pith. sign in

arxiv: 2604.15170 · v1 · submitted 2026-04-16 · 💻 cs.CV

OmniLight: One Model to Rule All Lighting Conditions

Youngjin Oh , Junyoung Park , Junhyeong Kwon , Nam Ik Cho This is my paper

Pith reviewed 2026-05-10 11:39 UTC · model grok-4.3

classification 💻 cs.CV
keywords image restorationshadow removallighting normalizationmixture of expertswavelet domaingeneralizationNTIRE challengeadverse lighting
0
0 comments X p. Extension

The pith

A single model using wavelet-domain experts restores images across diverse lighting conditions by training jointly on multiple datasets.

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

The paper compares a specialized per-dataset model called DINOLight against a unified model called OmniLight that incorporates a Wavelet Domain Mixture-of-Experts architecture. It demonstrates that the unified model can be trained across all datasets without suffering from negative transfer while still achieving strong performance on lighting restoration tasks. Both approaches rank at the top in the shadow removal, lighting normalization, and related tracks of the NTIRE 2026 Challenge. This comparison highlights how architecture choices affect the trade-off between specialization and generalization in real-world image restoration under varying illumination.

Core claim

OmniLight shows that a Wavelet Domain Mixture-of-Experts model trained jointly on multiple lighting restoration datasets can generalize effectively across cast shadows and irregular illumination without requiring separate models for each domain, matching the perceptual quality of specialized baselines while simplifying deployment.

What carries the argument

The Wavelet Domain Mixture-of-Experts (WD-MoE) architecture, which decomposes features into wavelet sub-bands and routes lighting-specific variations to different expert sub-networks during joint training.

If this is right

  • A single trained model suffices for multiple lighting restoration tasks instead of maintaining separate specialized networks.
  • Joint training on diverse lighting datasets improves robustness to unseen illumination patterns in downstream computer vision applications.
  • Wavelet-domain processing isolates lighting effects from scene content, enabling more stable feature routing in the experts.
  • Reduced model storage and inference overhead for real-world systems that encounter mixed lighting conditions.

Where Pith is reading between the lines

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

  • The same WD-MoE routing principle could be tested on other restoration problems such as low-light enhancement or color constancy without per-task retraining.
  • If the wavelet decomposition proves key to avoiding negative transfer, similar frequency-domain gating might apply to video or multi-modal restoration pipelines.
  • Deployment in edge devices would benefit from the unified model's lower memory footprint compared to an ensemble of specialized models.

Load-bearing premise

The mixture-of-experts routing in the wavelet domain can separate and process lighting variations from different datasets without one domain harming performance on others.

What would settle it

A joint-training experiment on the combined datasets where the unified WD-MoE model shows lower PSNR, SSIM, or perceptual scores on any single dataset's test set than the corresponding specialized DINOLight baseline.

Figures

Figures reproduced from arXiv: 2604.15170 by Junhyeong Kwon, Junyoung Park, Nam Ik Cho, Youngjin Oh.

Figure 1
Figure 1. Figure 1: Restoration examples of the baseline DINOLight [ [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of the proposed OmniLight. The [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Detailed illustration of the dual-branched OmniLight [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results of lighting-related image restoration of NTIRE 2026 Challenge [ [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: t-SNE visualization [51] of the routing guidance vector. Input DINOLight OmniLight [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Failure and success cases of restoration from NTIRE [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Adverse lighting conditions, such as cast shadows and irregular illumination, pose significant challenges to computer vision systems by degrading visibility and color fidelity. Consequently, effective shadow removal and ALN are critical for restoring underlying image content, improving perceptual quality, and facilitating robust performance in downstream tasks. However, while achieving state-of-the-art results on specific benchmarks is a primary goal in image restoration challenges, real-world applications often demand robust models capable of handling diverse domains. To address this, we present a comprehensive study on lighting-related image restoration by exploring two contrasting strategies. We leverage a robust framework for ALN, DINOLight, as a specialized baseline to exploit the characteristics of each individual dataset, and extend it to OmniLight, a generalized alternative incorporating our proposed Wavelet Domain Mixture-of-Experts (WD-MoE) that is trained across all provided datasets. Through a comparative analysis of these two methods, we discuss the impact of data distribution on the performance of specialized and unified architectures in lighting-related image restoration. Notably, both approaches secured top-tier rankings across all three lighting-related tracks in the NTIRE 2026 Challenge, demonstrating their outstanding perceptual quality and generalization capabilities. Our codes are available at https://github.com/OBAKSA/Lighting-Restoration.

