Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

EAMamba: Efficient All-Around Vision State Space Model for Image Restoration

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read EAMamba shows that splitting channels across scan heads lets a Vision Mamba restoration model cut FLOPs by 31-89% while keeping quality.

desk verdict A genuinely useful efficiency trick for Mamba-based restoration, with an incomplete baseline set that makes the headline FLOPs claim unproven against the full prior art. read the letter →

arxiv 2506.22246 v1 pith:OGT433TR submitted 2025-06-27 cs.CV

Yu-Cheng Lin , Yu-Syuan Xu , Hao-Wei Chen , Hsien-Kai Kuo , Chun-Yi Lee This is my paper · ORCID
classification cs.CV
keywords imagerestorationstatespacemodelsVisionMambaselectivescanmulti-headall-aroundscanningsuper-resolutiondenoising
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

EAMamba is a Vision Mamba architecture for image restoration that targets the dominant cost of reading an image in several directions. Earlier restoration Mambas run a full selective scan over all channels for every scan direction, so adding diagonal scans multiplies computation. EAMamba's Multi-Head Selective Scan Module splits the feature channels into groups, scans each group in a different direction, and concatenates the results, so the cost of extra directions is largely removed. With an all-around scan set (horizontal, vertical, diagonal, flipped diagonal, and reversals), the model reports 31-89% lower FLOPs than MambaIR and VMambaIR across super-resolution, denoising, deblurring, and dehazing, while keeping PSNR within 0.05 dB or improving it.

What carries the argument

Multi-Head Selective Scan (MHSS) and all-around scanning. MHSS splits an input feature into n channel groups; each group is transformed into a flattened one-dimensional sequence, run through a selective scan, inverse-transformed, and concatenated with the others. Because every head scans only C/n channels, running n heads costs about the same as one full-channel scan, so the number of scan directions stops being the main FLOPs driver. The all-around scanning strategy is the set of transformation patterns assigned to the heads: horizontal, vertical, diagonal, flipped diagonal, and their reversed orientations. Its job is to solve local pixel forgetting by ensuring that neighboring pixels in the 2D feature map stay close in at least one of the scanned sequences.

What would settle it

Run a matched comparison of EAMamba against the prior restoration Mamba designed to reduce scan-sequence overhead, using the same training data, patch schedule, and FLOPs measurement on RealSR $\times$4 and SIDD; if the FLOPs advantage over that method falls below 31% or PSNR drops below parity, the headline reduction claim fails. A second check: retrain EAMamba with only 2D scans at the same FLOPs budget; if PSNR does not drop, the all-around scan is not the source of the reported quality.

Watch

Extended reading notes

Core claim

The paper's central claim is that a channel-split Multi-Head Selective Scan Module (MHSSM) lets a Vision Mamba restoration network scan an image in many directions without paying the usual cost per sequence. Standard 2D selective scan processes all channels for each flattened one-dimensional sequence, so four or eight scan directions multiply FLOPs and parameters. MHSS partitions the channels into groups, gives each group one scan direction, and concatenates the group outputs, keeping total computation close to a single selective scan. The all-around scan set then supplies the spatial coverage a 2D scan misses: diagonal and flipped-diagonal directions keep spatially adjacent pixels closer in the token sequence, which the paper's effective receptive field visualizations show as improved local dependency preservation. Across four restoration tasks, the paper reports 31-89% FLOPs reductions relative to MambaIR and VMambaIR with essentially unchanged PSNR, and larger gains on motion deblurring and indoor dehazing.

Load-bearing premise

The headline efficiency gain depends on comparing with MambaIR and VMambaIR; a prior restoration method that already attacks the same scan cost is absent from the comparison, so the 31-89% reduction may not survive against it.

Editorial extensions

If this is right

  • Adding scan directions no longer multiplies compute, so multi-directional scanning becomes an affordable design choice rather than a cost to be minimized.
  • At 137 GFLOPs at 256 $\times$ 256, EAMamba is in the same compute class as Restormer while keeping Mamba's linear scaling with pixel count, narrowing the efficiency gap between state-space and attention-based restoration.
  • The 31-89% FLOPs reductions across four tasks imply that running every scan over the full channel map, not the selective scan itself, was the main efficiency bottleneck in earlier Vision Mamba restoration models.
  • The effective receptive field results indicate scan-direction diversity preserves local structure, which should transfer to other dense prediction tasks such as deraining, inpainting, and depth estimation.

