Pith. sign in

REVIEW 4 major objections 5 minor 79 references

NTIRE 2025 Challenge on RAW Image Restoration and Super-Resolution

T0 review · 4 major / 5 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 2025 RAW-image challenge report shows every submitted restoration and super-resolution model beat its baselines on the synthetic test sets, with the restoration win going to a Samsung AI team and the general 2x super-resolution win…

desk verdict A useful challenge record with concrete method write-ups, but the SR track's own Table 1 contradicts the 'state-of-the-art' claim and the paper needs a major correction before I'd trust its benchmark numbers. read the letter →

arxiv 2506.02197 v2 pith:OPSSUUJU submitted 2025-06-02 eess.IV cs.CV

Marcos V. Conde , Radu Timofte , Zihao Lu , Xiangyu Kong , Xiaoxia Xing , Fan Wang , Suejin Han , MinKyu Park
show 31 more authors
Tianyu Zhang Xin Luo Yeda Chen Dong Liu Li Pang Yuhang Yang Hongzhong Wang Xiangyong Cao Ruixuan Jiang Senyan Xu Siyuan Jiang Xueyang Fu Zheng-Jun Zha Tianyu Hao Yuhong He Ruoqi Li Yueqi Yang Xiang Yu Guanlan Hong Minmin Yi Yuanjia Chen Liwen Zhang Zijie Jin Cheng Li Lian Liu Wei Song Heng Sun Yubo Wang Jinghua Wang Jiajie Lu Watchara Ruangsan
This is my paper · ORCID
classification eess.IVcs.CV
keywords RAWimagerestorationsuper-resolutionBayerpatternsignalprocessingknowledgedistillationdenoisingdeblurringbenchmarkreport
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

At the heart of this report is a benchmark claim: on the 2025 challenge's synthetic test images, every submitted method for RAW restoration and 2x super-resolution beat the provided baselines, and the winners define what the report calls the current state-of-the-art in RAW restoration. The restoration tests are graded at three degradation levels, from sensor-noise-only to blur-plus-noise, and the super-resolution test asks networks to double the size of 4-channel Bayer-pattern RAW images. The practical significance is that well-trained, heavily compressed networks can denoise, deblur, and upscale RAW data before the nonlinear ISP steps where sRGB image quality is typically lost. The report itself is careful to note that the synthetic downsampling model is still an open realism gap.

What carries the argument

The machinery that carries the argument is the challenge benchmark itself: a paired dataset of clean, high-resolution, 4-channel RGGB Bayer RAW images degraded through the BSRAW pipeline, which applies real noise profiles, multiple PSF blur kernels, and a fixed downsample strategy to synthesize low-resolution inputs. Participants train on the clean images and submit restorations of held-out degraded test images; scores are PSNR and SSIM computed in the RAW domain, with a 200K-parameter cap defining the efficient track. Within that scaffold, the top solutions share a few reusable mechanisms: distillation from a large teacher into reparameterized student convolutions, transformer blocks with gated or partial convolutions, and progressive multi-stage training with frequency-domain losses. These mechanisms, not any single architecture, are what lift the submitted models above the baselines.

What would settle it

Take the winning models and evaluate them on real paired RAW captures that were not synthesized by the BSRAW pipeline, for example a smartphone RAW burst with known ground truth from a long-exposure tripod shot; if the PSNR and SSIM advantage over the baselines shrinks substantially or reverses on those real pairs, the reported state-of-the-art is an artifact of the synthetic test set rather than a property of real RAW restoration.

Watch

Extended reading notes

Core claim

The report's central discovery is that the gap between the 2024 baselines and 2025 submissions on this benchmark is real and large: in the restoration track, the general winner reaches 46.04 dB PSNR at level 1 and 40.10 dB at level 3, while the NAFNet baseline sits at 43.50 dB and 37.49 dB; in the super-resolution track, the general winner reaches 42.70 dB PSNR versus 42.853 dB for the 2024 BSRAW reference on the same 200-image test set, with the efficient winner at 42.12 dB using only 0.18M parameters. The submitted methods achieve this by working directly on RGGB Bayer RAW data, combining CNN and transformer blocks with knowledge distillation, reparameterization, and progressive training schedules. According to the authors, the outputs show no detectable color artifacts, and synthetic RAW super-resolution is effectively solved at this benchmark scale, with realistic downsampling named as the remaining open problem.

Load-bearing premise

The load-bearing premise is that the BSRAW degradation pipeline—its noise profiles, blur kernels, and simple downsampling—faithfully mimics what a real camera does to a RAW image, so that scores on the synthetic test set predict performance on real sensor data.

