Pith. sign in

REVIEW 4 major objections 6 minor 52 references

NTIRE 2025 the 2nd Restore Any Image Model (RAIM) in the Wild Challenge

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The 2nd RAIM challenge reports real-world image restoration gains over prior state of the art.

desk verdict Useful benchmark and method summaries, but the final-score formula does not reproduce the reported tables, so the SOTA claim is unverifiable as written. read the letter →

arxiv 2506.01394 v1 pith:NLZD3JWR submitted 2025-06-02 eess.IV cs.CV

Jie Liang , Radu Timofte , Qiaosi Yi , Zhengqiang Zhang , Shuaizheng Liu , Lingchen Sun , Rongyuan Wu , Xindong Zhang
show 2 more authors
Hui Zeng Lei Zhang
This is my paper
classification eess.IVcs.CV MSC 68T4568U1094A08
keywords imagerestorationbenchmarklow-lightjointdenoisinganddemosaicingdetailenhancementgenerationreal-worlddegradationspairedunpairedevaluationuserstudyefficiencyscoringsmartphoneISP
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper reports the second Restore Any Image Model (RAIM) challenge, a benchmark designed to test image-restoration models on real-world photos with complex, unknown degradations. The challenge ran two tracks — low-light joint denoising and demosaicing, and image detail enhancement/generation — each with paired data for quantitative scoring and unpaired real captures for subjective judging. It attracted nearly 300 registrations, 51 teams, and more than 600 submitted results. The paper claims the top-performing methods advanced the state of the art in image restoration, with all 20 or more expert judges agreeing on the best outputs. If the benchmark's simulations faithfully represent real smartphone captures, the challenge provides a reusable public testbed for comparing restoration quality and efficiency.

What carries the argument

The load-bearing mechanism is the challenge protocol itself: calibrated noise and PSF simulation feeding raw images through a real camera ISP to create paired reference data, unpaired real captures for subjective testing, a weighted composite metric (PSNR, SSIM, LPIPS, DISTS, NIQE) plus an efficiency score, and a final score $S_{final} = 0.4 \, S_2 + 0.6 \, S_n^3$ that blends the phase-2 quantitative score with the normalized phase-3 user-study score. This protocol is what lets the organizers claim that the top methods generalize to real captures rather than only to synthetic degradations.

What would settle it

A direct check would be to correlate each team's phase-2 composite score on the paired data with its phase-3 user-study ranking on the unpaired real captures; if the correlation is weak or negative, or if a phase-2 winner finishes near the bottom of the user study, then the simulated paired data are not a valid proxy for real-world restoration quality. A second check would compare the noise statistics (e.g., photon-transfer curve residuals) and measured PSF of the simulated inputs against the actual unpaired captures to quantify the domain gap.

Watch

Extended reading notes

Core claim

The central claim is that the 2nd RAIM challenge establishes a workable public benchmark for real-world image restoration and that the winning methods outperform previous approaches on both objective and subjective grounds. Paired data are synthesized by calibrating sensor noise and lens point-spread functions, applying degradations to raw images, and pushing them through a real camera ISP, while unpaired test images are true single-frame smartphone captures. Quantitative screening uses a weighted composite of PSNR, SSIM, LPIPS, DISTS, and NIQE; the top ten teams per track then enter a user study in which 20 judges rank anonymized outputs on the unpaired images, and the final score combines 40% of the phase-2 quantitative score with 60% of the normalized user-study score, alongside an efficiency score based on parameter count, FLOPs, and runtime.

Load-bearing premise

The paired evaluation data are generated by simulating sensor noise and lens blur on raw images and running them through a real ISP, and the whole screening procedure depends on these simulated pairs behaving like the real smartphone captures used in the subjective test.