Reading between the lines

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

  • Our inference: because MHSS decouples scan count from compute, the same channel-split design could make multi-directional state-space scans practical for video restoration or high-resolution images, settings the paper does not evaluate.
  • Our inference: the channel split trades per-head width for directional diversity, so quality likely depends on the number of heads and group size; a systematic sweep of those hyperparameters, absent from the paper, could change the optimal scan configuration.
  • Our inference: the all-around strategy can be read as an efficient state-space approximation of multi-orientation local feature aggregation; combining it with input-dependent scan routing or head weighting is a natural extension the paper leaves unexplored.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes EAMamba, a U-shaped image restoration architecture whose core token mixer, the Multi-Head Selective Scan Module (MHSSM), splits features along the channel dimension into groups and applies different 2D scan sequences per group before concatenation. An 'all-around' scan combining a standard 2D scan with a diagonal scan is used to address local pixel forgetting. On denoising, super-resolution, deblurring, and dehazing benchmarks, EAMamba reports FLOPs reductions of 89%, 41%, and 31% relative to MambaIR, MambaIR-UNet, and VMambaIR, respectively, with comparable PSNR/SSIM; the ablation in Table 8 shows that replacing a full-channel 2DSSM with MHSSM halves FLOPs with a negligible PSNR drop.

Significance. If the reported efficiency numbers hold in a complete comparison, the channel-split MHSS design is a simple and reusable mechanism that lowers the compute barrier for multi-directional Mamba restoration. Strengths: the core ablation is clean (Table 8); FLOPs are measured with fvcore on a fixed input resolution; the paper releases code; and the performance is evaluated across four standard restoration tasks. The main weakness is the incomplete baseline set for the 'existing low-level Vision Mamba methods' claim, which currently covers only MambaIR and VMambaIR.

major comments (3)
  1. [Section 2, Tables 1-3, Fig. 1] The efficiency claim over 'existing low-level Vision Mamba methods' is not established because the comparison set omits MambaIRv2 (arXiv:2411.15262), a prior method explicitly targeting the computational overhead of multi-directional 2D selective scans for image restoration. Please add MambaIRv2 to the efficiency comparison (or justify its exclusion), since the 31-89% reduction range and the 'new efficiency frontier' claim in Fig. 1 depend on the completeness of this baseline set.
  2. [Section 4.7, Table 7] The default scan combination '2D + Diagonal' appears to be selected based on the benchmark results shown in Table 7 rather than on a pre-registered validation criterion. For example, on GoPro the 2D+Hilbert combination outscores the chosen default (33.66 vs. 33.58 dB), and on SOTS-Indoor the 2D+Diagonal+Z-order combination is better (43.37 vs. 43.19 dB). State whether the default was fixed before seeing these test results; if not, provide a validation-based selection rule or discuss the sensitivity of the reported performance to this choice.
  3. [Section 3.3.1, Eq. (3)] The paper never specifies the number of scan groups n or how the scanning directions are assigned to groups, even though the efficiency argument of MHSS rests on this partitioning. As written, Eq. (3) defines the transform only in the abstract, and the statement that MHSS 'maintains computational complexity comparable to standard selective scan' is not derived. Please report n, describe the Transform/InverseTransform for each group, and provide a FLOPs expression in terms of C and n so the 2x reduction in Table 8 can be verified analytically.
minor comments (6)
  1. [Section 1] In the Introduction, 'which is a crucial for image restoration tasks' should read 'which is crucial for image restoration tasks'.
  2. [Section 3.3.1] Equation (3) uses the notation N = {i \in R : i = {1, 2, ..., n}}, which is mathematically incorrect; it should be N = {1, 2, ..., n}.
  3. [Figure 8 caption] In Fig. 8, 'VmambaIR' is written with a lowercase 'm' while the rest of the paper uses 'VMambaIR'.
  4. [Table 1] Table 1's header 'FLOPsCBSD68' lacks a separating space, and the units should be applied consistently across the table.
  5. [Section 4.1] The paper does not state whether the FLOPs values for all baselines in Tables 1-5 are measured with the same fvcore settings and input resolution; please clarify that all methods were measured identically.
  6. [References] Some references are incomplete or inconsistent with the venue of publication; for example, [53] cites the CoRR version of Mamba rather than the published NeurIPS version, and several other references similarly cite preprint versions.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the efficiency and performance claims are empirical and benchmark-based, with no claim reducing to its own inputs by construction.

full rationale

The paper's two central claims are (1) that the Multi-Head Selective Scan Module (MHSSM) avoids the FLOPs and parameter growth of full-channel multi-direction scanning, and (2) that all-around scanning improves restoration quality. Claim (1) is a direct architectural identity rather than a fitted result: Eq. (3) defines SS_N_in = Transform(Split(MHSS_in)), so each head scans only C/n channels, and concatenating n such heads performs the same total channel-computation as one full-channel scan while implementing multiple scan directions. Table 8 then verifies the consequence by measuring 286G FLOPs for the 2DSSM baseline and 137G for MHSSM, with FLOPs computed by the external fvcore tool. The 31-89% headline range is arithmetic from the benchmark tables: 137G versus 1290G (MambaIR, Table 1), 230G (MambaIR-UNet, Table 2), and 200G (VMambaIR, Table 3). Claim (2) is validated empirically on standard benchmarks: Table 6 compares scan strategies on SIDD, Table 8 isolates the contribution of MHSSM and all-around scanning on Urban100, and Table 7 reports every scan combination, with the 2D+Diagonal combination selected as the default because it 'generally yields good performance.' This is ordinary model selection with all configurations disclosed, not a fit renamed as a prediction. The ERF visualizations (Fig. 4, Fig. 11) serve as supporting evidence for the spatial-localization mechanism, not as circular definitions of the performance gain. No self-citation chain is load-bearing: MambaIR and VMambaIR are prior external baselines with independently published results, and no 'uniqueness theorem' is imported to force the architecture. The omission of MambaIRv2 (arXiv:2411.15262) from the comparison may narrow the external validity of the phrase 'existing low-level Vision Mamba methods,' but that is a completeness or correctness concern, not circularity, because no equation or parameter in this paper is defined in terms of the headline claim. The derivation chain is self-contained against external benchmarks, and no specific reduction of a predicted result to its own input could be identified.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no physical or conceptual entities beyond the MHSSM architecture. The main ledger entries are architecture hyperparameters chosen by experiment. The axioms are the background assumptions inherited from Mamba literature, the ERF-to-quality link, and the comparability of published baselines.