Editorial extensions

If this is right

  • A single network can now denoise, deblur, and 2x-upscale Bayer-pattern RAW in one pass, with the efficient winner using only 0.18M parameters and the general winner 1.94M, which is within reach of mobile inference budgets.
  • Knowledge distillation from a large teacher into reparameterized student models is a repeatable recipe: the restoration winner used it to win both the efficient and general tracks with a 0.19M-parameter efficient model.
  • The difference between the efficient and general winners (45.10 vs 46.04 dB at restoration level 1) is small, suggesting parameter count is not the main bottleneck on this benchmark.
  • Because all methods operate on 4-channel RAW before demosaicing, they avoid the irreversible sRGB color and tone-mapping transformations that typically limit RGB-domain restoration.
  • The authors' conclusion that synthetic RAW super-resolution is solved implies that remaining visible gains will come mostly from more realistic degradation models, not from bigger networks.

Reading between the lines

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

  • If the BSRAW downsampling model is as unrealistic as the authors concede, the leaderboard likely overstates how well these methods will transfer to real phones; a future real-camera pair set could collapse the spread among top submissions.
  • The report's qualitative 'no detectable color artifacts' claim could be quantified with a color-fidelity metric over the RAW output, which would give a sharper test of whether RAW-domain processing really preserves color.
  • The winning recipe suggests that on-device ISP pipelines could place a small restored-RAW network before demosaicing, but validating that would require the organizers to release deployment code, not just architecture descriptions.
  • A direct extension would be to run the same degradation pipeline at 3x or 4x scale; the current 2x setup may be easier than the general super-resolution problem, so the 'solved' conclusion should not be extrapolated without new tests.
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 / 5 minor

Summary. This paper reports the NTIRE 2025 challenge on RAW image restoration and super-resolution. It describes two competition tracks: a RAW super-resolution track (2x upscaling under synthetic blur and noise) and a RAW restoration track (denoising/deblurring at three synthetic degradation levels). The paper specifies the datasets (Adobe MIT5K/BSRAW for SR; Raw2RAW plus additional smartphone sensors for restoration), the synthetic degradation pipeline, the evaluation protocol (PSNR/SSIM in the RAW domain on 200 and 140 test images, respectively), and the baselines. It then summarizes the submitted methods and their results, with detailed sections authored by the participant teams. The headline claims are that the submitted solutions improve over the baselines and that the report presents the current state-of-the-art in RAW restoration, with Samsung AI winning both restoration tracks and USTC winning the SR general track.

Significance. If the reported results are taken at face value, the paper is a useful archival benchmark for RAW-domain restoration and super-resolution. Its strengths are the transparent description of the challenge setup, the use of multiple baselines, the inclusion of parameter and computational-cost figures, and the open code repository for the degradation pipeline (AISP). The restoration track does show consistent gains over the listed baselines, e.g., Samsung AI General improves PSNR from 44.20 (RawIR) to 46.04 at degradation level 1. However, the paper's significance is undercut by an internal contradiction in the super-resolution track (see major comment 1) and by the absence of any real-sensor test data. As printed, the 'state-of-the-art' claim holds only for the synthetic benchmark, and even there it fails for the SR track unless the baseline comparison is corrected or reinterpreted.

