pith. sign in

arxiv: 2605.25569 · v2 · pith:CM4EOR6Wnew · submitted 2026-05-25 · 💻 cs.CV

ControlLight: Towards Controllable, Consistent, and Generalizable Low-Light Enhancement

Yufeng Yang , Jianzhuang Liu , Jisheng Chu , Yuqi Peng , Xianfang Zeng , Jiancheng Huang , Shifeng Chen This is my paper

Pith reviewed 2026-06-29 23:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords low-light enhancementcontrollable enhancementflow matching lossimage datasetgeneralizationreal-world imagescomputer vision
0
0 comments X

The pith

ControlLight enables continuous control of enhancement strength in low-light images while keeping outputs consistent.

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

Existing deep learning methods for brightening low-light images train on limited data with one fixed target, which limits how well they work across real scenes and user needs. ControlLight builds a large dataset of real degraded images labeled with many continuous levels of illumination strength. It adds a misalignment-aware weighted flow matching loss to keep image structures aligned as the enhancement level varies. The result lets a user adjust brightness on the same photo toward better results without introducing new inconsistencies or artifacts. If the approach holds, it would make low-light enhancement more practical for editing and downstream vision tasks.

Core claim

The authors construct a large-scale dataset of real-world degraded images with continuous illumination-strength supervision and introduce a misalignment-aware weighted flow matching loss that preserves image structure across continuous enhancement strengths. ControlLight uses these to enable users to edit low-light images by controlling enhancement strength flexibly while preserving visual consistency and realism, achieving state-of-the-art performance and strong generalization.

What carries the argument

The misalignment-aware weighted flow matching loss that preserves image structure across continuous enhancement strengths.

If this is right

  • Users can flexibly control the strength of enhancement on real degraded low-light images.
  • Outputs remain visually consistent and realistic across different control strengths.
  • The method reaches state-of-the-art performance against existing low-light enhancement approaches.
  • Generalization to real-world scenarios improves over prior fixed-target methods.

Where Pith is reading between the lines

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

  • The continuous-supervision dataset could serve as a benchmark for testing other adjustable image-restoration models.
  • Interactive photo editors might adopt similar loss terms to let users drag a slider without breaking detail alignment.
  • The same controllability pattern could extend to related tasks such as denoising or contrast adjustment if paired with appropriate labels.

Load-bearing premise

The newly constructed large-scale dataset supplies reliable continuous illumination-strength supervision that, together with the misalignment-aware weighted flow matching loss, produces consistent outputs across different control strengths.

What would settle it

A side-by-side comparison where increasing the control strength on the same input produces visible structural misalignment or loss of realism that single-strength baselines avoid would falsify the consistency claim.

Figures

Figures reproduced from arXiv: 2605.25569 by Jiancheng Huang, Jianzhuang Liu, Jisheng Chu, Shifeng Chen, Xianfang Zeng, Yufeng Yang, Yuqi Peng.

