Pith. sign in

REVIEW 3 major objections 4 minor 113 references

ControlMambaIR: Conditional Controls with State-Space Model for Image Restoration

T0 review · 3 major / 4 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 ControlMambaIR: a Mamba-based condition network inside a diffusion model improves perceptual quality for deraining, deblurring, and denoising while keeping distortion metrics competitive.

desk verdict A plausible Mamba-conditioned diffusion architecture, but the central comparative claim is contradicted by the paper's own Gaussian-denoising table. read the letter →

arxiv 2506.02633 v1 pith:VBC425RQ submitted 2025-06-03 cs.CV

Cheng Yang , Lijing Liang , Zhixun Su This is my paper · ORCID
classification cs.CV
keywords imagerestorationMambanetworkdiffusionmodelconditionalcontrolderainingdeblurringdenoisingstate-space
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

ControlMambaIR is built around one idea: the Mamba state-space architecture, used as the conditional control network inside a diffusion model, provides finer-grained guidance than CNN- or attention-based condition networks for image restoration. The paper claims this design improves perceptual quality, measured by LPIPS and FID, on deraining, deblurring, and denoising benchmarks while staying competitive on the distortion metrics PSNR and SSIM. Ablations claim that directly predicting noise works better than predicting the clean image or using v-parameterization, and that Mamba outperforms attention and CNN modules in the same diffusion setup. The authors also claim data efficiency: without the Flickr2K training set, the model reaches better perceptual scores than baselines trained on more data. If correct, this points to state-space models as a cheap and effective condition backbone for diffusion-based restoration.

What carries the argument

The load-bearing object is ControlMambaIR's condition-control path: a parallel encoder that processes the low-quality image and injects its multi-scale features into the diffusion decoder through skip connections, alongside the encoder features from the noisy image. The blocks are Vision State-Space (VSS) blocks in the encoder and condition branch, Multi-Scale VSS (MSVSS) blocks in the decoder, and Temporal-Spatial Feature Interaction (TSFI) blocks that modulate features with timestep-dependent scale and shift. The decoder predicts the added noise $\epsilon_\theta(z_t,t,c_f)$ from the noisy input $z_t$, timestep $t$, and condition $c_f$. What this machinery is doing is testing whether a selective state-space model can deliver the fine-grained spatial conditioning that diffusion models lack in restoration; the paper's comparative claim is that it can, more cheaply than attention.

What would settle it

Retrain the strongest baselines, including Restormer, Uformer, and IR-SDE, on exactly the DIV2K, BSD500, and WaterlooED subset used here, with the same patch size, iteration count, and diffusion inference budget, and recompute LPIPS and FID over several seeds; the central claim fails if ControlMambaIR no longer leads on the perceptual metrics.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that substituting a Mamba-based condition network for the usual CNN- or attention-based controls in a diffusion restoration model consistently lowers LPIPS and FID across deraining, deblurring, and denoising benchmarks while preserving roughly comparable PSNR and SSIM. The supporting experiments show best-in-table perceptual scores on Rain100H, Rain100L, SIDD, and across Gaussian noise levels on CBSD68, Kodak24, and McMaster, at a reported 37G MACs. The module ablation attributes the gain to pairing diffusion with Mamba rather than with attention or no diffusion, and the prediction-target ablation identifies direct noise prediction as the best objective. The paper also asserts that this is achieved with roughly half the training data of compared denoisers because Flickr2K was not used.

Load-bearing premise

The load-bearing premise is that the comparison is fair: the numbers reported for other methods can be taken as the best those methods can do, even though they were trained on larger datasets than ControlMambaIR and with details that are not fully reported.

Editorial extensions

If this is right

  • The reported 37G MACs suggest that Mamba-based conditioning can deliver diffusion restoration at a fraction of the computational cost of transformer-based condition networks.
  • Direct noise prediction should be kept as the training objective for this architecture, since it beats image-start and v-parameterization on both distortion and perceptual metrics in the ablation.
  • The architecture transfers across deraining, deblurring, and denoising without task-specific redesign, so the conditional-control principle is not tied to one degradation type.
  • If the data-efficiency result is real, diffusion restoration with this condition network needs only about half the training data of the compared denoisers to reach better perceptual quality.