major comments (4)
  1. [Table 1 and Sec. 2.2] The stress-test concern is confirmed by the manuscript itself. Table 1 reports all 2025 SR results on 'the complete testing set (200 images) [19]', the same split used for the 2024 baselines. The best 2025 general model (USTC) reaches 42.70 dB PSNR / 0.9479 SSIM, and the best efficient model (SMFFRaw-S) reaches 42.12 dB / 0.9433, while the listed 2024 baselines RBSFormer and BSRAW score 43.649 / 0.987 and 42.853 / 0.986, respectively. Thus no submitted 2025 SR method beats the 2024 baselines on the printed numbers, and the SSIM gap is large. This directly contradicts the Sec. 2.2 statement that 'the top performing challenge solutions improve the baseline performance' and weakens the abstract's state-of-the-art claim for the SR task. Please either correct the table, explain an explicit protocol difference between the 2024 and 2025 evaluations, or revise the claims. As printed, this is an internal contradiction, not a matter of external convention.
  2. [Sec. 2.3] The claim that 'there are not detectable color artifacts' is unsupported. No artifact metric, no perceptual study, and no systematic visual inspection protocol is described; Figure 1 shows only a few example patches. Either add a quantitative artifact analysis (e.g., demosaicing error or color-difference metrics) or explicitly label this statement as an informal observation from the organizers.
  3. [Secs. 2.3 and 3.3] Both test sets are synthetic, generated by the organizers with the same BSRAW degradation pipeline that participants use for training. The paper itself concedes in Sec. 2.3 that 'more realistic downsampling remains an open challenge.' Therefore the benchmark results should be framed as performance on a specific synthetic degradation model, not as evidence that the methods generalize to real sensor data. The current abstract and Sec. 2.2 wording overstates the scope of the conclusions.
  4. [Tables 1 and 2] No error bars, repeated-run statistics, or significance tests are reported, and the test sets are small (200 and 140 images). Several top-method differences are small (e.g., USTC 42.70 dB vs. RawRTSR-L 42.58 dB in Table 1; Samsung AI 46.04 dB vs. Miers 45.72 dB in Table 2, level 1). The rankings should be accompanied by a statement of uncertainty, or the paper should explicitly acknowledge that the numerical ordering of the top entries may not be statistically meaningful.
minor comments (5)
  1. [Title and abstract] The title contains a typo ('RA W'), and the abstract contains 'thee challenge period'.
  2. [Sec. 4.3 and Table 6] Table 6 reports validation PSNR/SSIM values for SMFFRaw-Small and SMFFRaw-Large, but the 'Test 1MP' column is left as 'Unknown' and the validation numbers do not match the corresponding test entries in Table 1. Clarify which model names correspond to which Table 1 rows.
  3. [Sec. 4.5] There are typos in this section: 'dowsample' and 'probablity'.
  4. [References] Several references are duplicated: [6] and [7] are the same NAFNet paper, [32] and [33] are the same RBSFormer paper, and [46] and [47] are the same AdamW paper. Please merge the duplicates.
  5. [Sec. 5.5] The team is called 'Team ER-NAFNet' in the section heading and 'Team ChickenRun' in the text ('The architecture proposed by Team ChickenRun is illustrated in Fig. 20'). Please reconcile the naming.

Circularity Check

1 steps flagged · score 4.0 of 10

The 'state-of-the-art' claim is measured on a synthetic test set generated by the same authors' BSRAW pipeline, making the benchmark self-referential; participant rankings remain independent, and the Table 1 contradiction is a correctness issue rather than circularity.

  1. self definitional [Abstract; Section 2.1 Dataset; Section 2.2 Baselines; Section 2.3 Results]
    "This report presents the current state-of-the-art in RAW Restoration. ... The LR degraded images can be generated on-line during training using the degradation pipeline proposed in BSRAW [18]. ... The synthetic test dataset is generated by applying our degradation pipeline at different levels."

    The central 'state-of-the-art' claim is evaluated on a test set synthesized with the same BSRAW degradation pipeline that the challenge organizers (the same authors) proposed and also use as a baseline. The test distribution is therefore an input chosen by the authors rather than an independent external benchmark; a submitted network's rank largely measures how well it fits that particular synthetic degradation model. The paper itself qualifies the conclusion with 'more realistic downsampling remains an open challenge', acknowledging that the benchmark is a constructed model.

full rationale

The report is a challenge survey, not a first-principles derivation, so most of its content is a description of independently submitted methods and measured scores. The held-out test set, parameter counts, and fidelity metrics are objective and the rankings among participant methods are not circular. The main circularity concern is that the test set and the baseline both derive from the same prior work by the same authors (BSRAW, RawIR, NTIRE 2024), so the 'current state-of-the-art' claim is defined relative to a self-generated synthetic benchmark rather than an external ground truth. The paper's own limitation statement in Section 2.3 ('more realistic downsampling remains an open challenge') confirms the authors do not claim real-sensor transfer, which limits the circularity rather than eliminating it. Separately, Table 1 contradicts Section 2.2's statement that 'the top performing challenge solutions improve the baseline performance': for the SR track, USTC (42.70 dB) and SMFFRaw-S (42.12 dB) are below RBSFormer (43.649 dB) and BSRAW (42.853 dB) on the same 200-image test set. That is an internal factual inconsistency and a correctness risk, but it is not a circular derivation, so it does not raise the circularity score beyond 4.