Figure 1
Figure 1. Figure 1: Given a low-light input, ControlLight supports continuous adjustment of the enhancement strength from s = 0 to s = 1, producing smooth and controllable restoration consistent results across real-world scenes. Abstract Existing deep learning-based low-light enhancement methods are typically trained on limited datasets with single enhancement targets, which restricts their general￾ization ability and control… view at source ↗
Figure 2
Figure 2. Figure 2: Main data construction pipeline of Light100K. FLUX.2-klein-9B is used to generate normal-light references from real low-light images. We then apply Retinex-inspired decomposition and selective interpolation to construct highly consistent and continuous pseudo-paired data from each pair (I0, I1). These methods are generally trained under fixed supervision and therefore tend to produce results with a single … view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison of intermediate pseudo-GT construction at s = 0.5. Retinex-based interpolation yields more natural illumination transitions and better local contrast than direct alpha blending, making it better suited for continuous low-light enhancement. leaving approximately 20K high-quality paired samples. For each retained pair (I0, I1), we further construct a continuous pseudo-paired training group:… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of edge misalignment and the effect of weighted flow matching. Compared with standard flow matching, the proposed LwFM produces enhanced results with weaker edge-difference responses and better structural alignment to the input. the dominant differences between I0 and I1 mainly arise from brightness and color variations. After normalizing illumination, however, the remaining high-frequency re… view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the proposed framework. (a) During training, the low-light input image and a fixed restoration prompt are encoded and fed into FLUX.2-klein, where LoRA is used for efficient fine-tuning. The enhancement strength s modulates both the LoRA scaling factor and the pseudo ground-truth selection. (b) The edge mask is generated offline from input and target edges, producing the weight map Wfs. At infe… view at source ↗
Figure 6
Figure 6. Figure 6: Visual comparison on DICM Benchmark. The arrows indicate increasing enhancement strength from left to right. Compared with CLE Diffusion, our method produces smoother and more continuous enhancement transitions while better preserving natural color and scene structure. Input Retinexformer CIDNet CLE Diffusion LLFormer DarkIR QuadPrior Ours [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparison on RealIR-Bench. The arrows indicate increasing enhancement strength from left to right. Compared with traditional methods, our method achieves more natural restoration and better preserves scene structure. The arrows indicate increasing enhancement strength, showing that our model provides continuous and controllable low-light enhancement. 4 Experiments 4.1 Quantitative Metrics and Evalu… view at source ↗
Figure 8
Figure 8. Figure 8: Visual comparison on LOL-v1 benchmark. Our zero-shot results may deviate from the ground truth in color appearance, but they provide natural visual quality with preserved structures and textures. The outputs at different enhancement strengths s show smooth and approximately linear low-light enhancement control [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison with state-of-the-art low-light enhancement methods on the paired benchmarks LOL-v1 and LSWR. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison with universal continuous image editing methods on RealIR-Bench for low-light enhancement. All methods are evaluated under the same four-point control strengths. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative results of ControlLight on the reported benchmarks under four continuous enhancement strengths. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
read the original abstract

Existing deep learning-based low-light enhancement methods are typically trained on limited datasets with single enhancement targets, which restricts their generalization ability and controllability in real-world applications. To overcome these limitations, we propose ControlLight, a controllable, consistent, and generalizable framework for low-light enhancement. We first construct a large-scale dataset of real-world degraded images with continuous illumination-strength supervision. To further ensure consistent outputs under different control strengths, we introduce a misalignment-aware weighted flow matching loss that preserves image structure across continuous enhancement strengths. ControlLight allows users to edit real-world degraded low-light images toward satisfactory enhancement results by flexibly controlling the strength while preserving visual consistency and realism. Extensive experiments show that ControlLight achieves state-of-the-art performance against existing low-light enhancement approaches while demonstrating strong continuous controllability and generalization to real-world scenarios.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The paper proposes ControlLight, a framework for low-light image enhancement that targets controllability, consistency, and generalizability. It constructs a large-scale real-world dataset providing continuous illumination-strength supervision and introduces a misalignment-aware weighted flow matching loss to maintain structural consistency across varying control strengths. The method is presented as enabling flexible user control over enhancement while preserving realism, with claims of state-of-the-art performance and strong generalization to real-world scenarios.

Significance. If the central claims are substantiated by rigorous experiments, the work would represent a meaningful advance by shifting low-light enhancement from fixed-target training to continuous, user-controllable outputs. This could improve applicability in domains requiring adjustable enhancement levels, such as photography and surveillance, provided the new dataset and loss function demonstrably deliver the promised consistency and generalization without introducing artifacts.

