Pith. sign in

REVIEW 3 major objections 6 minor 70 references

MambaLIE: Scene Light Intensity-Boosted Low-Light Image Enhancement with State Space Model

T0 review · 3 major / 6 minor · reviewed 2026-07-12 · grok-4.5

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

Pith's one-line read A mean-filter light prior plus a hybrid state-space block can enhance low-light photos better than CNNs or Transformers while staying linear in cost.

desk verdict Solid engineering adaptation of Mamba to LIE with competitive numbers and clear ablations; the joint SOTA accuracy+speed+size claim is overstated relative to the tables. read the letter →

arxiv 2607.03013 v2 pith:WAMADCSX submitted 2026-07-03 cs.CV cs.AI

Wanshu Fan , Xiangyu Li , Cong Wang , Kin-man Lam , Xin Yang , Haiyan Zhang , Dongsheng Zhou This is my paper
classification cs.CVcs.AI
keywords low-lightimageenhancementstatespacemodelMambascenelightintensitypriorLocallyEnhancedconsumerelectronicslinearcomplexity
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

Low-light photos from phones and cameras lose structure and color, and existing CNN methods see only local patches while Transformers model global context at quadratic cost. This paper claims that a simple scene-light intensity map, obtained by mean filtering the dark input and then multiplicatively gating it, supplies useful structural guidance, and that a hybrid Locally Enhanced State Space Model (LESSM) can then restore the image with linear complexity. LESSM runs a bidirectional selective state-space branch for long-range dependencies in parallel with a convolutional local branch; the two are fused inside a U-Net-style encoder-decoder that also uses channel attention and a dual-gated feed-forward network. On four standard synthetic benchmarks and five real-world sets the resulting MambaLIE network reports higher PSNR/SSIM and better no-reference scores than recent CNN and Transformer baselines, while using fewer parameters and lower FLOPs. The practical claim is that the same accuracy-efficiency trade-off makes the method deployable on resource-constrained consumer devices.

What carries the argument

Locally Enhanced State Space Model (LESSM): a dual-branch block that splits features into a bidirectional selective SSM path (linear long-range modeling) and a convolutional local-enhancement path, then concatenates and residual-adds them; the block is further guided by a mean-filter light-intensity prior used as a multiplicative gate.

What would settle it

Replace the 5 imes5 mean prior with a learned illumination estimator or with no prior on the same LOLv1/MIT-Adobe FiveK splits; if the reported PSNR/SSIM gains disappear or reverse while model size stays comparable, the claim that the fixed mean-gate is essential collapses.

Watch

Extended reading notes

Core claim

MambaLIE shows that gating a low-light image by a mean-filtered scene-light intensity prior, then processing the gated input with a hybrid Locally Enhanced State Space Model that jointly models long-range illumination and local detail under linear complexity, yields higher accuracy and lower computational cost than both CNN and Transformer low-light enhancers on the standard synthetic and real-world benchmarks.

Load-bearing premise

That a fixed-size mean filter followed by simple convolutions produces a reliable enough scene-light map that, when used only as a multiplicative gate, correctly reveals structure without introducing bias under strongly non-uniform or extreme darkness.

Editorial extensions

If this is right

  • Consumer camera pipelines can replace heavier Transformer LIE modules with a linear-complexity SSM block without sacrificing measured quality.
  • The same mean-gate + LESSM pattern can be dropped into other low-level restoration U-Nets that need global illumination consistency.
  • Real-time mobile or edge inference of low-light enhancement becomes feasible at the reported parameter and FLOP counts.
  • Downstream detectors and segmenters that fail on dark frames should improve once the enhanced output is used as input, as the paper illustrates for object detection.

