural networks: An overview, Neural Networks 61 (2015) 85–117.doi:https://doi.org/10.1016/j.neunet.2014.09.003. URLhttps://www.sciencedirect.com/science/article/pii/S0893608014002135 22

  21. [21]

    Zhang, L

    R. Zhang, L. Guo, S. Huang, B. Wen, Rellie: Deep reinforcement learning for customized low- light image enhancement, arXiv preprint arXiv:2107.05830 (2021)

  22. [22]

    Jiang, Y

    H. Jiang, Y. Zheng, Learning to see moving objects in the dark, in: 2019 IEEE/CVF Inter- national Conference on Computer Vision (ICCV), 2019, pp. 7323–7332.doi:10.1109/ICCV. 2019.00742

  23. [23]

    K. G. Lore, A. Akintayo, S. Sarkar, Llnet: A deep autoencoder approach to natural low-light image enhancement, CoRR abs/1511.03995 (2015).arXiv:1511.03995. URLhttp://arxiv.org/abs/1511.03995

  24. [24]

    1662–1671

    Y.Wu, C.Pan, G.Wang, Y.Yang, J.Wei, C.Li, H.T.Shen, Learningsemantic-awareknowledge guidance for low-light image enhancement, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1662–1671

  25. [25]

    X. Xu, R. Wang, C.-W. Fu, J. Jia, Snr-aware low-light image enhancement, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 17714–17724

  26. [26]

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

  27. [27]

    Jiang, A

    H. Jiang, A. Luo, H. Fan, S. Han, S. Liu, Low-light image enhancement with wavelet-based diffusion models, ACM Transactions on Graphics (TOG) 42 (6) (2023) 1–14

  28. [28]

    Zhang, Z

    J. Zhang, Z. Li, J. Zhang, Y. Wang, Retinex-based self-conditioned diffusion model for low- light image enhancement, in: ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5.doi:10.1109/ICASSP49660.2025. 10890851

  29. [29]

    Y. Feng, S. Hou, H. Lin, Y. Zhu, P. Wu, W. Dong, J. Sun, Q. Yan, Y. Zhang, Difflight: Integrating content and detail for low-light image enhancement, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2024, pp. 6143–6152.doi: 10.1109/CVPRW63382.2024.00619

  30. [30]

    Jiang, A

    H. Jiang, A. Luo, X. Liu, S. Han, S. Liu, Lightendiffusion: Unsupervised low-light image en- hancement with latent-retinex diffusion models, in: European Conference on Computer Vision, Springer, 2025, pp. 161–179

  31. [31]

    H. Zhou, W. Dong, X. Liu, Y. Zhang, G. Zhai, J. Chen, Low-light image enhancement via generative perceptual priors, arXiv preprint arXiv:2412.20916 (2024)

  32. [32]

    Y. Si, H. Xu, X. Zhu, W. Zhang, Y. Dong, Y. Chen, H. Li, Scsa: Exploring the synergistic effects between spatial and channel attention, arXiv preprint arXiv:2407.05128 (2024)

  33. [33]

    M. Li, Y. Fu, T. Zhang, J. Liu, D. Dou, C. Yan, Y. Zhang, Latent diffusion enhanced rectangle transformer for hyperspectral image restoration, IEEE Transactions on Pattern Analysis and Machine Intelligence 47 (1) (2025) 549–564.doi:10.1109/TPAMI.2024.3475249

  34. [34]

    J. Pan, X. Liu, Y. Bai, D. Zhai, J. Jiang, D. Zhao, Illumination-aware low-light image enhance- ment with transformer and auto-knee curve, ACM Transactions on Multimedia Computing, Communications and Applications (2024). 23

  35. [35]

    Veluchamy, B

    M. Veluchamy, B. Subramani, Image contrast and color enhancement using adaptive gamma correction and histogram equalization, Optik (2019) 329–337doi:10.1016/j.ijleo.2019.02. 054. URLhttp://dx.doi.org/10.1016/j.ijleo.2019.02.054

  36. [36]

    Rahman, M

    S. Rahman, M. M. Rahman, M. Abdullah-Al-Wadud, G. D. Al-Quaderi, M. Shoyaib, An adap- tive gamma correction for image enhancement, EURASIP Journal on Image and Video Process- ing 2016 (2016) 1–13

  37. [37]

    Abdullah-Al-Wadud, M

    M. Abdullah-Al-Wadud, M. H. Kabir, M. A. A. Dewan, O. Chae, A dynamic histogram equaliza- tion for image contrast enhancement, IEEE transactions on consumer electronics 53 (2) (2007) 593–600

  38. [38]

    URLhttp://dx.doi.org/10.1016/j.dsp.2003.07.002

    H.Cheng, X.Shi, Asimpleandeffectivehistogramequalizationapproachtoimageenhancement, Digital Signal Processing (2004) 158–170doi:10.1016/j.dsp.2003.07.002. URLhttp://dx.doi.org/10.1016/j.dsp.2003.07.002

  39. [39]