major comments (2)
  1. [Abstract] Abstract: The abstract asserts SOTA results, continuous controllability, and generalization to real-world scenarios, but the provided manuscript text contains no quantitative tables, ablation studies, error analysis, or experimental controls to support these claims. Without access to the full methods, results, and dataset construction details, the load-bearing assertions cannot be evaluated and the central claims remain unsubstantiated.
  2. [Abstract] The weakest assumption identified—that the newly constructed dataset supplies reliable continuous illumination-strength supervision and that the misalignment-aware weighted flow matching loss produces consistent outputs—cannot be assessed because the manuscript text does not include the relevant sections on dataset construction, loss derivation, or validation experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their review and for highlighting the need for clear substantiation of the abstract claims. The full manuscript contains dedicated sections on dataset construction, loss derivation, quantitative results, ablations, and generalization experiments. We address the comments below and note that if the reviewed version appeared incomplete, we will ensure all supporting materials are explicitly referenced and highlighted in any revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The abstract asserts SOTA results, continuous controllability, and generalization to real-world scenarios, but the provided manuscript text contains no quantitative tables, ablation studies, error analysis, or experimental controls to support these claims. Without access to the full methods, results, and dataset construction details, the load-bearing assertions cannot be evaluated and the central claims remain unsubstantiated.

    Authors: The complete manuscript includes Section 3 on dataset construction with continuous illumination-strength supervision from real-world captures, Section 4.2 deriving the misalignment-aware weighted flow matching loss, and Section 5 with quantitative tables comparing against SOTA methods, ablation studies on loss components and controllability, error analysis, and generalization tests on unseen real-world data. These provide the required experimental controls and results. We disagree that the claims are unsubstantiated and will add cross-references from the abstract to these sections for clarity. revision: no

  2. Referee: [Abstract] The weakest assumption identified—that the newly constructed dataset supplies reliable continuous illumination-strength supervision and that the misalignment-aware weighted flow matching loss produces consistent outputs—cannot be assessed because the manuscript text does not include the relevant sections on dataset construction, loss derivation, or validation experiments.

    Authors: Section 3 details the dataset construction process, including how continuous illumination-strength labels were obtained via controlled real-world captures and alignment procedures. Section 4.2 provides the full derivation of the misalignment-aware weighted flow matching loss, explaining the weighting mechanism for structural consistency. Section 5.2 includes validation experiments measuring consistency metrics across control strengths. These sections directly address the assumptions; we will ensure they are not overlooked in future versions by adding explicit pointers. revision: no

Circularity Check

0 steps flagged

No significant circularity; claims rest on new dataset and loss

full rationale

The paper's derivation chain introduces a newly constructed large-scale dataset providing continuous illumination-strength supervision and a misalignment-aware weighted flow matching loss to ensure consistency. These are presented as original contributions supporting the controllability and generalization claims. No equation or step reduces by construction to fitted inputs from the same data, self-definition, or load-bearing self-citation chains. The abstract and described method remain self-contained against external benchmarks without the forbidden patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be identified from the provided text.

pith-pipeline@v0.9.1-grok · 5686 in / 1146 out tokens · 38808 ms · 2026-06-29T23:00:37.370194+00:00 · methodology

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

59 extracted references · 14 canonical work pages · 8 internal anchors

  1. [1]

    Continuous, subject-specific attribute control in t2i models by identifying semantic directions

    Stefan Andreas Baumann, Felix Krause, Michael Neumayr, Nick Stracke, Melvin Sevi, Vincent Tao Hu, and Björn Ommer. Continuous, subject-specific attribute control in t2i models by identifying semantic directions. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 13231– 13241, 2025

    2025

  2. [2]

    Retinexformer: One-stage retinex-based transformer for low-light image enhancement

    Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Timofte, and Yulun Zhang. Retinexformer: One-stage retinex-based transformer for low-light image enhancement. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12504–12513, 2023

    2023

  3. [3]

    Freemorph: Tuning-free generalized image morphing with diffusion model

    Yukang Cao, Chenyang Si, Jinghao Wang, and Ziwei Liu. Freemorph: Tuning-free generalized image morphing with diffusion model. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 18111–18120, 2025

    2025

  4. [4]

    Learning to see in the dark

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. Learning to see in the dark. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3291–3300, 2018

    2018

  5. [5]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. InForty-first International Conference on Machine Learning, 2024