Editorial extensions

If this is right

  • If the benchmark is faithful, future restoration models can be compared under a single protocol that mixes fidelity, perceptual quality, and efficiency on real-captured images.
  • The two-track design forces methods to handle raw low-light input and sRGB detail generation separately, so advances in each pipeline stage can be isolated.
  • The public datasets and the established scoring formula give a reproducible target for both academic and industrial teams targeting smartphone-class ISPs.
  • The efficiency scoring (parameters, FLOPs, runtime) makes deployment cost part of the evaluation, encouraging lightweight models alongside quality.
  • The user-study component provides a subjective check that objective scores alone would miss, so reported SOTA gains have a perceptual backing.

Reading between the lines

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

  • A testable extension: compute the rank correlation between phase-2 paired scores and phase-3 user-study scores across all submitted teams; a weak correlation would suggest the simulated paired data under-predict real-world perceptual preference.
  • The calibrated-PSF approach could be reused to build paired benchmarks for other smartphone imaging tasks such as burst HDR or telephoto super-resolution.
  • The challenge's top entries suggest a convergence on hybrid pipelines: transformer or attention backbones for global denoising followed by GAN-based local texture synthesis; future work could isolate which component contributes most to user preference.
  • Because teams were allowed any training data, the benchmark measures the whole pipeline (data, augmentation, architecture), not just the network; the reported SOTA advance may partly reflect better training data recipes rather than architectural novelty alone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. This paper reports the organization and results of the NTIRE 2025 second Restore Any Image Model (RAIM) in the Wild challenge. It describes two tracks (low-light raw joint denoising/demosaicing and RGB detail enhancement/generation), each with a paired-data phase for full-reference metrics and an unpaired phase evaluated by user study and efficiency. It reports participation counts, phase-2 quantitative scores, efficiency scores, user-study scores, and final ranks for 7 and 12 teams, and it summarizes the methods of the top six teams in each track. The central claim is that the top-performing methods advanced the state of the art in image restoration and received unanimous recognition from more than 20 expert judges.

Significance. If the challenge results are reproducible and the subjective protocol is sound, the paper provides a useful public benchmark for real-world image restoration, including paired and unpaired tracks, efficiency-aware ranking, and submissions from both industrial and academic teams. The release of datasets and the use of standard full-reference metrics are valuable contributions. However, the final-score computation cannot be verified from the equations and tables as written, and the user-study description is internally inconsistent; these issues currently block confidence in the ranking claims that support the state-of-the-art statement. The paper does not attempt any derivation from the reported data, so circularity is not a concern.