Reading between the lines

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

  • Because the prior is purely spatial and fixed-scale, adaptive or multi-scale light estimation may be needed before the method can handle mixed flash/ambient or very high-dynamic-range night scenes.
  • The bidirectional SSM scan is currently applied after a simple channel split; inserting content-aware routing or frequency-aware state selection could further reduce residual color shifts noted in the limitations.
  • The architecture’s linear scaling suggests it could be extended to video low-light enhancement by treating time as an additional selective dimension without quadratic memory growth.
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

3 major / 6 minor

Summary. The paper proposes MambaLIE, a U-Net-style low-light image enhancement network that (i) multiplies the input by a scene light-intensity prior obtained from a mean filter plus depthwise/pointwise convolutions, and (ii) replaces self-attention with a Locally Enhanced State Space Model (LESSM) that combines bidirectional selective SSM scans and a local convolutional branch inside Global Light Enhancement Mamba (GLEM) blocks. The authors claim superior accuracy, speed, and model size versus recent CNN and Transformer LIE methods on four paired synthetic benchmarks and five unpaired real-world sets, with ablations on the prior, LESSM branches, channel attention, DGFN, and GLEM depth.

Significance. Adapting selective state-space models to low-light enhancement with an explicit illumination gate and a hybrid global–local block is a timely and practically motivated contribution. Linear-complexity global modeling is attractive for consumer-device imaging, the design is ablated in reasonable detail (Tables V–X), and code is released. If the efficiency claims are fully substantiated and the accuracy claims are stated more carefully, the work would be a useful addition to the LIE literature and a concrete low-level vision use case for vision Mamba variants.

major comments (3)
  1. The abstract and conclusion assert that MambaLIE “outperforms state-of-the-art CNN-based and Transformer-based LIE methods … in terms of accuracy, speed, and model size.” Accuracy is competitive but not uniformly superior. On LOLv1 (Table I) PSNR is only +0.05 dB over PPformer while LPIPS is worse (0.153 vs 0.120); on LOLv2-Real (Table II) PSNR trails LLEMamba (22.06 vs 22.20) and LPIPS is substantially worse (0.187 vs 0.054). The joint SOTA claim should be revised to match the tables (e.g., “competitive or best on most metrics”) or the weaker metrics should be explained.
  2. Speed and model-size superiority rest almost entirely on the qualitative FLOPs–PSNR scatter in Fig. 15. The manuscript does not report a head-to-head table of parameters, FLOPs, and measured latency (same resolution, same hardware) against the main Transformer baselines cited in the claim—Retinexformer, LLFormer, and PPformer. Ablation Tables V and VIII give internal timings only. Without those numbers the “accuracy + speed + size” claim is not fully supported; please add a quantitative comparison table or narrow the claim.
  3. Section III-B, Eqs. (3)–(4) and Table V: the scene light-intensity prior is a fixed-size mean filter (default 5×5) used only as a multiplicative gate. The Limitations section and Fig. 16 already show color shifts and under-/over-exposure under extreme non-uniform lighting. Given that this prior is presented as a core contribution, the paper should either quantify failure modes (e.g., NIQE/PSNR stratified by darkness or non-uniformity) or more clearly bound the operating regime in which the prior is reliable, rather than treating the 5×5 choice as generally sufficient.
minor comments (6)
  1. Notation inconsistency: Eq. (1) writes Ic = N(I) while the surrounding text uses Ie for the enhanced image; align symbols.
  2. Typos: “BorwardSSM” (should be BackwardSSM), “nder-/over-exposed”, “severrly degarded”, “Attention Fusion Block Cross-layer” capitalization, and “Yangyu Li” vs “Xiangyu Li” in the author block.
  3. The manuscript ends with duplicated and partially unrelated author biographies (including people not listed as authors). Clean the back matter before camera-ready.
  4. Fig. 1 and Fig. 15 would benefit from absolute parameter counts and a short caption note on the resolution used for FLOPs so readers can compare fairly to Retinexformer/LLFormer.
  5. No error bars or multi-seed statistics are reported for the main tables; even a brief note on run-to-run variance would strengthen confidence in the small PSNR margins (e.g., +0.05 dB).
  6. Related-work discussion of LLEMamba is brief relative to the direct numerical comparison in Table II; a short architectural contrast would help readers place LESSM.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: purely empirical architecture + supervised training on public paired data; no derivation reduces to its inputs by construction.