    2024

  6. [6]

    Benito, Alvaro Garcia, and Marcos V

    Daniel Feijoo, Juan C. Benito, Alvaro Garcia, and Marcos V . Conde. Darkir: Robust low-light image restoration. InProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 10879–10889, June 2025

    2025

  7. [7]

    Concept sliders: Lora adaptors for precise control in diffusion models

    Rohit Gandikota, Joanna Materzy ´nska, Tingrui Zhou, Antonio Torralba, and David Bau. Concept sliders: Lora adaptors for precise control in diffusion models. InEuropean Conference on Computer Vision, pages 172–188. Springer, 2024

    2024

  8. [8]

    Seedream 3.0 Technical Report

    Yu Gao, Lixue Gong, Qiushan Guo, Xiaoxia Hou, Zhichao Lai, Fanshi Li, Liang Li, Xiaochen Lian, Chao Liao, Liyang Liu, et al. Seedream 3.0 technical report.arXiv preprint arXiv:2504.11346, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    Lime: Low-light image enhancement via illumination map estimation

    Xiaojie Guo, Yu Li, and Haibin Ling. Lime: Low-light image enhancement via illumination map estimation. IEEE Transactions on Image Processing, 26(2):982–993, 2016

    2016

  10. [10]

    R2rnet: Low-light image enhancement via real-low to real-normal network.Journal of Visual Communication and Image Representation, 90:103712, 2023

    Jiang Hai, Zhu Xuan, Ren Yang, Yutong Hao, Fengzhu Zou, Fang Lin, and Songchen Han. R2rnet: Low-light image enhancement via real-low to real-normal network.Journal of Visual Communication and Image Representation, 90:103712, 2023

    2023

  11. [11]

    Lora: Low-rank adaptation of large language models.ICLR, 1(2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 1(2):3, 2022

    2022

  12. [12]

    Diffusion model-based image editing: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Yi Huang, Jiancheng Huang, Yifan Liu, Mingfu Yan, Jiaxi Lv, Jianzhuang Liu, Wei Xiong, He Zhang, Liangliang Cao, and Shifeng Chen. Diffusion model-based image editing: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  13. [13]

    Springer, 2005

    Bernd Jähne.Digital image processing. Springer, 2005

    2005

  14. [14]

    Degrade is upgrade: Learning degradation for low-light image enhancement

    Kui Jiang, Zhongyuan Wang, Zheng Wang, Chen Chen, Peng Yi, Tao Lu, and Chia-Wen Lin. Degrade is upgrade: Learning degradation for low-light image enhancement. InProceedings of the AAAI conference on artificial intelligence, volume 36, pages 1078–1086, 2022

    2022

  15. [15]

    Enlightengan: Deep light enhancement without paired supervision.IEEE Transactions on Image Processing, 30:2340–2349, 2021

    Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision.IEEE Transactions on Image Processing, 30:2340–2349, 2021

    2021

  16. [16]

    Geditbench v2: A human-aligned benchmark for general image editing.arXiv preprint arXiv:2603.28547, 2026

    Zhangqi Jiang, Zheng Sun, Xianfang Zeng, Yufeng Yang, Xuanyang Zhang, Yongliang Wu, Wei Cheng, Gang Yu, Xu Yang, and Bihan Wen. Geditbench v2: A human-aligned benchmark for general image editing.arXiv preprint arXiv:2603.28547, 2026

    work page arXiv 2026

  17. [17]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 5148–5157, 2021

    2021

  18. [18]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742, 2025. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    The retinex theory of color vision.Scientific american, 237(6):108–129, 1977

    Edwin H Land. The retinex theory of color vision.Scientific american, 237(6):108–129, 1977

    1977

  20. [20]

    Contrast enhancement based on layered difference represen- tation of 2d histograms.IEEE Transactions on Image Processing, 22(12):5372–5384, 2013