major comments (4)
  1. [Section 3.2, Eq. (1), Tables 1–2] The final scores in Tables 1 and 2 are not reproducible from the stated formula. For Track 1, MiAlgo has 0.4*100.89 + 0.6*106 = 103.96, but Table 1 reports a final score of 112.18; adding the EScore of 15 gives 118.96, which still does not match. For Track 2, NulltoZero has 0.4*106.44 + 0.6*27.45 = 59.05, but Table 2 reports 113.26; adding the EScore of 20.9 gives 79.95. These discrepancies are far larger than rounding error. Because the final ranks are the primary quantitative support for the abstract's state-of-the-art claim, the authors must either correct Eq. (1) to the formula actually used (including how EScore enters, if it does) or recompute the final-score columns in both tables. They should also state explicitly whether the 'UScore in P3' column is the raw user-study score or the normalized score S^n_3.
  2. [Section 2.3.2 vs. Section 3.2 vs. Abstract] The number and role of human judges are described inconsistently: Section 2.3.2 says 10 experienced practitioners, Section 3.2 says 20 low-level vision-related students/engineers, and the abstract says 20+ expert judges. Additionally, the protocol in which each scorer selects a top three for each sample cannot by itself support the abstract's claim of 'unanimous recognition from all 20+ expert judges'; unanimous top-three membership is not the same as unanimous agreement on a single winner. Please report per-scorer counts or an inter-rater agreement statistic, and either support or soften the unanimity claim.
  3. [Section 2.4.3 vs. Section 3 vs. Eq. (1)] Section 2.4.3 states that final ranks are decided from the phase-2 quantitative scores, the efficiency of the model, and the subjective user-study results, and the tables include an EScore column; however, Eq. (1) omits EScore entirely. If efficiency contributes to the final ranking, the formula must say so; if it does not, the EScore column and the phase-3 text are misleading. The selection criterion is also inconsistent: Section 2.4.3 says the top ten teams by phase-2 quantitative score are selected, whereas Section 3 reports 7 and 12 teams invited in Tracks 1 and 2, respectively. These inconsistencies must be reconciled.
  4. [Sections 2.2.1 and 2.4] The paired data used for phase-2 quantitative screening are generated by simulating calibrated sensor noise and lens PSF on raw images and processing them through a real ISP, while the phase-3 unpaired data are real smartphone captures. Because phase-2 scores determine which teams advance to the subjective user study, the real-world validity of the rankings depends on how well the simulation matches the actual test distribution. Please provide a quantitative comparison (for example, noise and PSF statistics, or metric distributions on held-out real captures) or explicitly state this as a limitation; without such evidence, the selection into the user study may not reflect performance on the real unpaired data.
minor comments (6)
  1. [Section 7.2.1] The text says the model has approximately 0.303MB of parameters; this should be 0.303M parameters, since MB is a storage unit and not a parameter count.
  2. [Table 2 and Section 6] The team name is written as 'TACOSR' in Table 2 but as 'TACO SR' in Section 6, and 'W ANGTY' contains an apparent spacing error; please use consistent team names throughout.
  3. [Section 7.1.4] There are two consecutive paragraphs headed 'Training Details' for team TongJi-IPOE, and the second one mainly contains inference and efficiency details; please reorganize the subsection so that the headings match the content.
  4. [References] References [41] and [42] both cite the same Real-ESRGAN paper, references [20] and [52] cite the same perceptual-loss paper, and references [5] and [6] duplicate the NAFNet citation; these should be consolidated or clearly distinguished.
  5. [Figures 3 and 4] The captions say 'top 6 participating teams' but Tables 1 and 2 list 7 and 12 teams in phase 3, respectively; please specify which teams are shown and why those were selected.
  6. [Section 2.3.1] The composite score formula uses fixed caps (P SNR/50, SSIM/0.5, LPIPS/0.4, DISTS/0.3, NIQE/10) without motivation or citation; please cite the source of this measure or briefly justify the chosen weights and caps.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the challenge report contains no derivation chain; its score formulas are explicit evaluator definitions, and the reported rankings are measurements rather than fitted predictions.

full rationale

This paper is an organizational challenge report, and it does not claim to derive any quantity from the data it reports. The only formulas are the organizers' evaluation definitions: the SCORE formula in Section 2.3.1 (a fixed linear combination of PSNR, SSIM, LPIPS, DISTS, NIQE), the efficiency scoring in Section 2.3.3 (rank-based points), and the final-score rule in Section 3.2, Eq. (1) with Eq. (2), a weighted average of the Phase 2 quantitative score and a min-max normalized user-study score. These are definitions chosen by the organizers, not quantities fitted to a subset of the data and then predicted back. No parameter is learned from the reported scores, so no 'prediction' reduces to its own input by construction. The abstract's 'state of the art' and 'unanimous recognition' claims rest on measured metric values and user-study votes, not on a self-referential derivation. Self-citations ([30] for the previous RAIM challenge and [31] for this challenge itself) are used only to identify prior and current challenge data and the challenge listing; they do not supply an unverified load-bearing premise such as a uniqueness theorem or an adopted ansatz. A reviewer-facing caveat that is a correctness concern rather than circularity: the Final Score columns in Tables 1-2 do not reproduce Eq. (1) as written (e.g., Track 1 MiAlgo: 0.4x100.89 + 0.6x106 = 103.96 vs. reported 112.18; Track 2 NulltoZero: 0.4x106.44 + 0.6x27.45 = 59.05 vs. reported 113.26), and the number of judges is stated inconsistently (10 practitioners in Section 2.3.2, 20 students/engineers in Section 3.2, '20+ expert judges' in the abstract). Per the rubric, a non-reproducible scoring table and inconsistent participant counts are accounting and reporting issues, not reductions of a claim to its inputs, so they do not raise the circularity score.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The challenge report rests on domain assumptions about the realism of simulated paired data and the reliability of the user study, plus an arbitrary weighting scheme for the quantitative score. No new entities are introduced, and there is no derivation to audit.