full rationale

MambaLIE is a standard supervised low-light enhancement network. The scene-light prior (Eq. 3–4) is a fixed mean-filter + depthwise/pointwise convolution used only as a multiplicative gate; it is not fitted to any target metric and is ablated (Table V). LESSM (Eq. 5–15) is an architectural block whose parameters are learned end-to-end under an ordinary L1 loss (Eq. 2, 16). All quantitative claims are obtained by training on public paired sets (LOL-v1/v2, FiveK) and evaluating on held-out or real-world benchmarks; no parameter is fitted to a subset and then re-presented as a prediction of a related quantity. Self-citations are ordinary method papers and do not supply uniqueness theorems or load-bearing lemmas. Hyper-parameter choices (filter size, GLEM counts) are explicitly ablated rather than smuggled in. Consequently the derivation chain never collapses to its own inputs, and the circularity score is zero.

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

As a supervised deep-learning engineering paper the load-bearing content is architectural choices and hyper-parameters rather than mathematical axioms. The free parameters are the usual training and design knobs; the invented modules are the paper’s claimed technical contribution; domain assumptions are standard Retinex-style illumination modeling and the validity of public LIE benchmarks.

free parameters (4)
  • GLEM block counts per stage = {1,2,4,8}
    Default {1,2,4,8}; ablated in Table VIII; chosen for accuracy–speed trade-off.
  • scene-light mean-filter size = 5×5
    Default 5×5; ablated in Table V; controls spatial scale of the prior.
  • learning-rate schedule and epoch counts = 1e-4 to 1e-6, 1000/4000 epochs
    Adam 1e-4 → 1e-6 cosine; 1000 epochs on FiveK, 4000 on LOL variants; standard but free choices that affect final numbers.
  • patch size and batch size = 128×128, batch 12
    128×128 patches, batch 12; affect both performance and reported FLOPs.
assumptions (3)
  • domain assumption A spatially smoothed mean-filter map is a useful proxy for scene illumination that can be multiplicatively gated onto the input.
    Stated in Section III-B; classic Retinex-inspired modeling choice, not derived.
  • ad hoc to paper Bidirectional selective state-space layers plus a local conv branch capture the long-range and local dependencies needed for LIE better than pure CNN or MHSA under linear complexity.
    Core design hypothesis of LESSM (Section III-C); validated only by ablation, not by theory.
  • domain assumption Public paired LOL/FiveK splits and no-reference metrics (NIQE, PI, MUSIQ) are adequate proxies for real-world enhancement quality.
    Standard evaluation protocol of the subfield; used throughout Section IV.
invented entities (3)
  • Locally Enhanced State Space Model (LESSM)
    purpose: Hybrid global SSM + local conv block that replaces self-attention inside the U-Net.
    Defined in Section III-C and Fig. 3(b); the paper’s main architectural novelty.
  • Global Light Enhancement Mamba (GLEM)
    purpose: Wrapper that stacks LESSM, channel attention and dual-gated FFN.
    Defined in Section III-C; packaging of known components around LESSM.
  • Scene light intensity prior (mean-filter gated input)
    purpose: Provide explicit illumination guidance before the network.
    Eqs. 3–4; simple mean filter + convs, presented as a contribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MambaLIE: Scene Light Intensity-Boosted Low-Light Image Enhancement with State Space Model." pith.science (2026). https://pith.science/paper/WAMADCSX

@misc{pith2026260703013,
  author       = {Pith},
  title        = {Pith review of: MambaLIE: Scene Light Intensity-Boosted Low-Light Image Enhancement with State Space Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WAMADCSX}},
  note         = {Machine review of arXiv:2607.03013}
}
read the original abstract