Reading between the lines

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

  • The tables imply a perceptual-fidelity versus pixel-fidelity trade-off that the text does not spell out: where ControlMambaIR wins most on LPIPS and FID, it often trails on PSNR and SSIM, so the method is a better fit for human-facing viewing than for downstream tasks that need exact pixel accuracy.
  • Because the condition branch is structurally independent of the diffusion backbone, a natural extension the paper does not run is to swap the VSS blocks for another linear-time sequence model to test whether the gain comes from selective state spaces specifically or from any cheap conditional skip-connection path.
  • A fully matched head-to-head, with the same parameter count, same training set, same iteration budget, and same seed while varying only the condition backbone, would be the cleanest test of the Mamba-versus-attention claim, since the reported module ablation changes more than one factor at a time.
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 / 4 minor

Summary. The paper proposes ControlMambaIR, a U-shaped encoder-decoder architecture in which a Mamba-based control network (ControlNet-style) provides conditional low-quality-image features to a diffusion model that predicts added Gaussian noise. The method is evaluated on deraining (Rain100H/Rain100L), deblurring (GoPro), real-image denoising (SIDD), and Gaussian denoising (CBSD68/Kodak24/McMaster). The central claims are that the method consistently surpasses existing methods in perceptual metrics (LPIPS, FID) while maintaining comparable distortion metrics (PSNR, SSIM), and that the Mamba architecture is especially well suited as a conditional control network for diffusion models. Ablation studies compare prediction targets (noise, image start, v-parameterization) and module configurations (Diffusion, Attention, Mamba).

Significance. If the comparative claims were supported, the paper would offer a useful data point on combining state-space models with diffusion-based conditional image restoration, and the reported deraining results (Tables 1–2, e.g., Rain100H PSNR 33.86 vs. IR-SDE 31.65) are interesting. The architecture is clearly described and the ablation on prediction targets is a reasonable design study. However, the paper's central claim of 'comparable' distortion performance is contradicted by its own Gaussian-denoising table, and the perceptual-superiority claim rests on an unmatched comparison protocol with no error bars, no retrained baselines, and no released code or checkpoints. As presented, the evidence does not establish the paper's headline conclusions.

major comments (3)
  1. [Abstract, Sec. 7, Sec. 6.5, Table 5] The abstract states that the method 'maintain[s] comparable performance in image distortion metrics, including PSNR and SSIM,' and Sec. 7 concludes that it 'consistently outperforms existing methods in both distortion and perceptual quality measures.' These claims are contradicted by Table 5. For Gaussian denoising on CBSD68 at sigma=15, ControlMambaIR achieves 32.17 dB / 0.903 SSIM versus SwinIR's 34.42 dB / 0.936 SSIM; at sigma=50 on the same dataset, it achieves 26.46 dB / 0.745 versus Restormer's 28.60 dB / 0.813. Gaps of roughly 2–2.5 dB PSNR and 0.03–0.07 SSIM are not 'comparable' by conventional standards in image restoration. Section 6.5 itself concedes that the method 'is lower than the state-of-art methods in terms of PNSR and SSIM scores.' The central comparative claim is therefore internally inconsistent and must be corrected before the paper's conclusions can be accepted.
  2. [Sec. 6.6, Table 6] There is a direct numerical inconsistency in the ablation on prediction targets. The text states that the 'predict noise' target achieves 'a PSNR of 39.31dB, an SSIM of 0.948,' while Table 6 reports 39.11 dB PSNR and 0.930 SSIM for the same configuration. Since the 'predict noise' row in Table 6 is also the reported SIDD result in Table 4, this discrepancy undermines confidence in the reproducibility of the quantitative results. The authors should verify all reported numbers and correct the inconsistency.
  3. [Sec. 6.5, comparison protocol] The perceptual-superiority claim on Gaussian denoising is not established by the current comparison protocol. Section 6.5 states that ControlMambaIR was trained on DIV2K, BSD500, and WaterlooED, while the cited baselines (DnCNN, SwinIR, Restormer, etc.) were trained with additional data such as Flickr2K. The baselines are not retrained under matched training data, and no error bars or multiple-seed results are reported. Consequently, the LPIPS/FID advantages in Table 5 could be driven by training-data differences, inference budgets, or evaluation randomness rather than by the proposed architecture. A matched comparison with retrained baselines and variance estimates is needed to support the claim that Mamba-based conditional control outperforms CNN- and Attention-based alternatives in this setting.
minor comments (4)
  1. [Sec. 4, Eq. (7)] The discretization formula for B appears to be incorrect: the standard ZOH formula is B = (exp(Delta*A) - I) A^{-1} B (or an approximation), not (Delta*A)^{-1}(exp(Delta*A) - I)B. Please check the derivation and notation.
  2. [Sec. 6.5 and Table 6] The dataset name is consistently misspelled as 'SSID' in the text and in the Table 6 caption; the correct name is 'SIDD.'
  3. [Sec. 6.5 and Sec. 6.3] There are repeated typos: 'PNSR' should be 'PSNR' in Sec. 6.5, and 'denlurring' should be 'deblurring' in Sec. 6.3. Also, in Sec. 1, 'detaining' appears where 'denoising' is intended.
  4. [Sec. 7, Table 7] Table 7 lists DnCNN at 37G MACs and ControlMambaIR at 37G MACs, yet the text says 'the lowest MACs of 37G... achieved by our method.' Since the table shows a tie with DnCNN, the claim should be revised to 'tied for the lowest' or the MACs should be recomputed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical architecture paper whose claims are evaluated against external benchmarks, not derived from its own assumptions.