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 / 2 minor

Summary. The manuscript presents DINOLight as a specialized baseline for adverse lighting normalization (ALN) and shadow removal on individual datasets, and OmniLight, a unified model using a proposed Wavelet Domain Mixture-of-Experts (WD-MoE) architecture trained jointly across all datasets. It claims both approaches achieved top-tier rankings in the three lighting-related tracks of the NTIRE 2026 Challenge and discusses the impact of data distribution on specialized versus unified models for lighting restoration.

Significance. If the empirical results hold, demonstrating that a single WD-MoE model can generalize across diverse lighting conditions without negative transfer or per-domain specialization would be valuable for real-world robustness in computer vision. The open release of code at the cited GitHub repository is a clear strength supporting reproducibility.

major comments (2)
  1. [Abstract] Abstract: the central generalization claim (that OmniLight matches specialized performance without negative transfer) is load-bearing but unsupported by any reported quantitative metrics, per-track scores, ablation on expert routing, or direct DINOLight vs. OmniLight comparisons on individual domains; only qualitative rankings are stated.
  2. [Abstract] The skeptic concern is valid on current evidence: challenge rankings alone do not confirm the WD-MoE avoids degradation on any single lighting condition relative to its DINOLight counterpart, as no head-to-head PSNR/SSIM/perceptual scores or cross-domain transfer analysis is supplied.
minor comments (2)
  1. [Abstract] Expand the acronym ALN on first use.
  2. [Abstract] The abstract mentions a 'comparative analysis' but does not indicate where the supporting tables or figures appear.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for your thorough review and constructive comments on our manuscript. We appreciate the opportunity to clarify and strengthen the presentation of our results. Below, we provide point-by-point responses to the major comments.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central generalization claim (that OmniLight matches specialized performance without negative transfer) is load-bearing but unsupported by any reported quantitative metrics, per-track scores, ablation on expert routing, or direct DINOLight vs. OmniLight comparisons on individual domains; only qualitative rankings are stated.

    Authors: We agree that the abstract and manuscript would benefit from more explicit quantitative evidence to support the generalization claims. Although the NTIRE 2026 Challenge rankings are determined by quantitative metrics (PSNR, SSIM, and perceptual scores), we did not report the specific values or perform direct comparisons in the submitted version. In the revised manuscript, we will add a table presenting the per-track scores for both DINOLight and OmniLight, direct head-to-head comparisons on individual domains, and an ablation study on the expert routing in the WD-MoE architecture to demonstrate the absence of negative transfer. revision: yes

  2. Referee: [Abstract] The skeptic concern is valid on current evidence: challenge rankings alone do not confirm the WD-MoE avoids degradation on any single lighting condition relative to its DINOLight counterpart, as no head-to-head PSNR/SSIM/perceptual scores or cross-domain transfer analysis is supplied.

    Authors: We acknowledge the validity of this concern. To address it, the revised version will include the requested head-to-head quantitative comparisons and cross-domain transfer analysis between the specialized DINOLight models and the unified OmniLight model. This will provide concrete evidence regarding performance on individual lighting conditions. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper reports empirical results from standard supervised training of DINOLight (specialized per-dataset baseline) and OmniLight (WD-MoE unified model) on lighting restoration tasks, followed by their rankings in the external NTIRE 2026 Challenge. No mathematical derivations, predictions, or equations are described that reduce to fitted parameters or self-referential definitions by construction. Claims rest on challenge performance rather than any load-bearing self-citation chain or ansatz smuggled via prior work. This is a typical empirical ML paper with independent external validation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

The central claim rests on the empirical effectiveness of joint training with WD-MoE; no explicit free parameters, axioms, or invented physical entities are stated in the abstract, though standard deep-learning training assumptions apply.

invented entities (1)
  • Wavelet Domain Mixture-of-Experts (WD-MoE) no independent evidence
    purpose: To enable a single model to handle diverse lighting conditions by routing wavelet-decomposed features to specialized experts
    Introduced in the paper as the key architectural addition for generalization; no independent evidence outside this work is provided.

pith-pipeline@v0.9.0 · 5524 in / 1144 out tokens · 30970 ms · 2026-05-10T11:39:12.233227+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

74 extracted references · 74 canonical work pages · 2 internal anchors

  1. [1]