free parameters (4)
  • default scan combination = 2D + diagonal
    Chosen after evaluating multiple combinations on SIDD, RealSRx4, GoPro, and SOTS-Indoor (Table 7); alternative combinations can be better on specific datasets, so this choice is tuned to the benchmark set.
  • number of scan groups n in MHSS = not reported in main text
    The channel split count determines how many scan directions fit in one pass; without this value, the exact architecture is underspecified in the text.
  • channel expansion factor lambda in MHSSM = not reported in main text
    Width expansion in both MHSSM branches is a free design choice that affects parameter count and FLOPs; it is not specified.
  • channel MLP choice = Simple FFN
    Selected from Table 9 as the best efficiency-performance trade-off; GDFN is slightly stronger, so the choice is a tuned compromise.
assumptions (4)
  • standard math The selective scan operation on 1D sequences has linear complexity and is a valid token mixer (Mamba [53], VMamba [54]).
    EAMamba inherits Mamba's complexity and functionality claims from the cited literature rather than proving them.
  • domain assumption ERF gradient-flow maps indicate which local pixels are actually used by the model and correlate with restoration quality.
    Section 3.3.2 and Fig. 4/11 use ERF visualizations to diagnose 'local pixel forgetting' and to motivate all-around scanning; no formal metric links ERF coverage to PSNR.
  • domain assumption Baseline numbers from the original manuscripts were obtained under protocols comparable to EAMamba's training setup.
    Tables 1-5 mix numbers from original papers with numbers from EAMamba; any difference in data splits, training iterations, or metric code would change the comparison.
  • ad hoc to paper MambaIR and VMambaIR are the relevant low-level Vision Mamba baselines for the efficiency claim.
    The abstract and Section 2 restrict the comparison to [57,58], omitting MambaIRv2, which also targets scan overhead; this set choice is not justified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EAMamba: Efficient All-Around Vision State Space Model for Image Restoration." pith.science (2026). https://pith.science/paper/OGT433TR

@misc{pith2026250622246,
  author       = {Pith},
  title        = {Pith review of: EAMamba: Efficient All-Around Vision State Space Model for Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OGT433TR}},
  note         = {Machine review of arXiv:2506.22246}
}
read the original abstract

Image restoration is a key task in low-level computer vision that aims to reconstruct high-quality images from degraded inputs. The emergence of Vision Mamba, which draws inspiration from the advanced state space model Mamba, marks a significant advancement in this field. Vision Mamba demonstrates excellence in modeling long-range dependencies with linear complexity, a crucial advantage for image restoration tasks. Despite its strengths, Vision Mamba encounters challenges in low-level vision tasks, including computational complexity that scales with the number of scanning sequences and local pixel forgetting. To address these limitations, this study introduces Efficient All-Around Mamba (EAMamba), an enhanced framework that incorporates a Multi-Head Selective Scan Module (MHSSM) with an all-around scanning mechanism. MHSSM efficiently aggregates multiple scanning sequences, which avoids increases in computational complexity and parameter count. The all-around scanning strategy implements multiple patterns to capture holistic information and resolves the local pixel forgetting issue. Our experimental evaluations validate these innovations across several restoration tasks, including super resolution, denoising, deblurring, and dehazing. The results validate that EAMamba achieves a significant 31-89% reduction in FLOPs while maintaining favorable performance compared to existing low-level Vision Mamba methods.

Figures

Figures reproduced from arXiv: 2506.22246 by the authors.