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

The paper's central claim (the stated state-of-the-art) depends on the organizers' synthetic degradation model, their choice of fidelity metrics, and their dataset filtering. These are domain assumptions inherited from BSRAW and prior NTIRE challenges; the paper does not introduce new physical entities or fitted constants.

free parameters (3)
  • Degradation level 1 (test) = noise only, y = x + n
    Chosen by the organizers to define the easiest test condition; all results at this level depend on this choice.
  • Degradation level 2 (test) = 0.3 probability of blur, 0.5 probability of real noise
    Chosen by the organizers to define the intermediate test condition.
  • Degradation level 3 (test) = all images have realistic blur and noise
    Chosen by the organizers to define the hardest test condition.
assumptions (3)
  • domain assumption BSRAW degradation pipeline is representative of real RAW degradations.
    Used to generate all synthetic training and test data (Sections 2.1, 2.3, 3.1). The paper acknowledges "more realistic downsampling remains an open challenge" (Section 2.3).
  • domain assumption PSNR and SSIM measured in the RAW domain are valid fidelity metrics for ranking.
    All results are reported as PSNR/SSIM without uncertainty; no perceptual or task-based validation is performed.
  • domain assumption The filtered source datasets are representative of the target distribution.
    Sections 2.1 and 3.1 describe manual filtering for sharpness and low ISO, but no domain-shift analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NTIRE 2025 Challenge on RAW Image Restoration and Super-Resolution." pith.science (2026). https://pith.science/paper/OPSSUUJU

@misc{pith2026250602197,
  author       = {Pith},
  title        = {Pith review of: NTIRE 2025 Challenge on RAW Image Restoration and Super-Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OPSSUUJU}},
  note         = {Machine review of arXiv:2506.02197}
}
read the original abstract

This paper reviews the NTIRE 2025 RAW Image Restoration and Super-Resolution Challenge, highlighting the proposed solutions and results. New methods for RAW Restoration and Super-Resolution could be essential in modern Image Signal Processing (ISP) pipelines, however, this problem is not as explored as in the RGB domain. The goal of this challenge is two fold, (i) restore RAW images with blur and noise degradations, (ii) upscale RAW Bayer images by 2x, considering unknown noise and blur. In the challenge, a total of 230 participants registered, and 45 submitted results during thee challenge period. This report presents the current state-of-the-art in RAW Restoration.

Figures

Figures reproduced from arXiv: 2506.02197 by the authors.

Figure 1
Figure 1. Samples of the NTIRE 2025 RAW Image Super-Resolution Challenge testing set. and submit their results. Thus, the participants do not have access to the ground-truth images. We provide samples of the testing set in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The overall structure of the RawRTSR network, including 0.188M parameters and running at 3.64 ms on the A100 GPU. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The overall structure of the RawRTSR-L network: [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (18 more)
Figure 4
Figure 4. Figure 4: Team USTC framework for RAW image super resolution. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The proposed network architecture of SMFFRaw by Team XJTU. (a) Overall framework. (b) Hybrid Attention Feature Extraction [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The architecture of the RBSFormer [32] used by Team EGROUP for RAW image super-resolution. rate gradually decreases from the initial learning rate to 1 × 10−6 . All experiments are implemented by PyTorch 1.11.0 with two NVIDIA 4090 GPUs. We used batch-size of 8 and cro…
Figure 7
Figure 7. Figure 7: NAFBN proposed by Team NJU RSR [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Adopted NAFBlock used by Team NJU RSR. 0.5 to be applied. To achieve a better performance on rel￾atively lower resolution images, random dowsample with AvePool2d and bicubic interpolation is added during each patch cropping procedure with the probablity of 0.3. The tra…
Figure 9
Figure 9. Figure 9: MambaIRv2 structure Method Description We implemented the MambaIRv2[24] method on raw data with the aim of providing a baseline from a different perspective for the competition. This method did not use any additional data, and we simplified its architecture to obtain a…
Figure 10
Figure 10. Figure 10: Downsampling structure fore, if we perform bicubic interpolation using the nearest 16 points, with the interpolation point positioned at the cen￾ter of the downsampled red pixel, we believe this approach achieves more precise downsampling. As for other settings, we co…
Figure 11
Figure 11. Figure 11: Main branch of RepRawSR proposed by Team EffiLowCVer. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: The overall architecture of the proposed ECAN model, showing the main pipeline and details of the EfficientResidualBlock [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: RAW image samples from RawIR Dataset. 5. RAW Image Restoration Methods 5.1. Efficient RAW Image Restoration Team SamsungAI Xiangyu Kong 1 , Xiaoxia Xing 1 , Suejin Han 2 , MinKyu Park 2 1 Samsung R&D Institute China - Beijing (SRC-B) 2 The Department of Camera Innovat…
Figure 15
Figure 15. Figure 15: Architectures of ERIRNet-S, with reduced channels [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 14
Figure 14. Figure 14: Training Stage Description by Samsung AI [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 16
Figure 16. Figure 16: Architectures of ERIRNet-T, with ConvTranspose and [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]
Figure 17
Figure 17. Figure 17: CABATTSwinFIR proposed method by Team Miers (Xiaomi Inc.). [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]
Figure 18
Figure 18. Figure 18: Overall architecture proposed by the Team WIRTeam. [PITH_FULL_IMAGE:figures/full_fig_p018_18.png]
Figure 19
Figure 19. Figure 19: Overall architecture of the LMPR-Net method proposed by the Team WIRTeam. [PITH_FULL_IMAGE:figures/full_fig_p018_19.png]
Figure 20
Figure 20. Figure 20: The overall network architecture proposed by Team [PITH_FULL_IMAGE:figures/full_fig_p019_20.png]
Figure 21
Figure 21. Figure 21: Visual comparison using the NTIRE 2025 Challenge on RAW Image Restoration Challenge testing set (gp 2 11.npy). RAW images have 512 × 512 resolution and 4-channels (RGGB Bayer pattern). RAW images are visualized using bilinear demosaicing, gamma correction and tone map…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 48 canonical work pages