free parameters (1)
  • Evaluation metric weights and caps = PSNR 20/50, SSIM 15/0.5, LPIPS 20/0.4, DISTS 40/0.3, NIQE 30/10; final weights 0.4/0.6
    Hand-chosen weights in the challenge scoring formula (Section 2.3.1) and final ranking (Eq. 1). They determine the rankings and are not justified or fitted to external data.
assumptions (3)
  • domain assumption The simulated paired data, based on calibrated noise and PSF, are representative of real smartphone degradations.
    Section 2.2.1 uses simulated degradations to create paired data. If the simulation is unrepresentative, phase-2 quantitative rankings may not transfer to real-world captures.
  • domain assumption The subjective user study scores are a reliable measure of restoration quality.
    Section 2.3.2 and 3.2 use 10 to 20 practitioners and students to rank outputs without statistical validation. The paper states 'we found their opinions are similar' but provides no inter-rater agreement measure.
  • domain assumption The weighted metric formula is a valid proxy for image restoration quality.
    Section 2.3.1 defines the score with arbitrary weights and caps; the resulting rankings depend on these choices.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NTIRE 2025 the 2nd Restore Any Image Model (RAIM) in the Wild Challenge." pith.science (2026). https://pith.science/paper/NLZD3JWR

@misc{pith2026250601394,
  author       = {Pith},
  title        = {Pith review of: NTIRE 2025 the 2nd Restore Any Image Model (RAIM) in the Wild Challenge},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NLZD3JWR}},
  note         = {Machine review of arXiv:2506.01394}
}
read the original abstract

In this paper, we present a comprehensive overview of the NTIRE 2025 challenge on the 2nd Restore Any Image Model (RAIM) in the Wild. This challenge established a new benchmark for real-world image restoration, featuring diverse scenarios with and without reference ground truth. Participants were tasked with restoring real-captured images suffering from complex and unknown degradations, where both perceptual quality and fidelity were critically evaluated. The challenge comprised two tracks: (1) the low-light joint denoising and demosaicing (JDD) task, and (2) the image detail enhancement/generation task. Each track included two sub-tasks. The first sub-task involved paired data with available ground truth, enabling quantitative evaluation. The second sub-task dealt with real-world yet unpaired images, emphasizing restoration efficiency and subjective quality assessed through a comprehensive user study. In total, the challenge attracted nearly 300 registrations, with 51 teams submitting more than 600 results. The top-performing methods advanced the state of the art in image restoration and received unanimous recognition from all 20+ expert judges. The datasets used in Track 1 and Track 2 are available at https://drive.google.com/drive/folders/1Mgqve-yNcE26IIieI8lMIf-25VvZRs_J and https://drive.google.com/drive/folders/1UB7nnzLwqDZOwDmD9aT8J0KVg2ag4Qae, respectively. The official challenge pages for Track 1 and Track 2 can be found at https://codalab.lisn.upsaclay.fr/competitions/21334#learn_the_details and https://codalab.lisn.upsaclay.fr/competitions/21623#learn_the_details.

Figures

Figures reproduced from arXiv: 2506.01394 by the authors.