Images captured by consumer electronic devices, such as mobile phones and digital cameras, often suffer from low-light degradation due to sensor limitations and imaging pipelines, which degrades visual quality and affects downstream vision tasks. Existing methods based on Convolutional Neural Networks (CNNs) and Transformers have dominated current low-light image enhancement (LIE) due to their excellent ability to model hierarchical features. However, CNNs operate in local receptive fields that cannot model long-range dependencies, while Transformers overcome this problem but incur substantial computational costs. To address these challenges, we propose MambaLIE, a Scene Light Intensity-Boosted Low-Light Image Enhancement method based on a State Space Model (SSM). We first introduce scene light intensity to improve the structural distribution of illumination, which is then gated with the low-light input to guide enhancement. To better model the illumination while maintaining computational efficiency, we propose the Locally Enhanced State Space Model (LESSM) for efficient light enhancement. Our LESSM contains two branches: an SSM branch and a Local Enhanced branch, where the former is used to model the long-range dependencies with linear time complexity, while the latter is used to enhance local feature representations. Extensive experiments demonstrate that MambaLIE outperforms state-of-the-art CNN-based and Transformer-based LIE methods on four widely used synthetic benchmarks and five publicly available real-world benchmarks in terms of accuracy, speed, and model size, making it suitable for practical deployment on resource-constrained devices.

Figures

Figures reproduced from arXiv: 2607.03013 by the authors.

Figure 1
Figure 1. Visualization of receptive fields and enhancement [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the MambaLIE-based low-light image enhancement pipeline in consumer electronic imaging systems. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overall architecture of our proposed MambaLIE. Our MambaLIE first employs a mean filter to obtain the scene [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Visual comparison on LOLv1 [35]. MambaLIE not only restores bright areas without overexposure but also relights dark areas without introducing noise and color distortion. (a) Low-light (b) RetinexNet (c) KinD (d) RUAS (e) ELGAN (f) Restormer (g) LLFormer (h) LLIEDR (i)…
Figure 5
Figure 5. Figure 5: Results on the challenging dataset LOLv1 [ [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Results on the challenging dataset LOLv2-syn [ [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison on LOLv2-Real [49]. MambaLIE effectively preserves the brightness contrast between objects, ensuring that the distinction in lighting remains clear and natural throughout the image. MambaLIE’s strength in producing not only superior bright￾ness but al…
Figure 8
Figure 8. Figure 8: Results on the challenging dataset LOLv2-real [ [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Visual comparison of real-world images from MEF [ [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Visual comparison of real-world images from MEF [ [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Visual results on blurred low-light images from the Real-LOL-Blur [ [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: Visual results on noisy low-light scenes from the SIDD [ [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: Effect of scene intensity. Using scene intensity to gate [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]
Figure 14
Figure 14. Figure 14: LESSM vs. Transposed attention (Trans. attn.). Re [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 16
Figure 16. Figure 16: Limited enhancement quality under extreme low [PITH_FULL_IMAGE:figures/full_fig_p012_16.png]
Figure 15
Figure 15. Figure 15: Model Flops comparison on MIT-Adobe FiveK. Mam [PITH_FULL_IMAGE:figures/full_fig_p012_15.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 3 linked inside Pith

  1. [1]

    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 CVPR, 2020, pp. 1777–1786. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    2020

  2. [2]

    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, pp. 12 470–12 479

    2023

  3. [3]

    Design consider- ations of color image processing pipeline for digital cameras,

    W.-c. Kao, S.-h. Wang, L.-y. Chen, and S.-y. Lin, “Design consider- ations of color image processing pipeline for digital cameras,”IEEE Transactions on Consumer Electronics, vol. 52, no. 4, pp. 1144–1152, 2006

    2006

  4. [4]

    Unity makes strength: Coalition formation-based group-buying for timely UA V data collection,