    Chulwoo Lee, Chul Lee, and Chang-Su Kim. Contrast enhancement based on layered difference represen- tation of 2d histograms.IEEE Transactions on Image Processing, 22(12):5372–5384, 2013

    2013

  21. [21]

    Flow Matching for Generative Modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  22. [22]

    Step1X-Edit: A Practical Framework for General Image Editing

    Shiyu Liu, Yucheng Han, Peng Xing, Fukun Yin, Rui Wang, Wei Cheng, Jiaqi Liao, Yingming Wang, Honghao Fu, Chunrui Han, et al. Step1x-edit: A practical framework for general image editing.arXiv preprint arXiv:2504.17761, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    completely blind

    Anish Mittal, Rajiv Soundararajan, and Alan C Bovik. Making a “completely blind” image quality analyzer. IEEE Signal processing letters, 20(3):209–212, 2012

    2012

  24. [24]

    Kontinuous kontext: Continuous strength control for instruction-based image editing.arXiv preprint arXiv:2510.08532, 2025

    Rishubh Parihar, Or Patashnik, Daniil Ostashev, R Venkatesh Babu, Daniel Cohen-Or, and Kuan-Chieh Wang. Kontinuous kontext: Continuous strength control for instruction-based image editing.arXiv preprint arXiv:2510.08532, 2025

    work page arXiv 2025

  25. [25]

    Styleclip: Text-driven manipulation of stylegan imagery

    Or Patashnik, Zongze Wu, Eli Shechtman, Daniel Cohen-Or, and Dani Lischinski. Styleclip: Text-driven manipulation of stylegan imagery. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 2085–2094, October 2021

    2085

  26. [26]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    2023

  27. [27]

    Tara: Token-aware lora for composable personalization in diffusion models

    Yuqi Peng, Lingtao Zheng, Yufeng Yang, Yi Huang, Mingfu Yan, Jianzhuang Liu, and Shifeng Chen. Tara: Token-aware lora for composable personalization in diffusion models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 8385–8393, 2026

    2026

  28. [28]

    Contrast-limited adaptive histogram equalization: Speed and effectiveness stephen m

    Stephen M Pizer. Contrast-limited adaptive histogram equalization: Speed and effectiveness stephen m. pizer, r. eugene johnston, james p. ericksen, bonnie c. yankaskas, keith e. muller medical image display research group. InProceedings of the first conference on visualization in biomedical computing, Atlanta, Georgia, volume 337, page 2, 1990

    1990

  29. [29]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational Conference on Machine Learning, pages 8748–8763. PmLR, 2021

    2021

  30. [30]

    Gendeg: Diffusion-based degradation synthesis for generalizable all-in-one image restoration

    Sudarshan Rajagopalan, Nithin Gopalakrishnan Nair, Jay N Paranjape, and Vishal M Patel. Gendeg: Diffusion-based degradation synthesis for generalizable all-in-one image restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 28144–28154, 2025

    2025

  31. [31]

    Seedance 2.0: Advancing Video Generation for World Complexity

    Team Seedance, De Chen, Liyang Chen, Xin Chen, Ying Chen, Zhuo Chen, Zhuowei Chen, Feng Cheng, Tianheng Cheng, Yufeng Cheng, et al. Seedance 2.0: Advancing video generation for world complexity. arXiv preprint arXiv:2604.14148, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  32. [32]

    Seedream 4.0: Toward Next-generation Multimodal Image Generation

    Team Seedream, Yunpeng Chen, Yu Gao, Lixue Gong, Meng Guo, Qiushan Guo, Zhiyao Guo, Xiaoxia Hou, Weilin Huang, Yixuan Huang, et al. Seedream 4.0: Toward next-generation multimodal image generation.arXiv preprint arXiv:2509.20427, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  33. [33]

    Alchemist: Parametric control of material properties with diffusion models

    Prafull Sharma, Varun Jampani, Yuanzhen Li, Xuhui Jia, Dmitry Lagun, Fredo Durand, Bill Freeman, and Mark Matthews. Alchemist: Parametric control of material properties with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24130– 24141, 2024