  1. [19]

    Deep RAW image super-resolution

    Marcos V Conde, Florin Vasluianu, and Radu Timofte. Deep RAW image super-resolution. a NTIRE 2024 challenge sur- vey. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition Workshops, 2024. 2

    work page 2024

  2. [1]

    Semi-Supervised Raw-to-Raw Mapping

    Mahmoud Afifi and Abdullah Abuolaim. Semi-supervised raw-to-raw mapping.arXiv preprint arXiv:2106.13883,

    work page Pith review arXiv

  3. [2]

    Unprocessing images for learned raw denoising

    Tim Brooks, Ben Mildenhall, Tianfan Xue, Jiawen Chen, Dillon Sharlet, and Jonathan T Barron. Unprocessing images for learned raw denoising. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11036–11045, 2019. 1

    work page 2019

  4. [3]

    Learning photographic global tonal adjustment with a database of input / output image pairs

    Vladimir Bychkovsky, Sylvain Paris, Eric Chan, and Fr ´edo Durand. Learning photographic global tonal adjustment with a database of input / output image pairs. InCVPR, 2011. 2

    work page 2011

  5. [4]

    Do-conv: Depthwise over-parameterized convolutional layer.IEEE Transactions on Image Processing, 31:3726– 3736, 2022

    Jinming Cao, Yangyan Li, Mingchao Sun, Ying Chen, Dani Lischinski, Daniel Cohen-Or, Baoquan Chen, and Changhe Tu. Do-conv: Depthwise over-parameterized convolutional layer.IEEE Transactions on Image Processing, 31:3726– 3736, 2022. 17

    work page 2022

  6. [5]

    Two deterministic half-quadratic regular- ization algorithms for computed imaging

    Pierre Charbonnier, Laure Blanc-Feraud, Gilles Aubert, and Michel Barlaud. Two deterministic half-quadratic regular- ization algorithms for computed imaging. InProceedings of 1st international conference on image processing, pages 168–172. IEEE, 1994. 17

    work page 1994

  7. [6]

    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. 18

    work page 2022

  8. [7]

    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. 3, 4, 14

    work page 2022

Show all 79 references
  1. [8]

    Hat: Hybrid attention transformer for image restoration.arXiv preprint arXiv:2309.05239, 2023

    Xiangyu Chen, Xintao Wang, Wenlong Zhang, Xiangtao Kong, Yu Qiao, Jiantao Zhou, and Chao Dong. Hat: Hybrid attention transformer for image restoration.arXiv preprint arXiv:2309.05239, 2023. 15

    2023

  2. [9]

    Mofa: A model simplification roadmap for image restoration on mobile devices

    Xiangyu Chen, Ruiwen Zhen, Shuai Li, Xiaotian Li, and Guanghui Wang. Mofa: A model simplification roadmap for image restoration on mobile devices. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 1322–1332, 2023. 3, 4

    2023

  3. [10]

    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, 14

    2024

  4. [11]

    Bidirectional multi-scale implicit neural representations for image derain- ing

    Xiang Chen, Jinshan Pan, and Jiangxin Dong. Bidirectional multi-scale implicit neural representations for image derain- ing. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 25627–25636,

  5. [12]

    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) Work- shops, 2025. 2

    2025

  6. [13]

    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) Work- shops, 2025. 2

    2025

  7. [14]

    Nafssr: Stereo image super-resolution using nafnet, 2022

    Xiaojie Chu, Liangyu Chen, and Wenqing Yu. Nafssr: Stereo image super-resolution using nafnet, 2022. 9