    A high-quality denoising dataset for smartphone cameras

    Abdelrahman Abdelhamed, Stephen Lin, and Michael S Brown. A high-quality denoising dataset for smartphone cameras. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1692–1700, 2018. 1

    work page 2018

  2. [2]

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

    Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Radu Tim- ofte, and Yulun Zhang. Retinexformer: One-stage retinex- based transformer for low-light image enhancement. InPro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 12504–12513, 2023. 3, 5, 6

    work page 2023

  3. [3]

    Hinet: Half instance normalization network for image restoration

    Liangyu Chen, Xin Lu, Jie Zhang, Xiaojie Chu, and Cheng- peng Chen. Hinet: Half instance normalization network for image restoration. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 182–192, 2021. 5

    work page 2021

  4. [4]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. InEuropean confer- ence on computer vision, pages 17–33. Springer, 2022. 1, 3, 4, 5, 6

    work page 2022

  5. [5]

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

    Xiangyu Chen, Zheyuan Li, Yuandong Pu, Yihao Liu, Jiantao Zhou, Yu Qiao, and Chao Dong. A comparative study of image restoration networks for general backbone network design. InEuropean Conference on Computer Vision, pages 74–91. Springer, 2024. 4

    work page 2024

  6. [6]

    Learning contin- uous image representation with local implicit image function

    Yinbo Chen, Sifei Liu, and Xiaolong Wang. Learning contin- uous image representation with local implicit image function. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8628–8638, 2021. 1

    work page 2021

  7. [7]

    Nafssr: Stereo image super-resolution using nafnet

    Xiaojie Chu, Liangyu Chen, and Wenqing Yu. Nafssr: Stereo image super-resolution using nafnet. InProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, pages 1239–1248, 2022

    work page 2022

  8. [8]

    Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration

    Marcos V Conde, Ui-Jin Choi, Maxime Burchi, and Radu Timofte. Swin2sr: Swinv2 transformer for compressed image super-resolution and restoration. InEuropean conference on computer vision, pages 669–687. Springer, 2022. 1

    work page 2022

  9. [9]

    Instruc- tir: High-quality image restoration following human instruc- tions

    Marcos V Conde, Gregor Geigle, and Radu Timofte. Instruc- tir: High-quality image restoration following human instruc- tions. InEuropean Conference on Computer Vision, pages 1–21. Springer, 2024. 3

    work page 2024

  10. [10]

    Selective frequency network for image restoration

    Yuning Cui, Yi Tao, Zhenshan Bing, Wenqi Ren, Xinwei Gao, Xiaochun Cao, Kai Huang, and Alois Knoll. Selective frequency network for image restoration. InThe eleventh international conference on learning representations, 2023. 5, 6

    work page 2023

  11. [11]

    Towards ghost-free shadow removal via dual hierarchical aggrega- tion network and shadow matting gan

    Xiaodong Cun, Chi-Man Pun, and Cheng Shi. Towards ghost-free shadow removal via dual hierarchical aggrega- tion network and shadow matting gan. InProceedings of the AAAI conference on artificial intelligence, pages 10680– 10687, 2020. 1, 2, 5, 6

    work page 2020

  12. [12]

    Shadowrefiner: Towards mask-free shadow removal via fast fourier transformer

    Wei Dong, Han Zhou, Yuqiong Tian, Jingke Sun, Xiaohong Liu, Guangtao Zhai, and Jun Chen. Shadowrefiner: Towards mask-free shadow removal via fast fourier transformer. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 6208–6217, 2024. 1, 5, 6

    work page 2024

  13. [13]

    Uniprocessor: a text-induced unified low- level image processor

    Huiyu Duan, Xiongkuo Min, Sijing Wu, Wei Shen, and Guangtao Zhai. Uniprocessor: a text-induced unified low- level image processor. InEuropean Conference on Computer Vision, pages 180–199. Springer, 2024. 3

    work page 2024

  14. [14]

    Auto- exposure fusion for single-image shadow removal

    Lan Fu, Changqing Zhou, Qing Guo, Felix Juefei-Xu, Hongkai Yu, Wei Feng, Yang Liu, and Song Wang. Auto- exposure fusion for single-image shadow removal. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10571–10580, 2021. 2, 5, 6

    work page 2021

  15. [15]

    Prompt-based ingredient-oriented all-in- one image restoration.IEEE Transactions on Circuits and Systems for Video Technology, 34(10):9458–9471, 2024