    2024

  34. [34]

    Gemini: A Family of Highly Capable Multimodal Models

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  35. [35]

    LongCat-Image Technical Report

    Meituan LongCat Team, Hanghang Ma, Haoxian Tan, Jiale Huang, Junqiang Wu, Jun-Yan He, Lishuai Gao, Songlin Xiao, Xiaoming Wei, Xiaoqi Ma, Xunliang Cai, Yayong Guan, and Jie Hu. Longcat-image technical report.arXiv preprint arXiv:2512.07584, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  36. [36]

    Qwen3 technical report, 2025

    Qwen Team. Qwen3 technical report, 2025

    2025

  37. [37]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 11

    2017

  38. [38]

    Exploring clip for assessing the look and feel of images

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Exploring clip for assessing the look and feel of images. InProceedings of the AAAI conference on artificial intelligence, volume 37, pages 2555–2563, 2023

    2023

  39. [39]

    Seededit 3.0: Fast and high-quality generative image editing.arXiv preprint arXiv:2506.05083, 2025

    Peng Wang, Yichun Shi, Xiaochen Lian, Zhonghua Zhai, Xin Xia, Xuefeng Xiao, Weilin Huang, and Jian- chao Yang. Seededit 3.0: Fast and high-quality generative image editing.arXiv preprint arXiv:2506.05083, 2025

    work page arXiv 2025

  40. [40]

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

    Tao Wang, Kaihao Zhang, Tianrun Shen, Wenhan Luo, Bjorn Stenger, and Tong Lu. Ultra-high-definition low-light image enhancement: A benchmark and transformer-based method. InProceedings of the AAAI conference on artificial intelligence, volume 37, pages 2654–2662, 2023

    2023

  41. [41]

    Zero-reference low-light enhancement via physical quadruple priors

    Wenjing Wang, Huan Yang, Jianlong Fu, and Jiaying Liu. Zero-reference low-light enhancement via physical quadruple priors. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26057–26066, 2024

    2024

  42. [42]

    Zero-reference low-light enhancement via physical quadruple priors, 2024

    Wenjing Wang, Huan Yang, Jianlong Fu, and Jiaying Liu. Zero-reference low-light enhancement via physical quadruple priors, 2024

    2024

  43. [43]

    Low-light image enhancement with normalizing flow

    Yufei Wang, Renjie Wan, Wenhan Yang, Haoliang Li, Lap-Pui Chau, and Alex Kot. Low-light image enhancement with normalizing flow. InProceedings of the AAAI conference on artificial intelligence, volume 36, pages 2604–2612, 2022

    2022

  44. [44]

    Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 13(4):600–612, 2004

    2004

  45. [45]

    Mamballie: Implicit retinex-aware low light enhancement with global-then-local state space.Advances in neural information processing systems, 37:27440–27462, 2024

    Jiangwei Weng, Zhiqiang Yan, Ying Tai, Jianjun Qian, Jian Yang, and Jun Li. Mamballie: Implicit retinex-aware low light enhancement with global-then-local state space.Advances in neural information processing systems, 37:27440–27462, 2024

    2024

  46. [46]

    Qwen-image technical report, 2025

    Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, Yuxiang Chen, Zecheng Tang, Zekai Zhang, Zhengyi Wang, An Yang, Bowen Yu, Chen Cheng, Dayiheng Liu, Deqing Li, Hang Zhang, Hao Meng, Hu Wei, Jingyuan Ni, Kai Chen, Kuan Cao, Liang Peng, Lin Qu, Minggang Wu, Peng Wang, Shuting Yu, Tingkun...