Figure 1
Figure 1. Example Paired data for each track we have provided. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Example Unpaired data for each track we have provided. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visual comparisons between the input LQ image and the results from the top 6 participating teams in Track 1. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Visual comparisons between the input LQ image and the results from the top 6 participating teams in Track 2. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: The two-stage pipeline proposed by team MiAlgo. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 7
Figure 7. Figure 7: Overview of the Team TongJi-IPOE. 7.1.4 Team TongJi-IPOE Team TongJi-IPOE proposed a lightweight and efficient joint raw image denoising and demosaicing solution, named B2FNet (Branching to Fusion Network), designed for low-light image processing. Inspired by the low-l…
Figure 6
Figure 6. Figure 6: Pipeline of Team POLYU-AISP. LN: Layer Normaliza [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 8
Figure 8. Figure 8: Frame diagram of our method. GFLOPs for an input size of 1024 × 1024. Average inference time is 8.3 ms on an NVIDIA A100 GPU. Training Details. The method is implemented in PyTorch and runs on a single RTX 3090Ti GPU. The pipeline is simple and compact, and suitable fo…
Figure 10
Figure 10. Figure 10: The overall pipeline of the solution proposed by team [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Overview of the MSUnet proposed by Team TongJi [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 13
Figure 13. Figure 13: Overall framework proposed by iAM IR. (a) repre￾sents the U-Net architecture of NAFNet, (b) represents the original NAFNet block, and (c) represents our improved NAFNet block [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Structure comparison between SimpleGate and the [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: The tiny unet in phase3 applied by the team MiAlgo. [PITH_FULL_IMAGE:figures/full_fig_p016_15.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 47 canonical work pages

  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,

    work page

  2. [2]

    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. 10, 11

    work page 2018

  3. [3]

    Real-world blind super-resolution via feature matching with implicit high-resolution priors

    Chaofeng Chen, Xinyu Shi, Yipeng Qin, Xiaoming Li, Xiaoguang Han, Tao Yang, and Shihui Guo. Real-world blind super-resolution via feature matching with implicit high-resolution priors. InProceedings of the 30th ACM International Conference on Multimedia, pages 1329–1338,

    work page

  4. [4]

    Hinet: Half instance normalization network for image restoration

    Liangyu Chen, Xin Lu, Jie Zhang, Xiaojie Chu, and Chengpeng Chen. Hinet: Half instance normalization network for image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 182–192, 2021. 10

    work page 2021

  5. [5]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. InEuropean conference on computer vision, pages 17–33. Springer,

    work page

  6. [6]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. InComputer Vision–ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part VII, pages 17–33. Springer, 2022. 10

    work page 2022

  7. [7]

    Activating more pixels in image super-resolution transformer

    Xiangyu Chen, Xintao Wang, Jiantao Zhou, Yu Qiao, and Chao Dong. Activating more pixels in image super-resolution transformer. In2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 22367–22377, 2023. 13

    work page 2023

  8. [8]

    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. 9

    work page 2024

Show all 52 references
  1. [9]

    NTIRE 2025 challenge on image super-resolution (×4): Methods and results

    Zheng Chen, Kai Liu, Jue Gong, Jingkai Wang, Lei Sun, Zongwei Wu, Radu Timofte, Yulun Zhang, et al. NTIRE 2025 challenge on image super-resolution (×4): Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  2. [10]

    NTIRE 2025 challenge on real-world face restoration: Methods and results

    Zheng Chen, Jingkai Wang, Kai Liu, Jue Gong, Lei Sun, Zongwei Wu, Radu Timofte, Yulun Zhang, et al. NTIRE 2025 challenge on real-world face restoration: Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  3. [11]

    NTIRE 2025 challenge on raw image restoration and super-resolution

    Marcos Conde, Radu Timofte, et al. NTIRE 2025 challenge on raw image restoration and super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  4. [12]

    Raw image reconstruction from RGB on smartphones

    Marcos Conde, Radu Timofte, et al. Raw image reconstruction from RGB on smartphones. NTIRE 2025 challenge report. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  5. [13]

    Image quality assessment: Unifying structure and texture similarity.IEEE transactions on pattern analysis and machine intelligence, 44(5):2567–2581, 2020

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P Simoncelli. Image quality assessment: Unifying structure and texture similarity.IEEE transactions on pattern analysis and machine intelligence, 44(5):2567–2581, 2020. 14

    2020

  6. [14]

    NTIRE 2025 challenge on night photography rendering

    Egor Ershov, Sergey Korchagin, Alexei Khalin, Artyom Panshin, Arseniy Terekhin, Ekaterina Zaychenkova, Georgiy Lobarev, Vsevolod Plokhotnyuk, Denis Abramov, Elisey Zhdanov, Sofia Dorogova, Yasin Mamedov, Nikola Banic, Georgii Perevozchikov, Radu Timofte, et al. NTIRE 2025 chal...