    Hu Gao, Jing Yang, Ying Zhang, Ning Wang, Jingfan Yang, and Depeng Dang. Prompt-based ingredient-oriented all-in- one image restoration.IEEE Transactions on Circuits and Systems for Video Technology, 34(10):9458–9471, 2024. 3

    work page 2024

  16. [16]

    Pureformer: Transformer-based image denoising

    Arnim Gautam, Aditi Pawar, Aishwarya Joshi, Satya Narayan Tazi, Sachin Chaudhary, Praful Hambarde, Akshay Dudhane, Santosh Vipparthi, and Subrahamanyam Murala. Pureformer: Transformer-based image denoising. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1441–1449, 2025. 4

    work page 2025

  17. [17]

    Mambair: A simple baseline for image restoration with state-space model

    Hang Guo, Jinmin Li, Tao Dai, Zhihao Ouyang, Xudong Ren, and Shu-Tao Xia. Mambair: A simple baseline for image restoration with state-space model. InEuropean conference on computer vision, pages 222–241. Springer, 2024. 3, 5, 6

    work page 2024

  18. [18]

    Mambairv2: Attentive state space restoration

    Hang Guo, Yong Guo, Yaohua Zha, Yulun Zhang, Wenbo Li, Tao Dai, Shu-Tao Xia, and Yawei Li. Mambairv2: Attentive state space restoration. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 28124–28133,

    work page

  19. [19]

    Shadowformer: global context helps shadow removal

    Lanqing Guo, Siyu Huang, Ding Liu, Hao Cheng, and Bihan Wen. Shadowformer: global context helps shadow removal. InProceedings of the AAAI conference on artificial intelli- gence, pages 710–718, 2023. 1, 2, 5, 6

    work page 2023

  20. [20]

    Shadowdiffusion: When degradation prior meets diffusion model for shadow re- moval

    Lanqing Guo, Chong Wang, Wenhan Yang, Siyu Huang, Yufei Wang, Hanspeter Pfister, and Bihan Wen. Shadowdiffusion: When degradation prior meets diffusion model for shadow re- moval. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 14049–14058,

    work page

  21. [21]

    Boundary-aware divide and conquer: A diffusion- based solution for unsupervised shadow removal

    Lanqing Guo, Chong Wang, Wenhan Yang, Yufei Wang, and Bihan Wen. Boundary-aware divide and conquer: A diffusion- based solution for unsupervised shadow removal. InProceed- ings of the IEEE/CVF International Conference on Computer Vision, pages 13045–13054, 2023. 1

    work page 2023

  22. [22]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bern- hard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017. 6

    work page 2017

  23. [23]

    Direction-aware spatial context features for shadow detection

    Xiaowei Hu, Lei Zhu, Chi-Wing Fu, Jing Qin, and Pheng-Ann Heng. Direction-aware spatial context features for shadow detection. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7454–7462, 2018. 2

    work page 2018

  24. [24]

    Mask-shadowgan: Learning to remove shadows from unpaired data

    Xiaowei Hu, Yitong Jiang, Chi-Wing Fu, and Pheng-Ann Heng. Mask-shadowgan: Learning to remove shadows from unpaired data. InProceedings of the IEEE/CVF international conference on computer vision, pages 2472–2481, 2019. 1

    work page 2019

  25. [25]

    Dc- shadownet: Single-image hard and soft shadow removal using unsupervised domain-classifier guided network

    Yeying Jin, Aashish Sharma, and Robby T Tan. Dc- shadownet: Single-image hard and soft shadow removal using unsupervised domain-classifier guided network. InProceed- ings of the IEEE/CVF international conference on computer vision, pages 5027–5036, 2021

    work page 2021

  26. [26]

    Des3: Adaptive attention-driven self and soft shadow removal using vit similarity

    Yeying Jin, Wei Ye, Wenhan Yang, Yuan Yuan, and Robby T Tan. Des3: Adaptive attention-driven self and soft shadow removal using vit similarity. InProceedings of the AAAI Conference on Artificial Intelligence, pages 2634–2642, 2024. 1

    work page 2024

  27. [27]

    Idf: Iterative dynamic filtering networks for generalizable image denoising

    Dongjin Kim, Jaekyun Ko, Muhammad Kashif Ali, and Tae Hyun Kim. Idf: Iterative dynamic filtering networks for generalizable image denoising. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12180–12190, 2025. 1

    work page 2025

  28. [28]