Figure 1
Figure 1. Computational efficiency versus image quality across model ar [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Illustration of an all-around scanning approach that combines [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 5
Figure 5. An overview of the proposed EAMamba framework. EAMamba framework is an encoder-decoder architecture. Both the encoder and the decoder [PITH_FULL_IMAGE:figures/full_fig_p003_5.png] view at source ↗
Figures from the paper (6 more)
Figure 6
Figure 6. Figure 6: Illustration of the Multi-Head Selective Scan (MHSS) with our [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison of MPRNet [36], UFormer-B [50], Restormer [51], MambaIR-UNet [57], and our EAMamba on SIDD [59] validation set. The first row illustrates the results and the second row shows the normalized difference between the ground truth and the generated results…
Figure 8
Figure 8. Figure 8: Visual comparison of Restormer [51], MambaIR-UNet [57], VmambaIR [58], and our EAMamba on the real-world super-resolution RealSR dataset [61] at scaling factors of ×4. The cropped regions represent the normalized difference between the ground truth and the generated re…
Figure 9
Figure 9. Figure 9: Visual comparison of MPRNet [36], MAXIM-3S [85], Restormer [51], SFNet [39], and our EAMamba on GoPro [62] validation set. The first row illustrates the results and the second row shows the normalized difference between the ground truth and the generated results. Low Q…
Figure 10
Figure 10. Figure 10: Visual comparison of Dehamer [46], MAXIM-2S [85], DehazeFormer-L [47], and our EAMamba on SOTS-Indoor [63] test set. The first row illustrates the results and the second row shows the normalized difference between the ground truth and the generated results [PITH_FULL…
Figure 11
Figure 11. Figure 11: Illustration of the ERF results for different scanning strategies, [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. VEMamba: Efficient Isotropic Reconstruction of Volume Electron Microscopy with Axial-Lateral Consistent Mamba

    cs.CV 2026-03 conditional novelty 5.0 of 10

    VEMamba applies Mamba state-space models with axial-lateral chunked scanning and MoCo-based degradation learning to achieve efficient isotropic reconstruction of volume electron microscopy data.

Reference graph

Works this paper leans on

88 extracted references · 73 canonical work pages · cited by 1 Pith paper

  1. [1]

    Image super-resolution using deep convolutional net- works

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works. IEEE Trans. Pattern Analysis and Machine Intelli- gence (TPAMI), pages 295–307, 2016. 1, 3

    2016

  2. [2]

    Accurate image super-resolution using very deep convolutional net- works

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Accurate image super-resolution using very deep convolutional net- works. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 1646–1654, 2016

    2016

  3. [3]

    Enhanced deep residual networks for single image super-resolution

    Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. Enhanced deep residual networks for single image super-resolution. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition Workshop (CVPRW), pages 1132–1140, 2017

    2017

  4. [4]

    Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi

    Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew P. Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic single image super-resolution using a generative adversarial network. InProc. IEEE Conf. on Com- puter Vision and Pattern Recognition (CVPR) , pages 105– 114, 2017

    2017

  5. [5]

    Esrgan: En- hanced super-resolution generative adversarial networks

    Xintao Wang, Ke Yu, Shixiang Wu, Jinjin Gu, Yihao Liu, Chao Dong, Yu Qiao, and Chen Change Loy. Esrgan: En- hanced super-resolution generative adversarial networks. In Proc. European Conf. on Computer Vision Workshop (EC- CVW), pages 63–79, 2018

    2018

  6. [6]

    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. InProc. European Conf. on Computer Vision (ECCV), pages 294–310, 2018. 8

    2018

  7. [7]

    Residual dense network for image super-resolution

    Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, and Yun Fu. Residual dense network for image super-resolution. In Proc. IEEE Conf. on Computer Vision and Pattern Recog- nition (CVPR), pages 2472–2481, 2018. 3

    2018

  8. [8]

    Toward convolutional blind denoising of real pho- tographs

    Shi Guo, Zifei Yan, Kai Zhang, Wangmeng Zuo, and Lei Zhang. Toward convolutional blind denoising of real pho- tographs. In Proc. IEEE Conf. on Computer Vision and Pat- tern Recognition (CVPR), pages 1712–1722, 2019. 3, 5, 6

    2019

Show all 88 references
  1. [9]

    Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising

    Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, and Lei Zhang. Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising. IEEE Trans. Image Pro- cessing (TIP), pages 3142–3155, 2017. 5, 6

    2017

  2. [10]

    Learning deep CNN denoiser prior for image restoration

    Kai Zhang, Wangmeng Zuo, Shuhang Gu, and Lei Zhang. Learning deep CNN denoiser prior for image restoration. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), pages 2808–2817, 2017. 5

    2017

  3. [11]

    Ffdnet: To- ward a fast and flexible solution for cnn-based image denois- ing

    Kai Zhang, Wangmeng Zuo, and Lei Zhang. Ffdnet: To- ward a fast and flexible solution for cnn-based image denois- ing. IEEE Trans. Image Processing (TIP), pages 4608–4622,

  4. [12]

    Image de- noising using deep CNN with batch renormalization

    Chunwei Tian, Yong Xu, and Wangmeng Zuo. Image de- noising using deep CNN with batch renormalization. Neural Network, pages 461–473, 2020. 5

    2020

  5. [13]

    Plug-and-play image restoration with deep denoiser prior

    Kai Zhang, Yawei Li, Wangmeng Zuo, Lei Zhang, Luc Van Gool, and Radu Timofte. Plug-and-play image restoration with deep denoiser prior. IEEE Trans. Image Processing (TIP), pages 6360–6376, 2022. 3, 5

    2022

  6. [14]