    N. Qi, Y . Huang, W. Sun, S. Jin, T. A. Tsiftsis, Q. Wu, and X. Su, “Unity makes strength: Coalition formation-based group-buying for timely UA V data collection,” inGLOBECOM, 2022, pp. 3712–3717

    2022

  5. [5]

    Object detection in 20 years: A survey,

    Z. Zou, K. Chen, Z. Shi, Y . Guo, and J. Ye, “Object detection in 20 years: A survey,”Proc. IEEE, vol. 111, no. 3, pp. 257–276, 2023

    2023

  6. [6]

    Mseg3d: Multi-modal 3d semantic segmentation for autonomous driving,

    J. Li, H. Dai, H. Han, and Y . Ding, “Mseg3d: Multi-modal 3d semantic segmentation for autonomous driving,” inCVPR, 2023, pp. 21 694– 21 704

    2023

  7. [7]

    SRNMSM: A deep light-weight image super resolution network using multi-scale spatial and morphological feature generating residual blocks,

    A. Esmaeilzehi, M. O. Ahmad, and M. N. S. Swamy, “SRNMSM: A deep light-weight image super resolution network using multi-scale spatial and morphological feature generating residual blocks,”IEEE Trans. Broadcast., vol. 68, no. 1, pp. 58–68, 2022

    2022

  8. [8]

    Self-supervised pretraining for stereoscopic image super-resolution with parallax-aware masking,

    Z. Zhang, J. Lei, B. Peng, J. Zhu, and Q. Huang, “Self-supervised pretraining for stereoscopic image super-resolution with parallax-aware masking,”IEEE Trans. Broadcast., vol. 70, no. 2, pp. 482–491, 2024

    2024

Show all 70 references
  1. [9]

    Progressive domain translation defogging net- work for real-world fog images,

    Q. Guo and M. Zhou, “Progressive domain translation defogging net- work for real-world fog images,”IEEE Trans. Broadcast., vol. 68, no. 4, pp. 876–885, 2022

    2022

  2. [10]

    DMML: deep multi-prior and multi-discriminator learning for underwater image enhancement,

    A. Esmaeilzehi, Y . Ou, M. O. Ahmad, and M. N. S. Swamy, “DMML: deep multi-prior and multi-discriminator learning for underwater image enhancement,”IEEE Trans. Broadcast., vol. 70, no. 2, pp. 637–653, 2024

    2024

  3. [11]

    A simple and effective histogram equalization approach to image enhancement,

    H. Cheng and X. J. Shi, “A simple and effective histogram equalization approach to image enhancement,”DSP, vol. 14, no. 2, pp. 158–170, 2004

    2004

  4. [12]

    A real-time image processor with combining dynamic contrast ratio enhancement and inverse gamma correction for PDP,

    Z. Wang, Z. Liang, and C. Liu, “A real-time image processor with combining dynamic contrast ratio enhancement and inverse gamma correction for PDP,”Displays, vol. 30, no. 3, pp. 133–139, 2009

    2009

  5. [13]

    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,” inAAAI, 2023, pp. 2654–2662

    2023

  6. [14]

    Ppformer: Using pixel-wise and patch- wise cross-attention for low-light image enhancement,

    J. Dang, Y . Zhong, and X. Qin, “Ppformer: Using pixel-wise and patch- wise cross-attention for low-light image enhancement,”CVIU, vol. 241, p. 103930, 2024

    2024

  7. [15]

    Mamba: Linear-time sequence modeling with selective state spaces,

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,”arXiv:2312.00752, 2023

    2023 arXiv

  8. [16]

    Vmamba: Visual state space model,

    Y . Liu, Y . Tian, Y . Zhao, H. Yu, L. Xie, Y . Wang, Q. Ye, J. Jiao, and Y . Liu, “Vmamba: Visual state space model,” inNeurIPS, 2024

    2024

  9. [17]