    Transfer learning from synthetic to real-noise denois- ing with adaptive instance normalization

    Yoonsik Kim, Jae Woong Soh, Gu Yong Park, and Nam Ik Cho. Transfer learning from synthetic to real-noise denois- ing with adaptive instance normalization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3482–3492, 2020. 1

    work page 2020

  29. [29]

    Shadow removal via shadow image decomposition

    Hieu Le and Dimitris Samaras. Shadow removal via shadow image decomposition. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 8578–8587,

    work page

  30. [30]

    From shadow segmentation to shadow removal

    Hieu Le and Dimitris Samaras. From shadow segmentation to shadow removal. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XI 16, pages 264–281. Springer, 2020. 1

    work page 2020

  31. [31]

    All-in-one image restoration for unknown corruption

    Boyun Li, Xiao Liu, Peng Hu, Zhongqin Wu, Jiancheng Lv, and Xi Peng. All-in-one image restoration for unknown corruption. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17452–17462,

    work page

  32. [32]

    Efficient and explicit modelling of image hierarchies for image restora- tion

    Yawei Li, Yuchen Fan, Xiaoyu Xiang, Denis Demandolx, Rakesh Ranjan, Radu Timofte, and Luc Van Gool. Efficient and explicit modelling of image hierarchies for image restora- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 18278–18289,

    work page

  33. [33]

    Shadowmaskformer: Mask augmented patch embedding for shadow removal.IEEE Transactions on Artificial Intelligence,

    Zhuohao Li, Guoyang Xie, Guannan Jiang, and Zhichao Lu. Shadowmaskformer: Mask augmented patch embedding for shadow removal.IEEE Transactions on Artificial Intelligence,

    work page

  34. [34]

    Swinir: Image restoration using swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. InProceedings of the IEEE/CVF international conference on computer vision, pages 1833– 1844, 2021. 1, 3, 5, 6

    work page 2021

  35. [35]

    Regional atten- tion for shadow removal

    Hengxing Liu, Mingjia Li, and Xiaojie Guo. Regional atten- tion for shadow removal. InProceedings of the 32nd ACM International Conference on Multimedia, pages 5949–5957,

    work page

  36. [36]

    Shadow removal by a lightness-guided network with training on unpaired data.IEEE Transactions on Image Processing, 30:1853–1865, 2021

    Zhihao Liu, Hui Yin, Yang Mi, Mengyang Pu, and Song Wang. Shadow removal by a lightness-guided network with training on unpaired data.IEEE Transactions on Image Processing, 30:1853–1865, 2021. 1

    work page 2021

  37. [37]

    SGDR: Stochastic Gradient Descent with Warm Restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts.arXiv preprint arXiv:1608.03983,

    work page Pith review arXiv

  38. [38]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  39. [39]

    Latent feature-guided diffusion models for shadow removal

    Kangfu Mei, Luis Figueroa, Zhe Lin, Zhihong Ding, Scott Cohen, and Vishal M Patel. Latent feature-guided diffusion models for shadow removal. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 4313–4322, 2024. 1

    work page 2024

  40. [40]

    Dinolight: Robust ambient light normalization with self-supervised vi- sual prior integration.arXiv preprint arXiv:2603.12579, 2026

    Youngjin Oh, Junhyeong Kwon, and Nam Ik Cho. Dinolight: Robust ambient light normalization with self-supervised vi- sual prior integration.arXiv preprint arXiv:2603.12579, 2026. 1, 2, 3, 4, 5, 6, 7

    work page arXiv 2026

  41. [41]

    DINOv2: Learning Robust Visual Features without Supervision

    Maxime Oquab, Timoth´ee Darcet, Th´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193, 2023. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  42. [42]

    All- in-one image restoration for unknown degradations using adaptive discriminative filters for specific degradations

    Dongwon Park, Byung Hyun Lee, and Se Young Chun. All- in-one image restoration for unknown degradations using adaptive discriminative filters for specific degradations. In 2023 IEEE/CVF conference on computer vision and pattern recognition (CVPR), pages 5815–5824. IEEE, 2023. 3

    work page 2023

  43. [43]

    Benchmarking denoising al- gorithms with real photographs

    Tobias Plotz and Stefan Roth. Benchmarking denoising al- gorithms with real photographs. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1586–1595, 2017. 1

    work page 2017

  44. [44]