full rationale

ControlMambaIR is a conventional empirical architecture paper: it combines a U-Net with VSS/MSVSS blocks and a ControlNet conditioned on the low-quality image, trains with an L1 noise-prediction diffusion objective, and evaluates against standard benchmarks (Rain100H/L, GoPro, SIDD, CBSD68, Kodak24, McMaster). There is no derivation chain in which an output is defined in terms of its own prediction or in which a fitted parameter is renamed as a finding. The paper's central comparative claims are supported by tables of externally computed metrics; even where the paper concedes lower distortion performance (Section 6.5: 'ControlMambaIR is lower than the state-of-art methods in terms of PNSR and SSIM scores'), that concession is a correctness/validity concern, not circularity. The conclusion that Mamba is well-suited for conditional control is an ablation result (Table 8), not an imported uniqueness theorem or ansatz smuggled in by citation. The only self-citation, Yang et al. [62], appears in related work as context and is not load-bearing for the architecture or the main claims. No quantity in Equations (1)-(20) is defined in terms of the target result, and no prediction is statistically forced by construction. Accordingly, the paper receives a circularity score of 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim is an empirical performance claim. It rests on standard diffusion and state-space mathematics imported from prior work, the assumption that LQ conditioning is sufficient, and a comparison protocol that treats published baseline numbers as directly comparable to the authors' runs. No new entities are introduced.

free parameters (3)
  • Diffusion inference step budget = 100
    Chosen in Sec. 6.1 to limit inference cost; the trade-off between perceptual quality and distortion depends on this number.
  • Training iterations = 500,000
    Sec. 6.1: fixed budget, no convergence criterion or per-task adjustment reported.
  • Architecture channel width C and per-stage block counts L1-L4 = not reported
    The method section defines C and L_i symbolically but never gives values, making the model under-specified for replication.
assumptions (4)
  • domain assumption The low-quality image cf provides sufficient conditioning for noise prediction in the reverse diffusion process.
    Central design in Sec. 5.3 and Eq. (12): decoder concatenates ControlNet features f_i^c with encoder features to predict the noise. If the LQ image were insufficient, the main mechanism would fail.
  • domain assumption LPIPS and FID are appropriate primary metrics, and PSNR/SSIM can be sacrificed for perceptual gains.
    Sec. 6.1 argues PSNR/SSIM do not align with human perception; the paper then judges success mainly by LPIPS/FID.
  • domain assumption Published baseline numbers are directly comparable to the authors' runs.
    All comparison tables use numbers from prior papers without retraining; Sec. 6.5 admits the authors used different (less) training data than several baselines.
  • standard math Standard DDPM equations (Eqs. 1-5) and SSM discretization (Eqs. 6-9) from prior work are correct and applicable.
    Secs. 3-4 restate known results; the method inherits them without new proof.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ControlMambaIR: Conditional Controls with State-Space Model for Image Restoration." pith.science (2026). https://pith.science/paper/VBC425RQ

@misc{pith2026250602633,
  author       = {Pith},
  title        = {Pith review of: ControlMambaIR: Conditional Controls with State-Space Model for Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VBC425RQ}},
  note         = {Machine review of arXiv:2506.02633}
}
read the original abstract