    S. M. Pizer, E. P. Amburn, J. D. Austin, R. Cromartie, A. Geselowitz, T. Greer, B. ter Haar Romeny, J. B. Zimmerman, K. Zuiderveld, Adaptive histogram equalization and its vari- ations, Computer vision, graphics, and image processing 39 (3) (1987) 355–368

  40. [40]

    Lee, J.-L

    C.-H. Lee, J.-L. Shih, C.-C. Lien, C.-C. Han, Adaptive multiscale retinex for image contrast enhancement, in: 2013 International Conference on Signal-Image Technology & Internet-Based Systems, 2013, pp. 43–50.doi:10.1109/SITIS.2013.19

  41. [41]

    W. Wang, B. Li, J. Zheng, S. Xian, J. Wang, A fast multi-scale retinex algorithm for color image enhancement, in: 2008 International Conference on Wavelet Analysis and Pattern Recognition, Vol. 1, 2008, pp. 80–85.doi:10.1109/ICWAPR.2008.4635754

  42. [42]

    D. J. Jobson, Z.-u. Rahman, G. A. Woodell, A multiscale retinex for bridging the gap between color images and the human observation of scenes, IEEE Transactions on Image processing 6 (7) (1997) 965–976

  43. [43]

    J. Ma, X. Fan, J. Ni, X. Zhu, C. Xiong, Multi-scale retinex with color restoration image en- hancement based on gaussian filtering and guided filtering, International Journal of Modern Physics B 31 (16-19) (2017) 1744077

  44. [44]

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

  45. [45]

    L. Ma, T. Ma, R. Liu, X. Fan, Z. Luo, Toward fast, flexible, and robust low-light image enhance- ment, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5637–5646

  46. [46]

    Jiang, X

    Y. Jiang, X. Gong, D. Liu, Y. Cheng, C. Fang, X. Shen, J. Yang, P. Zhou, Z. Wang, En- lightengan: Deep light enhancement without paired supervision, IEEE transactions on image processing 30 (2021) 2340–2349

  47. [47]

    D. Zhou, Z. Yang, Y. Yang, Pyramid diffusion models for low-light image enhancement (2023). arXiv:2305.10028. URLhttps://arxiv.org/abs/2305.10028 24

  48. [48]

    Jiang, A

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

  49. [49]

    Chan, W.-C

    C.-Y. Chan, W.-C. Siu, Y.-H. Chan, H. Anthony Chan, Anlightendiff: Anchoring diffusion probabilistic model on low light image enhancement, IEEE Transactions on Image Processing 33 (2024) 6324–6339.doi:10.1109/TIP.2024.3486610

  50. [50]

    Alexey, An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv: 2010.11929 (2020)

    D. Alexey, An image is worth 16x16 words: Transformers for image recognition at scale, arXiv preprint arXiv: 2010.11929 (2020)

  51. [51]

    12504–12513

    Y.Cai, H.Bian, J.Lin, H.Wang, R.Timofte, Y.Zhang, Retinexformer: One-stageretinex-based transformer for low-light image enhancement, in: Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 12504–12513

  52. [52]

    G. Fan, Z. Yao, M. Gan, Illumination-aware and structure-guided transformer for low-light imageenhancement, ComputerVisionandImageUnderstanding252(2025)104276.doi:https: //doi.org/10.1016/j.cviu.2024.104276. URLhttps://www.sciencedirect.com/science/article/pii/S1077314224003576

  53. [53]

    W. Dong, Y. Min, H. Zhou, J. Chen, Towards scale-aware low-light enhancement via structure- guided transformer design (2025).arXiv:2504.14075. URLhttps://arxiv.org/abs/2504.14075

  54. [54]

    H. A. Ewaidat, Y. E. Brag, A. W. Y. E’layan, A. Almakhadmeh, Frequency-guided u-net: Leveraging attention filter gates and fast fourier transformation for enhanced medical image segmentation, arXiv preprint arXiv:2405.00683 (2024)

  55. [55]

    X. M. Li, Image enhancement in the fractional fourier domain, in: 2013 6th International Congress on Image and Signal Processing (CISP), Vol. 1, IEEE, 2013, pp. 299–303

  56. [56]

    C. Wang, H. Wu, 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

  57. [57]

    Huang, Y

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

  58. [58]

    Li, C.-L

    C. Li, C.-L. Guo, M. Zhou, Z. Liang, S. Zhou, R. Feng, C. C. Loy, Embedding fourier for ultra-high-definition low-light image enhancement, arXiv preprint arXiv:2302.11831 (2023)

  59. [59]

    Li, C.-L

    C. Li, C.-L. Guo, M. Zhou, Z. Liang, S. Zhou, R. Feng, C. C. Loy, Embedding fourier for ultra-high-definition low-light image enhancement, in: ICLR, 2023

  60. [60]

    X. Lv, S. Zhang, C. Wang, Y. Zheng, B. Zhong, C. Li, L. Nie, Fourier priors-guided diffusion for zero-shot joint low-light enhancement and deblurring, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 25378–25388

  61. [61]

    Zhang, P

    T. Zhang, P. Liu, M. Zhao, H. Lv, Dmfourllie: dual-stage and multi-branch fourier network for low-light image enhancement, in: Proceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 7434–7443. 25

  62. [62]