    Promptir: Prompting for all-in-one image restoration

    Vaishnav Potlapalli, Syed Waqas Zamir, Salman Khan, and Fahad Khan. Promptir: Prompting for all-in-one image restoration. InThirty-seventh Conference on Neural Informa- tion Processing Systems, 2023. 3

    work page 2023

  45. [45]

    Deshadownet: A multi-context embed- ding deep network for shadow removal

    Liangqiong Qu, Jiandong Tian, Shengfeng He, Yandong Tang, and Rynson WH Lau. Deshadownet: A multi-context embed- ding deep network for shadow removal. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4067–4075, 2017. 1, 2

    work page 2017

  46. [46]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InInternational Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer,

    work page

  47. [47]

    Promptnorm: Image geometry guides ambient light normalization

    David Serrano-Lozano, Francisco A Molina-Bakhos, Danna Xue, Yixiong Yang, Maria Pilligua, Ramon Baldrich, Maria Vanrell, and Javier Vazquez-Corral. Promptnorm: Image geometry guides ambient light normalization. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 905–916, 2025. 3, 4, 5, 6

    work page 2025

  48. [48]

    Vmambair: Visual state space model for image restoration.IEEE Transactions on Circuits and Systems for Video Technology, 2025

    Yuan Shi, Bin Xia, Xiaoyu Jin, Xing Wang, Tianyu Zhao, Xin Xia, Xuefeng Xiao, and Wenming Yang. Vmambair: Visual state space model for image restoration.IEEE Transactions on Circuits and Systems for Video Technology, 2025. 3

    work page 2025

  49. [49]

    Meta-transfer learning for zero-shot super-resolution

    Jae Woong Soh, Sunwoo Cho, and Nam Ik Cho. Meta-transfer learning for zero-shot super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3516–3525, 2020. 1

    work page 2020

  50. [50]

    Transweather: Transformer-based restoration of images degraded by adverse weather conditions

    Jeya Maria Jose Valanarasu, Rajeev Yasarla, and Vishal M Patel. Transweather: Transformer-based restoration of images degraded by adverse weather conditions. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2353–2363, 2022. 3

    work page 2022

  51. [51]

    Visualizing data using t-sne.Journal of machine learning research, 9(11),

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne.Journal of machine learning research, 9(11),

    work page

  52. [52]

    Shadow removal with paired and unpaired learning

    Florin-Alexandru Vasluianu, Andr´es Romero, Luc Van Gool, and Radu Timofte. Shadow removal with paired and unpaired learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 826–835,

    work page

  53. [53]

    Wsrd: A novel benchmark for high resolution image shadow removal

    Florin-Alexandru Vasluianu, Tim Seizinger, and Radu Tim- ofte. Wsrd: A novel benchmark for high resolution image shadow removal. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pages 1826–1835, 2023. 1, 2, 5, 6

    work page 2023

  54. [54]

    Towards image ambient lighting normalization

    Florin-Alexandru Vasluianu, Tim Seizinger, Zongwei Wu, Rakesh Ranjan, and Radu Timofte. Towards image ambient lighting normalization. InEuropean Conference on Computer Vision, pages 385–404. Springer, 2024. 1, 2, 5, 6

    work page 2024

  55. [55]

    Ntire 2024 image shadow removal challenge report

    Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Zongwei Wu, Cailian Chen, Radu Timofte, Wei Dong, Han Zhou, Yuqiong Tian, Jun Chen, et al. Ntire 2024 image shadow removal challenge report. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6547–6570, 2024. 1, 2

    work page 2024

  56. [56]

    After the party: Navigating the map- ping from color to ambient lighting

    Florin-Alexandru Vasluianu, Tim Seizinger, Zongwei Wu, and Radu Timofte. After the party: Navigating the map- ping from color to ambient lighting. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 9218–9229, 2025. 1, 2, 3, 5, 6

    work page 2025

  57. [57]

    Ntire 2025 ambient lighting normalization challenge report

    Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Zongwei Wu, Radu Timofte, Yuanfei Bao, Xingbo Wang, Xin Lu, Jiarong Yang, Anya Hu, et al. Ntire 2025 ambient lighting normalization challenge report. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 1289–1300, 2025. 2

    work page 2025

  58. [58]

    Learning- Based Ambient Lighting Normalization: NTIRE 2026 Chal- lenge Results and Findings