This paper proposes ControlMambaIR, a novel image restoration method designed to address perceptual challenges in image deraining, deblurring, and denoising tasks. By integrating the Mamba network architecture with the diffusion model, the condition network achieves refined conditional control, thereby enhancing the control and optimization of the image generation process. To evaluate the robustness and generalization capability of our method across various image degradation conditions, extensive experiments were conducted on several benchmark datasets, including Rain100H, Rain100L, GoPro, and SSID. The results demonstrate that our proposed approach consistently surpasses existing methods in perceptual quality metrics, such as LPIPS and FID, while maintaining comparable performance in image distortion metrics, including PSNR and SSIM, highlighting its effectiveness and adaptability. Notably, ablation experiments reveal that directly noise prediction in the diffusion process achieves better performance, effectively balancing noise suppression and detail preservation. Furthermore, the findings indicate that the Mamba architecture is particularly well-suited as a conditional control network for diffusion models, outperforming both CNN- and Attention-based approaches in this context. Overall, these results highlight the flexibility and effectiveness of ControlMambaIR in addressing a range of image restoration perceptual challenges.

Figures

Figures reproduced from arXiv: 2506.02633 by the authors.

Figure 1
Figure 1. Illustration of the visual process of the reverse-time image restoration [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The ControlMambaIR network is designed for image restoration tasks, [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 2
Figure 2. The overall architecture of our proposed ControlMambaIR. The [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figures from the paper (8 more)
Figure 3
Figure 3. Figure 3: Overview of the Multi-Scale Vision State-Space (MSVSS) Block. (a) [PITH_FULL_IMAGE:figures/full_fig_p017_3.png]
Figure 4
Figure 4. Figure 4: Visual results of our ControlMambaIR method and other deraining [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: Visual results of our ControlMambaIR method and other deraining [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Visual results of our ControlMambaIR method compared to other [PITH_FULL_IMAGE:figures/full_fig_p025_6.png]
Figure 7
Figure 7. Figure 7: Visual results of our ControlMambaIR method compared to other [PITH_FULL_IMAGE:figures/full_fig_p027_7.png]
Figure 8
Figure 8. Figure 8: Visual results of our ControlMambaIR method and other denoising [PITH_FULL_IMAGE:figures/full_fig_p029_8.png]
Figure 9
Figure 9. Figure 9: Visual results of our ControlMambaIR method and other denoising [PITH_FULL_IMAGE:figures/full_fig_p029_9.png]
Figure 10
Figure 10. Figure 10: Visual results of our ControlMambaIR method and other denoising [PITH_FULL_IMAGE:figures/full_fig_p030_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

113 extracted references · 49 canonical work pages

  1. [1]

    L. I. Rudin, S. Osher, E. Fatemi, Nonlinear total variation based noise removal algorithms, Physica D: Nonlinear Phenomena 60 (1992) 259– 268

    1992

  2. [2]

    W. H. Richardson, Bayesian-based iterative method of image restoration, Journal of the Optical Society of America 62 (1972) 55–59

    1972

  3. [3]

    L. B. Lucy, An iterative technique for the rectification of observed distributions, The Astronomical Journal 79 (1974) 745–754

    1974

  4. [4]

    Geman, D

    S. Geman, D. Geman, Stochastic relaxation, gibbs distributions, and the bayesian restoration of images, IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-6 (1984) 721–741

    1984

  5. [5]

    T. F. Chan, C.-K. Wong, Total variation blind deconvolution, IEEE transactions on image processing : a publication of the IEEE Signal Processing Society 7 3 (1998) 370–5. 34

    1998

  6. [6]

    L. Ma, J. Yu, T. Zeng, Sparse representation prior and total variation- based image deblurring under impulse noise, SIAM J. Imaging Sci. 6 (2013) 2258–2284

    2013

  7. [7]

    S. Rani, S. Jindal, B. Kaur, A brief review on image restoration tech- niques, International Journal of Computer Applications 150 (2016) 30–33

    2016

  8. [8]

    Huang, J

    Y. Huang, J. Huang, J. Liu, M. Yan, Y. Dong, J. Lv, S. Chen, Wavedm: Wavelet-baseddiffusionmodelsforimagerestoration, IEEETransactions on Multimedia 26 (2023) 7058–7073

    2023

Show all 113 references
  1. [9]

    Buades, B

    A. Buades, B. Coll, J.-M. Morel, A non-local algorithm for image denoising, 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05) 2 (2005) 60–65 vol. 2

    2005

  2. [10]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, ArXiv abs/1505.04597 (2015)

    2015 arXiv

  3. [11]

    Szegedy, S

    C. Szegedy, S. Ioffe, V. Vanhoucke, A. A. Alemi, Inception-v4, inception- resnet and the impact of residual connections on learning, ArXiv abs/1602.07261 (2016)