    22252–22261

    Z.Fu, Y.Yang, X.Tu, Y.Huang, X.Ding, K.-K.Ma, Learningasimplelow-lightimageenhancer frompairedlow-lightinstances, in: ProceedingsoftheIEEE/CVFconferenceoncomputervision and pattern recognition, 2023, pp. 22252–22261

  63. [63]

    X. Yi, H. Xu, H. Zhang, L. Tang, 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. 12302–12311

  64. [64]

    Zhang, X

    Y. Zhang, X. Guo, J. Ma, W. Liu, J. Zhang, Beyond brightening low-light images, International Journal of Computer Vision 129 (2021) 1013–1037

  65. [65]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image seg- mentation, in: N. Navab, J. Hornegger, W. M. Wells, A. F. Frangi (Eds.), Medical Image Com- puting and Computer-Assisted Intervention – MICCAI 2015, Springer International Publishing, Cham, 2015, pp. 234–241

  66. [66]

    Prasetyo, R

    H. Prasetyo, R. B. Ashidiqy, U. Salamah, Cmaunext: An efficient neural network based on large kernel and multi-dimensional attention module for breast tumor segmentation, in: 2024 IEEE International Conference on Smart Mechatronics (ICSMech), 2024, pp. 89–94.doi:10.1109/ ICSMech62936.2024.10812276

  67. [67]

    W. Yang, W. Wang, H. Huang, S. Wang, J. Liu, Sparse gradient regularized deep retinex network for robust low-light image enhancement, IEEE Transactions on Image Processing 30 (2021) 2072–2086.doi:10.1109/TIP.2021.3050850

  68. [68]

    C. Chen, Q. Chen, M. Do, V. Koltun, Seeing motion in the dark, in: 2019 IEEE/CVF Inter- national Conference on Computer Vision (ICCV), 2019, pp. 3184–3193.doi:10.1109/ICCV. 2019.00328

  69. [69]

    3291–3300.doi:10.1109/CVPR.2018

    C.Chen, Q.Chen, J.Xu, V.Koltun, Learningtoseeinthedark, in: 2018IEEE/CVFConference on Computer Vision and Pattern Recognition, 2018, pp. 3291–3300.doi:10.1109/CVPR.2018. 00347

  70. [70]

    R. Wang, X. Xu, C.-W. Fu, J. Lu, B. Yu, J. Jia, Seeing dynamic scene in the dark: A high- quality video dataset with mechatronic alignment, in: 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 9680–9689.doi:10.1109/ICCV48922.2021.00956

  71. [72]

    H. Zeng, J. Cai, L. Li, Z. Cao, L. Zhang, Learning image-adaptive 3d lookup tables for high performance photo enhancement in real-time., IEEE Trans. Pattern Anal. Mach. Intell. 44 (4) (2022) 2058–2073. URLhttp://dblp.uni-trier.de/db/journals/pami/pami44.html#ZengCLCZ22 26

  72. [73]

    W. Yang, S. Wang, Y. Fang, Y. Wang, J. Liu, Band representation-based semi-supervised low- light image enhancement: Bridging the gap between signal fidelity and perceptual quality, IEEE Transactions on Image Processing 30 (2021) 3461–3473.doi:10.1109/TIP.2021.3062184

  73. [74]

    R. Wang, Q. Zhang, C.-W. Fu, X. Shen, W.-S. Zheng, J. Jia, Underexposed photo enhancement using deep illumination estimation, in: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 6842–6850.doi:10.1109/CVPR.2019.00701

  74. [76]

    H. Chen, Y. Wang, T. Guo, C. Xu, Y. Deng, Z. Liu, S. Ma, C. Xu, C. Xu, W. Gao, Pre-trained image processing transformer, in: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 12294–12305.doi:10.1109/CVPR46437.2021.01212

  75. [78]

    X. Li, W. Wang, X. Feng, M. Li, Deep parametric retinex decomposition model for low-light imageenhancement, ComputerVisionandImageUnderstanding241(2024)103948.doi:https: //doi.org/10.1016/j.cviu.2024.103948. URLhttps://www.sciencedirect.com/science/article/pii/S1077314224000298

  76. [79]

    K. Xu, X. Yang, B. Yin, R. W. Lau, Learning to restore low-light images via decomposition-and- enhancement, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 2278–2287.doi:10.1109/CVPR42600.2020.00235

  77. [80]

    Zhang, J.Zhang, X.Guo, Kindling thedarkness: Apractical low-light image enhancer(2019)

    Y. Zhang, J.Zhang, X.Guo, Kindling thedarkness: Apractical low-light image enhancer(2019). arXiv:1905.04161. URLhttps://arxiv.org/abs/1905.04161

  78. [81]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, M. Yang, Restormer: Efficient trans- former for high-resolution image restoration, in: 2022 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), 2022, pp. 5718–5729.doi:10.1109/CVPR52688.2022. 00564

  79. [82]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, M.-H. Yang, L. Shao, Learning enriched features for fast image restoration and enhancement, IEEE Transactions on Pattern Analysis and Machine Intelligence 45 (2) (2023) 1934–1948.doi:10.1109/TPAMI.2022.3167175. 27