    2022

  8. [15]

    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. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  9. [16]

    Raw image reconstruc- tion from RGB on smartphones

    Marcos Conde, Radu Timofte, et al. Raw image reconstruc- tion from RGB on smartphones. NTIRE 2025 challenge re- port. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) Workshops,

    2025

  10. [17]

    Perceptual image enhancement for smartphone real-time applications

    Marcos V Conde, Florin Vasluianu, Javier Vazquez-Corral, and Radu Timofte. Perceptual image enhancement for smartphone real-time applications. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1848–1858, 2023. 1

    2023

  11. [18]

    Bsraw: Improving blind raw image super-resolution

    Marcos V Conde, Florin Vasluianu, and Radu Timofte. Bsraw: Improving blind raw image super-resolution. InPro- ceedings of the IEEE/CVF Winter Conference on Applica- tions of Computer Vision, pages 8500–8510, 2024. 1, 2, 3, 6, 8, 9, 15

    2024

  12. [20]

    To- ward efficient deep blind raw image restoration

    Marcos V Conde, Florin Vasluianu, and Radu Timofte. To- ward efficient deep blind raw image restoration. In2024 IEEE International Conference on Image Processing (ICIP), pages 1725–1731. IEEE, 2024. 2, 4, 18, 19

    2024

  13. [21]

    NTIRE 2025 challenge on night photography rendering

    Egor Ershov, Sergey Korchagin, Alexei Khalin, Artyom Pan- shin, 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 ch...

    2025

  14. [22]

    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 Con- ference on Computer Vision and Pattern Recognition (...

    2025

  15. [23]

    Syenet: A simple yet effective net- work for multiple low-level vision tasks with real-time per- formance on mobile device

    Weiran Gou, Ziyao Yi, Yan Xiang, Shaoqing Li, Zibin Liu, Dehui Kong, and Ke Xu. Syenet: A simple yet effective net- work for multiple low-level vision tasks with real-time per- formance on mobile device. InProceedings of the IEEE/CVF International Conference on Computer Vision...

    2023

  16. [24]

    Mambairv2: Attentive state space restoration.arXiv preprint arXiv:2411.15269,

    Hang Guo, Yong Guo, Yaohua Zha, Yulun Zhang, Wenbo Li, Tao Dai, Shu-Tao Xia, and Yawei Li. Mambairv2: Attentive state space restoration.arXiv preprint arXiv:2411.15269,

    arXiv

  17. [25]

    NTIRE 2025 challenge on text to image generation model quality assess- ment

    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 assess- ment. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) Workshops,

    2025

  18. [26]

    Convolutional neural networks for direct text deblurring

    Michal Hradi ˇs, Jan Kotera, Pavel Zemcık, and FilipˇSroubek. Convolutional neural networks for direct text deblurring. In Proceedings of BMVC, 2015. 6

    2015

  19. [27]

    Squeeze-and-excitation net- works

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation net- works. InCVPR, pages 7132–7141, 2018. 13

    2018

  20. [28]

    Replac- ing mobile camera isp with a single deep learning model

    Andrey Ignatov, Luc Van Gool, and Radu Timofte. Replac- ing mobile camera isp with a single deep learning model. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 536–537,

  21. [29]

    Learned smartphone isp on mobile npus with deep learning, mobile ai 2021 challenge: Report

    Andrey Ignatov, Cheng-Ming Chiang, Hsien-Kai Kuo, Anas- tasia Sycheva, and Radu Timofte. Learned smartphone isp on mobile npus with deep learning, mobile ai 2021 challenge: Report. InCVPR Workshops, pages 2503–2514, 2021. 1

    2021

  22. [30]

    NTIRE 2025 challenge on video quality enhancement for video con- ferencing: Datasets, methods and results

    Varun Jain, Zongwei Wu, Quan Zou, Louis Florentin, Hen- rik Turbell, Sandeep Siddhartha, Radu Timofte, et al. NTIRE 2025 challenge on video quality enhancement for video con- ferencing: Datasets, methods and results. InProceedings of the IEEE/CVF Conference on Computer Vision ...