    2016 arXiv

  4. [12]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Un- terthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, An image is worth 16x16 words: Transformers for image recognition at scale, ArXiv abs/2010.11929 (2020)

    2020 arXiv

  5. [13]

    S. H. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, M. Shah, Transformers in vision: A survey, ACM Computing Surveys (CSUR) 54 (2021) 1–41

    2021

  6. [14]

    Z. Wang, X. Cun, J. Bao, J. Liu, Uformer: A general u-shaped trans- former for image restoration, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021) 17662–17672

    2021

  7. [15]

    C. Guo, Q. Yan, S. Anwar, R. Cong, W. Ren, C. Li, Image dehaz- ing transformer with transmission-aware 3d position embedding, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022) 5802–5810. 35

    2022

  8. [16]

    Zhang, W

    K. Zhang, W. Zuo, Y. Chen, D. Meng, L. Zhang, Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising, IEEE Transactions on Image Processing 26 (2016) 3142–3155

    2016

  9. [17]

    Zhang, W

    K. Zhang, W. Zuo, L. Zhang, Ffdnet: Toward a fast and flexible solution for cnn-based image denoising, IEEE Transactions on Image Processing 27 (2017) 4608–4622

    2017

  10. [18]

    Y. Pan, C. Ren, X. Wu, J. Huang, X. He, Real image denoising via guided residual estimation and noise correction, IEEE Transactions on Circuits and Systems for Video Technology 33 (2023) 1994–2000

    2023

  11. [19]

    M. Yao, D. He, X. Li, F. Li, Z. Xiong, Toward interactive self-supervised denoising, IEEE Transactions on Circuits and Systems for Video Tech- nology 33 (2023) 5360–5374

    2023

  12. [20]

    J. Xu, D. Ren, L. Zhang, D. Zhang, Patch group based bayesian learning for blind image denoising, in: ACCV Workshops, 2016

    2016

  13. [21]

    C. Dong, C. C. Loy, K. He, X. Tang, Learning a deep convolutional net- work for image super-resolution, in: European Conference on Computer Vision, 2014

    2014

  14. [22]

    C. Dong, C. C. Loy, K. He, X. Tang, Image super-resolution using deep convolutional networks, IEEE Transactions on Pattern Analysis and Machine Intelligence 38 (2014) 295–307

    2014

  15. [23]

    J. Kim, J. K. Lee, K. M. Lee, Accurate image super-resolution using very deep convolutional networks, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2015) 1646–1654

    2015

  16. [24]

    C. Wang, X. Xing, Z. Su, J. Chen, Dcsfn: Deep cross-scale fusion network for single image rain removal, Proceedings of the 28th ACM International Conference on Multimedia (2020)

    2020

  17. [25]

    C. Wang, J. shan Pan, X. Wu, Online-updated high-order collaborative networks for single image deraining, in: AAAI Conference on Artificial Intelligence, 2022

    2022

  18. [26]

    X. Cui, C. Wang, D. Ren, Y. Chen, P. Zhu, Semi-supervised image deraining using knowledge distillation, IEEE Transactions on Circuits and Systems for Video Technology 32 (2022) 8327–8341. 36

    2022

  19. [27]

    D. Ren, W. Zuo, Q. Hu, P. F. Zhu, D. Meng, Progressive image deraining networks: A better and simpler baseline, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2019) 3932–3941

    2019

  20. [28]

    D. Ren, W. Shang, P. Zhu, Q. Hu, D. Meng, W. Zuo, Single image deraining using bilateral recurrent network, IEEE Transactions on Image Processing 29 (2020) 6852–6863

    2020

  21. [29]

    D. Ren, W. Zuo, D. Zhang, L. Zhang, M.-H. Yang, Simultaneous fidelity and regularization learning for image restoration, IEEE Transactions on Pattern Analysis and Machine Intelligence 43 (2018) 284–299

    2018

  22. [30]

    L. Cai, S. Li, D. Ren, P. Wang, Dual recursive network for fast image deraining, 2019 IEEE International Conference on Image Processing (ICIP) (2019) 2756–2760

    2019

  23. [31]

    X. Xu, J. shan Pan, Y. Zhang, M.-H. Yang, Motion blur kernel esti- mation via deep learning, IEEE Transactions on Image Processing 27 (2018) 194–205

    2018

  24. [32]