    Florin-Alexandru Vasluianu, Tim Seizinger, Jeffrey Chen, Zhuyun Zhou, Zongwei Wu, Radu Timofte, et al. Learning- Based Ambient Lighting Normalization: NTIRE 2026 Chal- lenge Results and Findings . InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2026. 1, 2, 6, 7

    work page 2026

  59. [59]

    Advances in Single-Image Shadow Removal: Results from the NTIRE 2026 Challenge

    Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Zongwei Wu, Radu Timofte, et al. Advances in Single-Image Shadow Removal: Results from the NTIRE 2026 Challenge . InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2026. 1, 2, 6, 7

    work page 2026

  60. [60]

    Style-guided shadow removal

    Jin Wan, Hui Yin, Zhenyao Wu, Xinyi Wu, Yanting Liu, and Song Wang. Style-guided shadow removal. InEuropean Conference on Computer Vision, pages 361–378. Springer,

    work page

  61. [61]

    Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal

    Jifeng Wang, Xiang Li, and Jian Yang. Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1788–1797, 2018. 1, 2

    work page 2018

  62. [62]

    Re- covering realistic texture in image super-resolution by deep spatial feature transform

    Xintao Wang, Ke Yu, Chao Dong, and Chen Change Loy. Re- covering realistic texture in image super-resolution by deep spatial feature transform. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 606–615, 2018. 1, 4

    work page 2018

  63. [63]

    Multiscale structural similarity for image quality assessment

    Zhou Wang, Eero P Simoncelli, and Alan C Bovik. Multiscale structural similarity for image quality assessment. InThe Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003, pages 1398–1402. Ieee, 2003. 5

    work page 2003

  64. [64]

    Uformer: A general u-shaped transformer for image restoration

    Zhendong Wang, Xiaodong Cun, Jianmin Bao, Wengang Zhou, Jianzhuang Liu, and Houqiang Li. Uformer: A general u-shaped transformer for image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17683–17693, 2022. 5, 6

    work page 2022

  65. [65]

    Homoformer: Homogenized transformer for image shadow removal

    Jie Xiao, Xueyang Fu, Yurui Zhu, Dong Li, Jie Huang, Kai Zhu, and Zheng-Jun Zha. Homoformer: Homogenized transformer for image shadow removal. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25617–25626, 2024. 1, 2

    work page 2024

  66. [66]

    Omnisr: Shadow removal under direct and indirect lighting

    Jiamin Xu, Zelong Li, Yuxin Zheng, Chenyu Huang, Renshu Gu, Weiwei Xu, and Gang Xu. Omnisr: Shadow removal under direct and indirect lighting. InProceedings of the AAAI Conference on Artificial Intelligence, pages 8887–8895, 2025. 1, 3, 5, 6

    work page 2025

  67. [67]

    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. 1

    work page 2024

  68. [68]

    Com- plexity experts are task-discriminative learners for any image restoration

    Eduard Zamfir, Zongwei Wu, Nancy Mehta, Yuedong Tan, Danda Pani Paudel, Yulun Zhang, and Radu Timofte. Com- plexity experts are task-discriminative learners for any image restoration. InProceedings of the Computer Vision and Pat- tern Recognition Conference, pages 12753–12763, 2025. 3, 4, 5

    work page 2025

  69. [69]

    Multi-stage progressive image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao. Multi-stage progressive image restoration. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14821–14831, 2021. 5, 6

    work page 2021

  70. [70]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5728–5739,

    work page

  71. [71]

    All-in-one multi-degradation image restoration net- work via hierarchical degradation representation

    Cheng Zhang, Yu Zhu, Qingsen Yan, Jinqiu Sun, and Yanning Zhang. All-in-one multi-degradation image restoration net- work via hierarchical degradation representation. InProceed- ings of the 31st ACM international conference on multimedia, pages 2285–2293, 2023. 3

    work page 2023

  72. [72]

    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. 6

    work page 2018

  73. [73]

    Spa- former: Transformer image shadow detection and removal via spatial attention.arXiv preprint arXiv:2206.10910, 2022

    Xiao Feng Zhang, Chao Chen Gu, and Shan Ying Zhu. Spa- former: Transformer image shadow detection and removal via spatial attention.arXiv preprint arXiv:2206.10910, 2022. 1

    work page arXiv 2022

  74. [74]

    Bijective mapping network for shadow removal

    Yurui Zhu, Jie Huang, Xueyang Fu, Feng Zhao, Qibin Sun, and Zheng-Jun Zha. Bijective mapping network for shadow removal. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5627–5636,

    work page