    Deblurgan: Blind mo- tion deblurring using conditional adversarial networks

    Orest Kupyn, V olodymyr Budzan, Mykola Mykhailych, Dmytro Mishkin, and Jiri Matas. Deblurgan: Blind mo- tion deblurring using conditional adversarial networks. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), pages 8183–8192, 2018. 3

    2018

  7. [15]

    Scale-recurrent network for deep image deblurring

    Xin Tao, Hongyun Gao, Xiaoyong Shen, Jue Wang, and Ji- aya Jia. Scale-recurrent network for deep image deblurring. In Proc. IEEE Conf. on Computer Vision and Pattern Recog- nition (CVPR), pages 8174–8182, 2018. 7

    2018

  8. [16]

    Deblurgan-v2: Deblurring (orders-of-magnitude) faster and better

    Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (orders-of-magnitude) faster and better. In Proc. IEEE Int. Conf. on Computer Vi- sion (ICCV), pages 8877–8886, 2019. 7

    2019

  9. [17]

    Deblurring by realis- tic blurring

    Kaihao Zhang, Wenhan Luo, Yiran Zhong, Lin Ma, Bj ¨orn Stenger, Wei Liu, and Hongdong Li. Deblurring by realis- tic blurring. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 2734–2743, 2020. 7

    2020

  10. [18]

    Deep stacked hierarchical multi-patch network for im- age deblurring

    Hongguang Zhang, Yuchao Dai, Hongdong Li, and Piotr Ko- niusz. Deep stacked hierarchical multi-patch network for im- age deblurring. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 5978–5986, 2019. 7

    2019

  11. [19]

    Rethinking coarse-to-fine approach in sin- gle image deblurring

    Sung-Jin Cho, Seo-Won Ji, Jun-Pyo Hong, Seung-Won Jung, and Sung-Jea Ko. Rethinking coarse-to-fine approach in sin- gle image deblurring. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 4621–4630, 2021. 3, 7

    2021

  12. [20]

    Xueyang Fu, Jiabin Huang, Xinghao Ding, Yinghao Liao, and John W. Paisley. Clearing the skies: A deep network ar- chitecture for single-image rain removal. IEEE Trans. Image Processing (TIP), pages 2944–2956, 2017. 3

    2017

  13. [21]

    Semi-supervised transfer learning for image rain re- moval

    Wei Wei, Deyu Meng, Qian Zhao, Zongben Xu, and Ying Wu. Semi-supervised transfer learning for image rain re- moval. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 3877–3886, 2019

    2019

  14. [22]

    He Zhang and Vishal M. Patel. Density-aware single im- age de-raining using a multi-stream dense network. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 695–704, 2018

    2018

  15. [23]

    Rajeev Yasarla and Vishal M. Patel. Uncertainty guided multi-scale residual learning-using a cycle spinning CNN for single image de-raining. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) , pages 8405–8414, 2019

    2019

  16. [24]

    Recurrent squeeze-and-excitation context aggregation net for single image deraining

    Xia Li, Jianlong Wu, Zhouchen Lin, Hong Liu, and Hongbin Zha. Recurrent squeeze-and-excitation context aggregation net for single image deraining. In Proc. European Conf. on Computer Vision (ECCV), pages 262–277, 2018

    2018

  17. [25]

    Progressive image deraining networks: A better and simpler baseline

    Dongwei Ren, Wangmeng Zuo, Qinghua Hu, Pengfei Zhu, and Deyu Meng. Progressive image deraining networks: A better and simpler baseline. In Proc. IEEE Conf. on Com- puter Vision and Pattern Recognition (CVPR), pages 3937– 3946, 2019

    2019

  18. [26]

    Multi-scale progressive fusion network for single image deraining

    Kui Jiang, Zhongyuan Wang, Peng Yi, Chen Chen, Baojin Huang, Yimin Luo, Jiayi Ma, and Junjun Jiang. Multi-scale progressive fusion network for single image deraining. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), pages 8343–8352, 2020. 3

    2020

  19. [27]

    Single image dehazing via multi- scale convolutional neural networks

    Wenqi Ren, Si Liu, Hua Zhang, Jin-shan Pan, Xiaochun Cao, and Ming-Hsuan Yang. Single image dehazing via multi- scale convolutional neural networks. In Proc. European Conf. on Computer Vision (ECCV) , pages 154–169, 2016. 3

    2016

  20. [28]

    Dehazenet: An end-to-end system for single image haze removal

    Bolun Cai, Xiangmin Xu, Kui Jia, Chunmei Qing, and Dacheng Tao. Dehazenet: An end-to-end system for single image haze removal. IEEE Trans. Image Processing (TIP), pages 5187–5198, 2016. 7

    2016

  21. [29]

    Aod-net: All-in-one dehazing network

    Boyi Li, Xiulian Peng, Zhangyang Wang, Jizheng Xu, and Dan Feng. Aod-net: All-in-one dehazing network. In Proc. IEEE Int. Conf. on Computer Vision (ICCV) , pages 4780– 4788, 2017. 7

    2017

  22. [30]