    2025

  7. [15]

    NTIRE 2025 challenge on cross-domain few-shot object detection: Methods and results

    Yuqian Fu, Xingyu Qiu, Bin Ren Yanwei Fu, Radu Timofte, Nicu Sebe, Ming-Hsuan Yang, Luc Van Gool, et al. NTIRE 2025 challenge on cross-domain few-shot object detection: Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CV...

    2025

  8. [16]

    Div8k: Diverse 8k resolution image dataset

    Shuhang Gu, Andreas Lugmayr, Martin Danelljan, Manuel Fritsche, Julien Lamour, and Radu Timofte. Div8k: Diverse 8k resolution image dataset. In2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), pages 3512–3516. IEEE, 2019. 15

    2019

  9. [17]

    NTIRE 2025 challenge on text to image generation model quality assessment

    Shuhao Han, Haotian Fan, Fangyuan Kong, Wenjie Liao, Chunle Guo, Chongyi Li, Radu Timofte, et al. NTIRE 2025 challenge on text to image generation model quality assessment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  10. [18]

    Aim 2020 challenge on learned image signal processing pipeline

    Andrey Ignatov, Radu Timofte, Zhilu Zhang, Ming Liu, Haolin Wang, Wangmeng Zuo, and et al. Aim 2020 challenge on learned image signal processing pipeline. In Computer Vision–ECCV 2020 Workshops: Glasgow, UK, August 23–28, 2020, Proceedings, Part III 16, pages 152–170, 2020. 9, 14, 16

    2020

  11. [19]

    NTIRE 2025 challenge on video quality enhancement for video conferencing: Datasets, methods and results

    Varun Jain, Zongwei Wu, Quan Zou, Louis Florentin, Henrik Turbell, Sandeep Siddhartha, Radu Timofte, et al. NTIRE 2025 challenge on video quality enhancement for video conferencing: Datasets, methods and results. In Proceedings of the IEEE/CVF Conference on Computer Vision and...

    2025

  12. [20]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14, pages 694–711. Springer, 2016. 9

    2016

  13. [21]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410, 2019. 15

    2019

  14. [22]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    arXiv

  15. [23]

    Artifact generation when using perceptual loss for image deblurring.Authorea Preprints, 2023

    Patrick Krawczyk, Marvin Gaertner, Andreas Jansche, Timo Bernthaler, and Gerhard Schneider. Artifact generation when using perceptual loss for image deblurring.Authorea Preprints, 2023. 9

    2023

  16. [24]

    Fast and accurate image super-resolution with deep laplacian pyramid networks

    Wei-Sheng Lai, Jia-Bin Huang, Narendra Ahuja, and Ming-Hsuan Yang. Fast and accurate image super-resolution with deep laplacian pyramid networks. IEEE transactions on pattern analysis and machine intelligence, 41(11):2599–2613, 2018. 9

    2018

  17. [25]

    NTIRE 2025 challenge on efficient burst hdr and restoration: Datasets, methods, and results

    Sangmin Lee, Eunpil Park, Angel Canelo, Hyunhee Park, Youngjo Kim, Hyungju Chun, Xin Jin, Chongyi Li, Chun-Le Guo, Radu Timofte, et al. NTIRE 2025 challenge on efficient burst hdr and restoration: Datasets, methods, and results. InProceedings of the IEEE/CVF Conference on Comp...

    2025

  18. [26]

    NTIRE 2025 challenge on day and night raindrop removal for dual-focused images: Methods and results

    Xin Li, Yeying Jin, Xin Jin, Zongwei Wu, Bingchen Li, Yufei Wang, Wenhan Yang, Yu Li, Zhibo Chen, Bihan Wen, Robby Tan, Radu Timofte, et al. NTIRE 2025 challenge on day and night raindrop removal for dual-focused images: Methods and results. InProceedings of the IEEE/CVF Confe...