    S4ND: modeling images and videos as multidimensional signals with state spaces,

    E. Nguyen, K. Goel, A. Gu, G. W. Downs, P. Shah, T. Dao, S. Baccus, and C. R ´e, “S4ND: modeling images and videos as multidimensional signals with state spaces,” inNeurIPS, 2022

    2022

  10. [18]

    Vision mamba: Efficient visual representation learning with bidirectional state space model,

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, and X. Wang, “Vision mamba: Efficient visual representation learning with bidirectional state space model,” inICML, vol. 235, 2024, pp. 62 429–62 442

    2024

  11. [19]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in CVPR, 2018, pp. 7132–7141

    2018

  12. [20]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” inNeurIPS, 2017, pp. 5998–6008

    2017

  13. [21]

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

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

    2013

  14. [22]

    An adaptive gamma correction for image enhancement,

    S. Rahman, M. Rahman, M. Abdullah-Al-Wadud, G. D. Al-Quaderi, and M. Shoyaib, “An adaptive gamma correction for image enhancement,” EJIVP, vol. 2016, p. 35, 2016

    2016

  15. [23]

    Naturalness preserved enhance- ment algorithm for non-uniform illumination images,

    S. Wang, J. Zheng, H. Hu, and B. Li, “Naturalness preserved enhance- ment algorithm for non-uniform illumination images,”IEEE TIP, vol. 22, no. 9, pp. 3538–3548, 2013

    2013

  16. [24]

    RIRO: from retinex- inspired reconstruction optimization model to deep low-light image enhancement unfolding network,

    L. Zhao, B. Chen, J. Zhang, A. Wang, and H. Bai, “RIRO: from retinex- inspired reconstruction optimization model to deep low-light image enhancement unfolding network,”IEEE TCI, vol. 10, pp. 969–983, 2024

    2024

  17. [25]

    Pixel-wise gamma correction mapping for low-light image enhancement,

    X. Li, M. Liu, and Q. Ling, “Pixel-wise gamma correction mapping for low-light image enhancement,”IEEE TCSVT, vol. 34, no. 2, pp. 681–694, 2024

    2024

  18. [26]

    Learning a single convolutional layer model for low light image enhancement,

    Y . Zhang, B. Teng, D. Yang, Z. Chen, H. Ma, G. Li, and W. Ding, “Learning a single convolutional layer model for low light image enhancement,”IEEE TCSVT, vol. 34, no. 7, pp. 5995–6008, 2024

    2024

  19. [27]

    Lfienet: Light field image enhancement network by fusing exposures of lf-dslr image pairs,

    W. Ye, T. Yan, J. Gao, and Y . Yang, “Lfienet: Light field image enhancement network by fusing exposures of lf-dslr image pairs,”IEEE TCI, vol. 9, pp. 620–635, 2023

    2023

  20. [28]

    Snr-aware low-light image enhancement,

    X. Xu, R. Wang, C. Fu, and J. Jia, “Snr-aware low-light image enhancement,” inCVPR, 2022, pp. 17 693–17 703

    2022

  21. [29]

    Dual-stream adaptive convergent low-light image enhancement network based on frequency perception,

    G. Han, K. Wu, F. Zeng, J. Liu, and S. Kwong, “Dual-stream adaptive convergent low-light image enhancement network based on frequency perception,”IEEE TCI, vol. 9, pp. 1152–1164, 2023

    2023

  22. [30]

    On the parameterization and initialization of diagonal state space models,

    A. Gu, K. Goel, A. Gupta, and C. R ´e, “On the parameterization and initialization of diagonal state space models,” inNeurIPS, 2022

    2022

  23. [31]

    Efficiently modeling long sequences with structured state spaces,

    A. Gu, K. Goel, and C. R ´e, “Efficiently modeling long sequences with structured state spaces,” inICLR, 2022

    2022

  24. [32]

    Image super- resolution using very deep residual channel attention networks,

    Y . Zhang, K. Li, K. Li, L. Wang, B. Zhong, and Y . Fu, “Image super- resolution using very deep residual channel attention networks,” in ECCV, vol. 11211, 2018, pp. 294–310