    Grid- dehazenet: Attention-based multi-scale network for image dehazing

    Xiaohong Liu, Yongrui Ma, Zhihao Shi, and Jun Chen. Grid- dehazenet: Attention-based multi-scale network for image dehazing. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 7313–7322, 2019. 7

    2019

  23. [31]

    Multi-scale boosted de- hazing network with dense feature fusion

    Hang Dong, Jinshan Pan, Lei Xiang, Zhe Hu, Xinyi Zhang, Fei Wang, and Ming-Hsuan Yang. Multi-scale boosted de- hazing network with dense feature fusion. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 2154–2164, 2020. 7

    2020

  24. [32]

    Ffa-net: Feature fusion attention network for single image dehazing

    Xu Qin, Zhilin Wang, Yuanchao Bai, Xiaodong Xie, and Huizhu Jia. Ffa-net: Feature fusion attention network for single image dehazing. In Proc. AAAI Conf. on Artificial In- telligence (AAAI), pages 11908–11915, 2020. 7

    2020

  25. [33]

    Con- trastive learning for compact single image dehazing

    Haiyan Wu, Yanyun Qu, Shaohui Lin, Jian Zhou, Ruizhi Qiao, Zhizhong Zhang, Yuan Xie, and Lizhuang Ma. Con- trastive learning for compact single image dehazing. InProc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 10551–10560, 2021. 7

    2021

  26. [34]

    Perceiving and mod- eling density for image dehazing

    Tian Ye, Yunchen Zhang, Mingchao Jiang, Liang Chen, Yun Liu, Sixiang Chen, and Erkang Chen. Perceiving and mod- eling density for image dehazing. In Proc. European Conf. on Computer Vision (ECCV), pages 130–145, 2022. 3, 7

    2022

  27. [35]

    Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao

    Syed Waqas Zamir, Aditya Arora, Salman H. Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao. Learning enriched features for real image restora- tion and enhancement. In Proc. European Conf. on Com- puter Vision (ECCV), pages 492–511, 2020. 5, 6

    2020

  28. [36]

    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 Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 14821–14831, 2021. 5, 6, 7

    2021

  29. [37]

    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. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition Workshop (CVPRW) , pages 182– 192, 2021. 6, 7

    2021

  30. [38]

    Kuldeep Purohit, Maitreya Suin, A. N. Rajagopalan, and Vishnu Naresh Boddeti. Spatially-adaptive image restoration using distortion-guided networks. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 2289–2299, 2021. 7

    2021

  31. [39]

    Selec- tive frequency network for image restoration

    Yuning Cui, Yi Tao, Zhenshan Bing, Wenqi Ren, Xinwei Gao, Xiaochun Cao, Kai Huang, and Alois Knoll. Selec- tive frequency network for image restoration. In Proc. Int. Conf. on Learning Representations (ICLR), 2023. 1, 7

    2023

  32. [40]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  33. [41]

    Second-order attention network for single im- age super-resolution

    Tao Dai, Jianrui Cai, Yongbing Zhang, Shu-Tao Xia, and Lei Zhang. Second-order attention network for single im- age super-resolution. In Proc. IEEE Conf. on Computer Vi- sion and Pattern Recognition (CVPR), pages 11065–11074,

  34. [42]

    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. In Proc. IEEE Conf. on Computer Vi- sion and Pattern Recognition (CVPR), pages 22367–22377,

  35. [43]

    Stripformer: Strip transformer for fast image deblurring

    Fu-Jen Tsai, Yan-Tsung Peng, Yen-Yu Lin, Chung-Chi Tsai, and Chia-Wen Lin. Stripformer: Strip transformer for fast image deblurring. In Proc. European Conf. on Computer Vision (ECCV), pages 146–162, 2022. 3, 7

    2022

  36. [44]

    Image de-raining transformer

    Jie Xiao, Xueyang Fu, Aiping Liu, Feng Wu, and Zheng- Jun Zha. Image de-raining transformer. IEEE Trans. Pattern Analysis and Machine Intelligence (TPAMI) , pages 12978– 12995, 2023. 3

    2023

  37. [45]

    Learn- ing A sparse transformer network for effective image derain- ing

    Xiang Chen, Hao Li, Mingqiang Li, and Jinshan Pan. Learn- ing A sparse transformer network for effective image derain- ing. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 5896–5905, 2023. 3

    2023

  38. [46]

    Image dehazing transformer with transmission-aware 3d position embedding

    Chunle Guo, Qixin Yan, Saeed Anwar, Runmin Cong, Wenqi Ren, and Chongyi Li. Image dehazing transformer with transmission-aware 3d position embedding. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 5802–5810, 2022. 3, 7

    2022

  39. [47]

    Vision transformers for single image dehazing

    Yuda Song, Zhuqing He, Hui Qian, and Xin Du. Vision transformers for single image dehazing. IEEE Trans. Image Processing (TIP), pages 1927–1941, 2023. 3, 7

    1927

  40. [48]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. In Proc. IEEE Int. Conf. on Computer Vision Workshop (ICCVW), pages 1833–1844, 2021. 3, 5