    2025

  47. [47]

    Q-bench: A benchmark for general-purpose foundation models on low-level vision

    Haoning Wu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Annan Wang, Chunyi Li, Wenxiu Sun, Qiong Yan, Guangtao Zhai, and Weisi Lin. Q-bench: A benchmark for general-purpose foundation models on low-level vision. InICLR, 2024

    2024

  48. [48]

    Recoro: Re gion-co ntrollable ro bust light enhancement with user-specified imprecise masks

    Dejia Xu, Hayk Poghosyan, Shant Navasardyan, Yifan Jiang, Humphrey Shi, and Zhangyang Wang. Recoro: Re gion-co ntrollable ro bust light enhancement with user-specified imprecise masks. InProceedings of the 30th ACM International Conference on Multimedia, pages 1376–1386, 2022

    2022

  49. [49]

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

    Qingsen Yan, Yixu Feng, Cheng Zhang, Guansong Pang, Kangbiao Shi, Peng Wu, Wei Dong, Jinqiu Sun, and Yanning Zhang. Hvi: A new color space for low-light image enhancement. InProceedings of the computer vision and pattern recognition conference, pages 5678–5687, 2025

    2025

  50. [50]

    Maniqa: Multi-dimension attention network for no-reference image quality assessment

    Sidi Yang, Tianhe Wu, Shuwei Shi, Shanshan Lao, Yuan Gong, Mingdeng Cao, Jiahao Wang, and Yujiu Yang. Maniqa: Multi-dimension attention network for no-reference image quality assessment. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1191–1200, 2022

    2022

  51. [51]

    Sparse gradient regularized deep retinex network for robust low-light image enhancement.IEEE Transactions on Image Processing, 30:2072–2086, 2021

    Wenhan Yang, Wenjing Wang, Haofeng Huang, Shiqi Wang, and Jiaying Liu. Sparse gradient regularized deep retinex network for robust low-light image enhancement.IEEE Transactions on Image Processing, 30:2072–2086, 2021

    2072

  52. [52]

    Realrestorer: Towards generalizable real-world image restoration with large-scale image editing models.arXiv preprint arXiv:2603.25502, 2026

    Yufeng Yang, Xianfang Zeng, Zhangqi Jiang, Fukun Yin, Jianzhuang Liu, Wei Cheng, Shiyu Liu, Yuqi Peng, Gang YU, Shifeng Chen, et al. Realrestorer: Towards generalizable real-world image restoration with large-scale image editing models.arXiv preprint arXiv:2603.25502, 2026

    work page arXiv 2026

  53. [53]

    Cle diffusion: Controllable light enhancement diffusion model

    Yuyang Yin, Dejia Xu, Chuangchuang Tan, Ping Liu, Yao Zhao, and Yunchao Wei. Cle diffusion: Controllable light enhancement diffusion model. InProceedings of the 31st ACM International Conference on Multimedia, pages 8145–8156, 2023. 12

    2023

  54. [54]

    Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild

    Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, and Chao Dong. Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25669–25680, 2024

    2024

  55. [55]

    Slideredit: Continuous image editing with fine-grained instruction control.arXiv preprint arXiv:2511.09715, 2025

    Arman Zarei, Samyadeep Basu, Mobina Pournemat, Sayan Nag, Ryan Rossi, and Soheil Feizi. Slideredit: Continuous image editing with fine-grained instruction control.arXiv preprint arXiv:2511.09715, 2025

    work page arXiv 2025

  56. [56]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 586–595, 2018

    2018

  57. [57]

    Kindling the darkness: A practical low-light image enhancer

    Yonghua Zhang, Jiawan Zhang, and Xiaojie Guo. Kindling the darkness: A practical low-light image enhancer. InProceedings of the 27th ACM international conference on multimedia, pages 1632–1640, 2019

    2019

  58. [58]

    Pyramid diffusion models for low-light image enhancement

    Dewei Zhou, Zongxin Yang, and Yi Yang. Pyramid diffusion models for low-light image enhancement. arXiv preprint arXiv:2305.10028, 2023

    work page arXiv 2023

  59. [59]

    a dark photo

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. InProceedings of the IEEE International Conference on Computer Vision, pages 2223–2232, 2017. 13 Appendix A Continuous Pseudo-Paired Data Construction Details During the construction of Light100K, we first collect ...

    2017