    2025

  23. [31]

    Rbsformer: En- hanced transformer network for raw image super-resolution

    Siyuan Jiang, Senyan Xu, and Xingfu Wang. Rbsformer: En- hanced transformer network for raw image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 6479–6488, 2024. 8

    2024

  24. [32]

    Rbsformer: En- hanced transformer network for raw image super-resolution

    Siyuan Jiang, Senyan Xu, and Xingfu Wang. Rbsformer: En- hanced transformer network for raw image super-resolution. In2024 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW), pages 6479–6488,

  25. [33]

    Rbsformer: En- hanced transformer network for raw image super-resolution

    Siyuan Jiang, Senyan Xu, and Xingfu Wang. Rbsformer: En- hanced transformer network for raw image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6479–6488, 2024. 2, 6, 8

    2024

  26. [34]

    A software platform for manipulating the camera imaging pipeline

    Hakki Can Karaimer and Michael S Brown. A software platform for manipulating the camera imaging pipeline. In ECCV, pages 429–444, 2016. 1

    2016

  27. [35]

    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

  28. [36]

    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. In Proceedings of the IEEE/CVF Conference on Com...

    2025

  29. [37]

    Promptcir: blind compressed image restoration with prompt learning

    Bingchen Li, Xin Li, Yiting Lu, Ruoyu Feng, Mengxi Guo, Shijie Zhao, Li Zhang, and Zhibo Chen. Promptcir: blind compressed image restoration with prompt learning. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6442–6452, 2024. 16

    2024

  30. [38]

    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

  31. [39]

    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 Com- pu...

    2025

  32. [40]

    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. InPro- ce...

    2025

  33. [41]

    Multi-level feature fusion network for lightweight stereo image super-resolution

    Yunxiang Li, Wenbin Zou, Qiaomu Wei, Feng Huang, and Jing Wu. Multi-level feature fusion network for lightweight stereo image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6489–6498, 2024. 6

    2024

  34. [42]

    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 Vi- sion a...

    2025

  35. [43]

    Learning raw image denoising with bayer pattern unification and bayer preserving augmentation

    Jiaming Liu, Chi-Hao Wu, Yuzhi Wang, Qin Xu, Yuqian Zhou, Haibin Huang, Chuan Wang, Shaofan Cai, Yifan Ding, Haoqiang Fan, et al. Learning raw image denoising with bayer pattern unification and bayer preserving augmentation. InProceedings of the IEEE/CVF Conference on Computer...

    2019

  36. [44]

    NTIRE 2025 XGC quality assessment chal- lenge: Methods and results

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

    2025

  37. [45]

    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 Vi- sion ...

    2025

  38. [46]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 19

    2017 arXiv

  39. [47]

    Decoupled weight de- cay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight de- cay regularization. InInternational Conference on Learning Representations, 2019. 5

    2019

  40. [48]

    Rewrite the stars

    Xu Ma, Xiyang Dai, Yue Bai, Yizhou Wang, and Yun Fu. Rewrite the stars. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 5694–5703, 2024. 6

    2024

  41. [49]

    Intriguing findings of frequency selection for image deblurring

    Xintian Mao, Yiming Liu, Fengze Liu, Qingli Li, Wei Shen, and Yan Wang. Intriguing findings of frequency selection for image deblurring. InProceedings of the AAAI Conference on Artificial Intelligence, pages 1905–1913, 2023. 6

    1905

  42. [50]

    Modelling the scene dependent imaging in cameras with a deep neural network

    Seonghyeon Nam and Seon Joo Kim. Modelling the scene dependent imaging in cameras with a deep neural network. InProceedings of the IEEE International Conference on Computer Vision, pages 1717–1725, 2017. 1

    2017

  43. [51]

    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 Infor- mation Processing Systems, 2023. 17

    2023

  44. [52]

    The tenth NTIRE 2025 efficient super- resolution challenge report

    Bin Ren, Hang Guo, Lei Sun, Zongwei Wu, Radu Timo- fte, 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

  45. [53]

    Neural blind deconvolution using deep priors

    Dongwei Ren, Kai Zhang, Qilong Wang, Qinghua Hu, and Wangmeng Zuo. Neural blind deconvolution using deep priors. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3341–3350,

  46. [54]

    Efficient super-resolution algorithm using overlapping bicubic inter- polation

    Watchara Ruangsang and Supavadee Aramvith. Efficient super-resolution algorithm using overlapping bicubic inter- polation. In2017 IEEE 6th Global Conference on Consumer Electronics (GCCE), pages 1–2. IEEE, 2017. 12

    2017

  47. [55]

    Multi-fusnet of cross channel network for image super-resolution.IEEE Access, 11:56287–56299, 2023

    Watchara Ruangsang, Supavadee Aramvith, and Takao Onoye. Multi-fusnet of cross channel network for image super-resolution.IEEE Access, 11:56287–56299, 2023. 12, 13

    2023

  48. [56]

    NTIRE 2025 challenge on UGC video enhancement: Meth- ods and results

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

    2025

  49. [57]

    Mobilenetv2: Inverted residuals and linear bottlenecks

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zh- moginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 4510–4520, 2018. 13

    2018

  50. [58]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Husz ´ar, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE conference on compu...