    2021

  41. [49]

    Pre-trained image processing transformer

    Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yip- ing Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. In Proc. IEEE Conf. on Computer Vision and Pattern Recogni- tion (CVPR), pages 12299–12310, 2021. 6, 7

    2021

  42. [50]

    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. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 17662–17672, 2022. 3, 6, 7

    2022

  43. [51]

    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. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 5728–5739, 2022. 3, 5, 6, 7, 8

    2022

  44. [52]

    Accurate image restoration with attention retractable transformer

    Jiale Zhang, Yulun Zhang, Jinjin Gu, Yongbing Zhang, Linghe Kong, and Xin Yuan. Accurate image restoration with attention retractable transformer. In Proc. Int. Conf. on Learning Representations (ICLR), 2023. 1

    2023

  45. [53]

    Mamba: Linear-time sequence mod- eling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence mod- eling with selective state spaces. CoRR, abs/2312.00752,

    arXiv

  46. [54]

    Vmamba: Visual state space model

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, and Yunfan Liu. Vmamba: Visual state space model. CoRR, abs/2401.10166, 2024. 1, 2, 3, 4, 5

    2024 arXiv

  47. [55]

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

    Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang. Vision mamba: Efficient visual representation learning with bidirectional state space model. In Proc. Int. Conf. on Machine Learning (ICML) ,

  48. [56]

    Mambavision: A hybrid mamba-transformer vision backbone

    Ali Hatamizadeh and Jan Kautz. Mambavision: A hybrid mamba-transformer vision backbone. CoRR, abs/2407.08083, 2024

    2024 arXiv

  49. [57]

    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. CoRR, abs/2402.15648, 2024. 2, 3, 5, 6, 8

    2024 arXiv

  50. [58]

    Vmambair: Visual state space model for image restoration

    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. CoRR, abs/2403.11423, 2024. 1, 3, 6

    2024 arXiv

  51. [59]

    Abdelrahman Abdelhamed, Stephen Lin, and Michael S. Brown. A high-quality denoising dataset for smartphone cameras. In Proc. IEEE Conf. on Computer Vision and Pat- tern Recognition (CVPR), pages 1692–1700, 2018. 2, 5, 6, 8

    2018

  52. [60]

    Metaformer is actually what you need for vision

    Weihao Yu, Mi Luo, Pan Zhou, Chenyang Si, Yichen Zhou, Xinchao Wang, Jiashi Feng, and Shuicheng Yan. Metaformer is actually what you need for vision. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 10819–10829, 2022. 3

    2022

  53. [61]

    Toward real-world single image super-resolution: A new benchmark and a new model

    Jianrui Cai, Hui Zeng, Hongwei Yong, Zisheng Cao, and Lei Zhang. Toward real-world single image super-resolution: A new benchmark and a new model. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 3086–3095, 2019. 2, 5, 6, 8

    2019

  54. [62]

    Deep multi-scale convolutional neural network for dynamic scene deblurring

    Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 257–265, 2017. 6, 7, 8

    2017

  55. [63]

    Benchmarking single- image dehazing and beyond

    Boyi Li, Wenqi Ren, Dengpan Fu, Dacheng Tao, Dan Feng, Wenjun Zeng, and Zhangyang Wang. Benchmarking single- image dehazing and beyond. IEEE Trans. Image Processing (TIP), pages 492–505, 2019. 2, 7, 8

    2019

  56. [64]

    Acti- vating wider areas in image super-resolution

    Cheng Cheng, Hang Wang, and Hongbin Sun. Acti- vating wider areas in image super-resolution. CoRR, abs/2403.08330, 2024. 3

    2024 arXiv

  57. [65]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Inter- vention (MICCAI), pages 234–241, 2015. 4

    2015

  58. [66]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization. CoRR, abs/1607.06450, 2016. 4

    2016 arXiv

  59. [67]

    Sigmoid- weighted linear units for neural network function approxi- mation in reinforcement learning

    Stefan Elfwing, Eiji Uchibe, and Kenji Doya. Sigmoid- weighted linear units for neural network function approxi- mation in reinforcement learning. Neural Networks, pages 3–11, 2018. 4

    2018

  60. [68]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. In Proc. European Conf. on Computer Vision (ECCV), pages 17–33, 2022. 5, 8

    2022

  61. [69]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In Proc. Int. Conf. on Learning Representa- tions (ICLR), 2019. 5

    2019

  62. [70]

    com / facebookresearch / fvcore

    https : / / github . com / facebookresearch / fvcore. 5

  63. [71]

    Martin, Charless C

    David R. Martin, Charless C. Fowlkes, Doron Tal, and Jiten- dra Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 416–425, 2001. 5

    2001

  64. [72]

    Color demosaicking by local directional interpolation and nonlocal adaptive thresholding

    Lei Zhang, Xiaolin Wu, Antoni Buades, and Xin Li. Color demosaicking by local directional interpolation and nonlocal adaptive thresholding. J. Electronic Imaging, 2011. 5