    2025

  19. [27]

    NTIRE 2025 challenge on short-form ugc video quality assessment and enhancement: Kwaisr dataset and study

    Xin Li, Xijun Wang, Bingchen Li, Kun Yuan, Yizhen Shao, Suhang Yao, Ming Sun, Chao Zhou, Radu Timofte, and Zhibo Chen. NTIRE 2025 challenge on short-form ugc video quality assessment and enhancement: Kwaisr dataset and study. InProceedings of the IEEE/CVF Conference on Compute...

    2025

  20. [28]

    NTIRE 2025 challenge on short-form ugc video quality assessment and enhancement: Methods and results

    Xin Li, Kun Yuan, Bingchen Li, Fengbin Guan, Yizhen Shao, Zihao Yu, Xijun Wang, Yiting Lu, Wei Luo, Suhang Yao, Ming Sun, Chao Zhou, Zhibo Chen, Radu Timofte, et al. NTIRE 2025 challenge on short-form ugc video quality assessment and enhancement: Methods and results. InProceed...

    2025

  21. [29]

    Lsdir: A large scale dataset for image restoration

    Yawei Li, Kai Zhang, Jingyun Liang, Jiezhang Cao, Ce Liu, Rui Gong, Yulun Zhang, Hao Tang, Yun Liu, Denis Demandolx, et al. Lsdir: A large scale dataset for image restoration. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1775–1787,...

    2023

  22. [30]

    Gustafsson, Zheng Zhao, Jens Sj¨olund, Thomas B

    Jie Liang, Radu Timofte, Qiaosi Yi, Shuaizheng Liu, Lingchen Sun, Rongyuan Wu, Xindong Zhang, Hui Zeng, Lei Zhang, Yibin Huang, Shuai Liu, Yongqiang Li, Chaoyu Feng, Xiaotao Wang, Lei Lei, Yuxiang Chen, Xiangyu Chen, Qiubo Chen, Fengyu Sun, Mengying Cui, Jiaxu Chen, Zhenyu Hu,...

    2024

  23. [31]

    NTIRE 2025 the 2nd restore any image model (RAIM) in the wild challenge

    Jie Liang, Radu Timofte, Qiaosi Yi, Zhengqiang Zhang, Shuaizheng Liu, Lingchen Sun, Rongyuan Wu, Xindong Zhang, Hui Zeng, Lei Zhang, et al. NTIRE 2025 the 2nd restore any image model (RAIM) in the wild challenge. In Proceedings of the IEEE/CVF Conference on Computer Vision and...

    2025

  24. [32]

    NTIRE 2025 XGC quality assessment challenge: Methods and results

    Xiaohong Liu, Xiongkuo Min, Qiang Hu, Xiaoyun Zhang, Jie Guo, et al. NTIRE 2025 XGC quality assessment challenge: Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  25. [33]

    NTIRE 2025 challenge on low light image enhancement: Methods and results

    Xiaoning Liu, Zongwei Wu, Florin-Alexandru Vasluianu, Hailong Yan, Bin Ren, Yulun Zhang, Shuhang Gu, Le Zhang, Ce Zhu, Radu Timofte, et al. NTIRE 2025 challenge on low light image enhancement: Methods and results. In Proceedings of the IEEE/CVF Conference on Computer Vision an...

    2025

  26. [34]

    The tenth NTIRE 2025 efficient super-resolution challenge report

    Bin Ren, Hang Guo, Lei Sun, Zongwei Wu, Radu Timofte, Yawei Li, et al. The tenth NTIRE 2025 efficient super-resolution challenge report. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  27. [35]

    NTIRE 2025 challenge on UGC video enhancement: Methods and results

    Nickolay Safonov, Alexey Bryntsev, Andrey Moskalenko, Dmitry Kulikov, Dmitriy Vatolin, Radu Timofte, et al. NTIRE 2025 challenge on UGC video enhancement: Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  28. [36]

    NTIRE 2025 challenge on event-based image deblurring: Methods and results

    Lei Sun, Andrea Alfarano, Peiqi Duan, Shaolin Su, Kaiwei Wang, Boxin Shi, Radu Timofte, Danda Pani Paudel, Luc Van Gool, et al. NTIRE 2025 challenge on event-based image deblurring: Methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2025