    2016

  51. [59]

    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

  52. [60]

    The tenth ntire 2025 image denoising challenge report

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

    2025

  53. [61]

    A+: Adjusted anchored neighborhood regression for fast super- resolution

    Radu Timofte, Vincent De Smet, and Luc Van Gool. A+: Adjusted anchored neighborhood regression for fast super- resolution. InComputer Vision–ACCV 2014: 12th Asian Conference on Computer Vision, Singapore, Singapore, November 1-5, 2014, Revised Selected Papers, Part IV 12, page...

    2014

  54. [62]

    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. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025. 2

    2025

  55. [63]

    NTIRE 2025 ambi- ent lighting normalization challenge

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

    2025

  56. [64]

    Chan, Ke Yu, Chao Dong, and Chen Change Loy

    Xintao Wang, Kelvin C.K. Chan, Ke Yu, Chao Dong, and Chen Change Loy. Edvr: Video restoration with enhanced deformable convolutional networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019. 8

    2019

  57. [65]

    Practical deep raw image denoising on mobile devices

    Yuzhi Wang, Haibin Huang, Qin Xu, Jiaming Liu, Yiqun Liu, and Jue Wang. Practical deep raw image denoising on mobile devices. InComputer Vision–ECCV 2020: 16th Eu- ropean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part VI, pages 1–16. Springer, 2020. 3

    2020

  58. [66]

    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. InPro- ceedings of the IEEE/CVF Conference on Computer Vision an...

    2025

  59. [67]

    Invertible im- age signal processing

    Yazhou Xing, Zian Qian, and Qifeng Chen. Invertible im- age signal processing. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 6287–6296, 2021. 1

    2021

  60. [68]

    Towards real scene super-resolution with raw images

    Xiangyu Xu, Yongrui Ma, and Wenxiu Sun. Towards real scene super-resolution with raw images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1723–1731, 2019. 1, 2

    2019

  61. [69]

    Exploiting raw images for real-scene super-resolution

    Xiangyu Xu, Yongrui Ma, Wenxiu Sun, and Ming-Hsuan Yang. Exploiting raw images for real-scene super-resolution. IEEE transactions on pattern analysis and machine intelli- gence, 44(4):1905–1921, 2020. 2

    1905

  62. [70]

    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

  63. [71]

    Casr: Efficient cascade network structure with channel aligned method for 4k real-time single image super-resolution

    Kihwan Yoon, Ganzorig Gankhuyag, Jinman Park, Haengseon Son, and Kyoungwon Min. Casr: Efficient cascade network structure with channel aligned method for 4k real-time single image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni...

    2024

  64. [72]

    Inceptionnext: When inception meets convnext

    Weihao Yu, Pan Zhou, Shuicheng Yan, and Xinchao Wang. Inceptionnext: When inception meets convnext. InProceed- ings of the IEEE/cvf conference on computer vision and pat- tern recognition, pages 5672–5683, 2024. 6

    2024

  65. [73]

    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, Ste- fano Mattoccia, et al. NTIRE 2025 challenge on hr depth from images of specular and transparent surfaces. InPro- ceedings of the IEEE/CVF Conference on Comput...

    2025

  66. [74]

    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. InCVPR,

  67. [75]

    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,

  68. [76]

    Swinfir: Revisiting the swinir with fast fourier convolution and improved training for image super- resolution.arXiv preprint arXiv:2208.11247, 2022

    Dafeng Zhang, Feiyu Huang, Shizhuo Liu, Xiaobing Wang, and Zhezhu Jin. Swinfir: Revisiting the swinir with fast fourier convolution and improved training for image super- resolution.arXiv preprint arXiv:2208.11247, 2022. 8, 15

    2022 arXiv

  69. [77]

    Deep unfold- ing network for image super-resolution

    Kai Zhang, Luc Van Gool, and Radu Timofte. Deep unfold- ing network for image super-resolution. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3217–3226, 2020. 18

    2020

  70. [78]

    Shufflenet: An extremely efficient convolutional neural net- work for mobile devices

    Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, and Jian Sun. Shufflenet: An extremely efficient convolutional neural net- work for mobile devices. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 6848–6856, 2018. 6

    2018

  71. [79]

    Image super-resolution using very deep residual channel attention networks

    Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu. Image super-resolution using very deep residual channel attention networks. InEuropean Confer- ence on Computer Vision, pages 286–301, 2018. 13

    2018

Pith tools

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