    2018

  25. [33]

    Layer normalization,

    L. J. Ba, J. R. Kiros, and G. E. Hinton, “Layer normalization,” arXiv:1607.06450, 2016

    2016 arXiv

  26. [34]

    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,” inACM MM, 2019, pp. 1632–1640

    2019

  27. [35]

    Deep retinex decomposition for low-light enhancement,

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

    2018

  28. [36]

    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,” inCVPR, 2011, pp. 97–104

    2011

  29. [37]

    A fusion-based enhancing method for weakly illuminated images,

    X. Fu, D. Zeng, Y . Huang, Y . Liao, X. Ding, and J. W. Paisley, “A fusion-based enhancing method for weakly illuminated images,”Signal Process., vol. 129, pp. 82–96, 2016

    2016

  30. [38]

    A weighted vari- ational model for simultaneous reflectance and illumination estimation,

    X. Fu, D. Zeng, Y . Huang, X. S. Zhang, and X. Ding, “A weighted vari- ational model for simultaneous reflectance and illumination estimation,” inCVPR, 2016, pp. 2782–2790

    2016

  31. [39]

    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 TPAMI, vol. 44, no. 8, pp. 4225–4238, 2022

    2022

  32. [40]

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

    R. Liu, L. Ma, J. Zhang, X. Fan, and Z. Luo, “Retinex-inspired unrolling with cooperative prior architecture search for low-light image enhancement,” inCVPR, 2021, pp. 10 561–10 570

    2021

  33. [41]

    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 TIP, vol. 30, pp. 2340–2349, 2021

    2021

  34. [42]

    Retinexdip: A unified deep framework for low-light image enhancement,

    Z. Zhao, B. Xiong, L. Wang, Q. Ou, L. Yu, and F. Kuang, “Retinexdip: A unified deep framework for low-light image enhancement,”IEEE TCSVT, vol. 32, no. 3, pp. 1076–1088, 2022

    2022

  35. [43]

    Efinet: Restoration for low-light images via enhancement-fusion iterative network,

    C. Liu, F. Wu, and X. Wang, “Efinet: Restoration for low-light images via enhancement-fusion iterative network,”IEEE TCSVT, vol. 32, no. 12, pp. 8486–8499, 2022

    2022

  36. [44]

    Uformer: A general u-shaped transformer for image restoration,

    Z. Wang, X. Cun, J. Bao, W. Zhou, J. Liu, and H. Li, “Uformer: A general u-shaped transformer for image restoration,” inCVPR, 2022, pp. 17 662–17 672

    2022

  37. [45]

    Restormer: Efficient transformer for high-resolution image restoration,

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M. Yang, “Restormer: Efficient transformer for high-resolution image restoration,” inCVPR. IEEE, 2022, pp. 5718–5729

    2022

  38. [46]

    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,” inECCV, 2024

    2024

  39. [47]

    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 TCSVT, vol. 34, no. 4, pp. 2861–2875, 2024

    2024

  40. [48]

    Deep parametric retinex decomposition model for low-light image enhancement,

    X. Li, W. Wang, X. Feng, and M. Li, “Deep parametric retinex decomposition model for low-light image enhancement,”CVIU, vol. 241, p. 103948, 2024

    2024

  41. [49]

    Sparse gradient regularized deep retinex network for robust low-light image enhance- ment,

    W. Yang, W. Wang, H. Huang, S. Wang, and J. Liu, “Sparse gradient regularized deep retinex network for robust low-light image enhance- ment,”IEEE TIP, vol. 30, pp. 2072–2086, 2021

  42. [50]

    Beyond brightening low-light images,

    Y . Zhang, X. Guo, J. Ma, W. Liu, and J. Zhang, “Beyond brightening low-light images,”IJCV, vol. 129, no. 4, pp. 1013–1037, 2021

    2021

  43. [51]