  29. [37]

    The tenth ntire 2025 image denoising challenge report

    Lei Sun, Hang Guo, Bin Ren, Luc Van Gool, Radu Timofte, Yawei Li, et al. The tenth ntire 2025 image denoising challenge report. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  30. [38]

    NTIRE 2025 image shadow removal challenge report

    Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Cailian Chen, Zongwei Wu, Radu Timofte, et al. NTIRE 2025 image shadow removal challenge report. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  31. [39]

    NTIRE 2025 ambient lighting normalization challenge

    Florin-Alexandru Vasluianu, Tim Seizinger, Zhuyun Zhou, Zongwei Wu, Radu Timofte, et al. NTIRE 2025 ambient lighting normalization challenge. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  32. [40]

    Swift parameter-free attention network for efficient super-resolution

    Cheng Wan, Hongyuan Yu, Zhiqi Li, Yihang Chen, Yajun Zou, Yuqing Liu, Xuanwu Yin, and Kunlong Zuo. Swift parameter-free attention network for efficient super-resolution. In2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 6246–6256, 2024. 13

    2024

  33. [41]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InInternational Conference on Computer Vision Workshops (ICCVW). 13, 15, 16

  34. [42]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InProceedings of the IEEE/CVF international conference on computer vision, pages 1905–1914, 2021. 9

    1905

  35. [43]

    NTIRE 2025 challenge on light field image super-resolution: Methods and results

    Yingqian Wang, Zhengyu Liang, Fengyuan Zhang, Lvli Tian, Longguang Wang, Juncheng Li, Jungang Yang, Radu Timofte, Yulan Guo, et al. NTIRE 2025 challenge on light field image super-resolution: Methods and results. In Proceedings of the IEEE/CVF Conference on Computer Vision and...

    2025

  36. [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. 14

    2004

  37. [45]

    One-step effective diffusion network for real-world image super-resolution.Advances in Neural Information Processing Systems, 37:92529–92553, 2024

    Rongyuan Wu, Lingchen Sun, Zhiyuan Ma, and Lei Zhang. One-step effective diffusion network for real-world image super-resolution.Advances in Neural Information Processing Systems, 37:92529–92553, 2024. 12

    2024

  38. [46]

    Seesr: Towards semantics-aware real-world image super-resolution

    Rongyuan Wu, Tao Yang, Lingchen Sun, Zhengqiang Zhang, Shuai Li, and Lei Zhang. Seesr: Towards semantics-aware real-world image super-resolution. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 25456–25467, 2024. 12

    2024

  39. [47]

    NTIRE 2025 challenge on single image reflection removal in the wild: Datasets, methods and results

    Kangning Yang, Jie Cai, Ling Ouyang, Florin-Alexandru Vasluianu, Radu Timofte, Jiaming Ding, Huiming Sun, Lan Fu, Jinlong Li, Chiu Man Ho, Zibo Meng, et al. NTIRE 2025 challenge on single image reflection removal in the wild: Datasets, methods and results. InProceedings of the...

    2025

  40. [48]

    NTIRE 2025 challenge on hr depth from images of specular and transparent surfaces

    Pierluigi Zama Ramirez, Fabio Tosi, Luigi Di Stefano, Radu Timofte, Alex Costanzino, Matteo Poggi, Samuele Salti, Stefano Mattoccia, et al. NTIRE 2025 challenge on hr depth from images of specular and transparent surfaces. In Proceedings of the IEEE/CVF Conference on Computer ...

    2025

  41. [49]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14821–14831, 2021. 14

    2021

  42. [50]

    Designing a practical degradation model for deep blind image super-resolution

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timofte. Designing a practical degradation model for deep blind image super-resolution. InProceedings of the IEEE/CVF international conference on computer vision, pages 4791–4800, 2021. 15

    2021

  43. [51]

    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. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018. 14

    2018

  44. [52]

    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. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018. 9

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.