    2011

  65. [73]

    Kodak lossless true color image suite

    Rich Franzen. Kodak lossless true color image suite. http: //r0k.us/graphics/kodak/, 1999. 5

    1999

  66. [74]

    Egiazarian

    Kostadin Dabov, Alessandro Foi, Vladimir Katkovnik, and Karen O. Egiazarian. Image denoising by sparse 3-d transform-domain collaborative filtering. IEEE Trans. Image Processing (TIP), pages 2080–2095, 2007. 5, 6

    2007

  67. [75]

    Real image denoising with feature attention

    Saeed Anwar and Nick Barnes. Real image denoising with feature attention. In Proc. IEEE Int. Conf. on Computer Vi- sion (ICCV), pages 3155–3164, 2019. 6

    2019

  68. [76]

    Variational denoising network: Toward blind noise modeling and removal

    Zongsheng Yue, Hongwei Yong, Qian Zhao, Deyu Meng, and Lei Zhang. Variational denoising network: Toward blind noise modeling and removal. In Proc. Conf. on Neural In- formation Processing Systems (NeurIPS), pages 1688–1699,

  69. [77]

    Spatial- adaptive network for single image denoising

    Meng Chang, Qi Li, Huajun Feng, and Zhihai Xu. Spatial- adaptive network for single image denoising. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors, Proc. European Conf. on Computer Vision (ECCV), pages 171–187, 2020. 6

    2020

  70. [78]

    Dual adversarial network: Toward real-world noise removal and noise generation

    Zongsheng Yue, Qian Zhao, Lei Zhang, and Deyu Meng. Dual adversarial network: Toward real-world noise removal and noise generation. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors, Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, U...

    2020

  71. [79]

    Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao

    Syed Waqas Zamir, Aditya Arora, Salman H. Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao. Cycleisp: Real image restoration via improved data synthesis. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 2693–2702, 2020. 6

    2020

  72. [80]

    Adaptive consistency prior based deep network for image de- noising

    Chao Ren, Xiaohai He, Chuncheng Wang, and Zhibo Zhao. Adaptive consistency prior based deep network for image de- noising. In Proc. IEEE Conf. on Computer Vision and Pat- tern Recognition (CVPR) , pages 8596–8606. Computer Vi- sion Foundation / IEEE, 2021. 6

    2021

  73. [81]

    Dynamic atten- tive graph learning for image restoration

    Chong Mou, Jian Zhang, and Zhuoyuan Wu. Dynamic atten- tive graph learning for image restoration. In Proc. IEEE Int. Conf. on Computer Vision (ICCV), pages 4308–4317. IEEE,

  74. [82]

    NTIRE 2017 chal- lenge on single image super-resolution: Dataset and study

    Eirikur Agustsson and Radu Timofte. NTIRE 2017 chal- lenge on single image super-resolution: Dataset and study. In Proc. IEEE Conf. on Computer Vision and Pattern Recog- nition Workshop (CVPRW), pages 1122–1131, 2017. 5

    2017

  75. [83]

    NTIRE 2017 challenge on sin- gle image super-resolution: Methods and results

    Radu Timofte, Eirikur Agustsson, Luc Van Gool, Ming- Hsuan Yang, and Lei Zhang. NTIRE 2017 challenge on sin- gle image super-resolution: Methods and results. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition Workshop (CVPRW), pages 1110–1121, 2017. 5

    2017

  76. [84]

    Waterloo ex- ploration database: New challenges for image quality assess- ment models

    Kede Ma, Zhengfang Duanmu, Qingbo Wu, Zhou Wang, Hongwei Yong, Hongliang Li, and Lei Zhang. Waterloo ex- ploration database: New challenges for image quality assess- ment models. IEEE Trans. Image Processing (TIP) , pages 1004–1016, 2017. 5

    2017

  77. [85]

    Bovik, and Yinxiao Li

    Zhengzhong Tu, Hossein Talebi, Han Zhang, Feng Yang, Peyman Milanfar, Alan C. Bovik, and Yinxiao Li. MAXIM: multi-axis MLP for image processing. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pages 5759–5770, 2022. 6, 7

    2022

  78. [86]

    Human-aware mo- tion deblurring

    Ziyi Shen, Wenguan Wang, Xiankai Lu, Jianbing Shen, Haibin Ling, Tingfa Xu, and Ling Shao. Human-aware mo- tion deblurring. CoRR, abs/2001.06816, 2020. 6, 7

    2001 arXiv

  79. [87]

    Sin- gle image super-resolution from transformed self-exemplars

    Jia-Bin Huang, Abhishek Singh, and Narendra Ahuja. Sin- gle image super-resolution from transformed self-exemplars. In Proc. IEEE Conf. on Computer Vision and Pattern Recog- nition (CVPR), pages 5197–5206, 2015. 8

    2015

  80. [88]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Il- lia Polosukhin. Attention is all you need. In Proc. Conf. on Neural Information Processing Systems (NeurIPS), pages 5998–6008, 2017. 8

    2017

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.