    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,” inCVPR, 2022, pp. 5891–5900

    2022

  44. [52]

    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,” inACM MM, 2023, pp. 7459–7469

    2023

  45. [53]

    Low-light image enhancement via breaking down the darkness,

    X. Guo and Q. Hu, “Low-light image enhancement via breaking down the darkness,”IJCV, vol. 131, no. 1, pp. 48–66, 2023

    2023

  46. [54]

    Implicit neural representation for cooperative low-light image enhancement,

    S. Yang, M. Ding, Y . Wu, Z. Li, and J. Zhang, “Implicit neural representation for cooperative low-light image enhancement,” inICCV, 2023, pp. 12 872–12 881

    2023

  47. [55]

    Cspn: A category- specific processing network for low-light image enhancement,

    H. Wu, C. Wang, L. Tu, C. Patsch, and Z. Jin, “Cspn: A category- specific processing network for low-light image enhancement,”IEEE TCSVT, pp. 1–1, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

    2024

  48. [56]

    Pseudo-supervised low-light image enhancement with mutual learning,

    Y . Luo, B. You, G. Yue, and J. Ling, “Pseudo-supervised low-light image enhancement with mutual learning,”IEEE TCSVT, vol. 34, no. 1, pp. 85–96, 2024

    2024

  49. [57]

    DA-DRN: A degradation-aware deep retinex network for low-light image enhancement,

    X. Wei, X. Lin, and Y . Li, “DA-DRN: A degradation-aware deep retinex network for low-light image enhancement,”DSP, vol. 144, p. 104256, 2024

    2024

  50. [58]

    Llemamba: Low- light enhancement via relighting-guided mamba with deep unfolding network,

    X. Zhang, H. Zeng, J. Pan, Q. Shen, and Y . Chen, “Llemamba: Low- light enhancement via relighting-guided mamba with deep unfolding network,”arXiv:2406.01028, 2024

    2024 arXiv

  51. [59]

    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 TIP, vol. 26, no. 2, pp. 982–993, 2017

    2017

  52. [60]

    Perceptual quality assessment for multi- exposure image fusion,

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

    2015

  53. [61]

    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,” inECCV, vol. 13666, 2022, pp. 573–589

    2022

  54. [62]

    A high-quality denoising dataset for smartphone cameras,

    A. Abdelhamed, S. Lin, and M. S. Brown, “A high-quality denoising dataset for smartphone cameras,” inCVPR, 2018, pp. 1692–1700

    2018

  55. [63]

    Scope of validity of psnr in im- age/video quality assessment,

    Q. Huynh-Thu and M. Ghanbari, “Scope of validity of psnr in im- age/video quality assessment,”Electronics Letters, vol. 44, no. 13, pp. 800–801, 2008

    2008

  56. [64]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE TIP, vol. 13, no. 4, pp. 600–612, 2004

    2004

  57. [65]

    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, pp. 586–595

    2018

  58. [66]

    Making a

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

    2013

  59. [67]

    Learning a no-reference quality metric for single-image super-resolution,

    C. Ma, C. Yang, X. Yang, and M. Yang, “Learning a no-reference quality metric for single-image super-resolution,”CVIU, vol. 158, pp. 1–16, 2017

    2017

  60. [68]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” inICLR, 2015

    2015

  61. [69]

    SGDR: stochastic gradient descent with warm restarts,

    I. Loshchilov and F. Hutter, “SGDR: stochastic gradient descent with warm restarts,” inICLR, 2017

    2017

  62. [70]

    Enhanced deep residual networks for single image super-resolution,

    B. Lim, S. Son, H. Kim, S. Nah, and K. M. Lee, “Enhanced deep residual networks for single image super-resolution,” inCVPR, 2017, pp. 1132–1140. IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTATIONAL INTELLIGENCE, VOL. XX, NO. XX, AUGUST 2023 14 Linwei Chen was born in Jiangsu,...

    2017

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.