    X. Tao, H. Gao, Y. Wang, X. Shen, J. Wang, J. Jia, Scale-recurrent network for deep image deblurring, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 8174–8182

    2018

  25. [33]

    Zhang, J

    J. Zhang, J. shan Pan, J. S. J. Ren, Y. Song, L. Bao, R. W. H. Lau, M.-H. Yang, Dynamic scene deblurring using spatially variant recurrent neural networks, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 2521–2529

    2018

  26. [34]

    D. Ren, W. Shang, Y. Yang, W. Zuo, Aggregating nearest sharp features via hybrid transformers for video deblurring, Information Sciences (2023)

    2023

  27. [35]

    S. W. Zamir, A. Arora, S. H. Khan, M. Hayat, F. S. Khan, M.-H. Yang, L. Shao, Multi-stage progressive image restoration, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021) 14816–14826

    2021

  28. [36]

    Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, B. Guo, Swin transformer: Hierarchical vision transformer using shifted windows, 37 2021 IEEE/CVF International Conference on Computer Vision (ICCV) (2021) 9992–10002

    2021

  29. [37]

    Liang, J

    J. Liang, J. Cao, G. Sun, K. Zhang, L. V. Gool, R. Timofte, Swinir: Imagerestorationusingswintransformer, 2021IEEE/CVFInternational Conference on Computer Vision Workshops (ICCVW) (2021) 1833–1844

    2021

  30. [38]

    S. W. Zamir, A. Arora, S. H. Khan, M. Hayat, F. S. Khan, M.-H. Yang, Restormer: Efficient transformer for high-resolution image restoration, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR) (2021) 5718–5729

    2021

  31. [39]

    S. Chen, T. Ye, Y. Liu, E. Chen, Dual-former: Hybrid self-attention transformer for efficient image restoration, ArXiv abs/2210.01069 (2022)

    2022 arXiv

  32. [40]

    Z. Chen, Y. Zhang, J. Gu, Y. Zhang, L. Kong, X. Yuan, Cross ag- gregation transformer for image restoration, ArXiv abs/2211.13654 (2022)

    2022 arXiv

  33. [41]

    Arjovsky, S

    M. Arjovsky, S. Chintala, L. Bottou, Wasserstein gan, ArXiv abs/1701.07875 (2017)

    2017 arXiv

  34. [42]

    L. Dinh, D. Krueger, Y. Bengio, Nice: Non-linear independent compo- nents estimation, arXiv: Learning (2014)

    2014

  35. [43]

    Ledig, L

    C. Ledig, L. Theis, F. Huszár, J. Caballero, A. P. Aitken, A. Tejani, J. Totz, Z. Wang, W. Shi, Photo-realistic single image super-resolution using a generative adversarial network, 2017 IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR) (2016) 105–114

    2016

  36. [44]

    Z. Chen, Z. Zeng, H. lan Shen, X. Zheng, P. Dai, P. Ouyang, Dn-gan: Denoising generative adversarial networks for speckle noise reduction in optical coherence tomography images, Biomed. Signal Process. Control. 55 (2020)

    2020

  37. [45]

    J. Yu, Z. L. Lin, J. Yang, X. Shen, X. Lu, T. S. Huang, Generative image inpainting with contextual attention, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 5505–5514. 38

    2018

  38. [46]

    Kupyn, V

    O. Kupyn, V. Budzan, M. Mykhailych, D. Mishkin, J. Matas, Deblurgan: Blind motion deblurring using conditional adversarial networks, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2017) 8183–8192

    2017

  39. [47]

    L. Dinh, J. N. Sohl-Dickstein, S. Bengio, Density estimation using real nvp, ArXiv abs/1605.08803 (2016)

    2016 arXiv

  40. [48]

    D. P. Kingma, P. Dhariwal, Glow: Generative flow with invertible 1x1 convolutions, ArXiv abs/1807.03039 (2018)

    2018 arXiv

  41. [49]

    Y. Kim, D. Son, Noise conditional flow model for learning the super- resolution space, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (2021) 424–432

    2021

  42. [50]

    X. Wei, H. V. Gorp, L. Gonzalez-Carabarin, D. Freedman, Y. C. Eldar, R. J. G. van Sloun, Image denoising with deep unfolding and normalizing flows, ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (2022) 1551–1555

    2022

  43. [51]

    Y. Ren, X. Yu, R. Zhang, T. H. Li, S. Liu, G. Li, Structureflow: Image inpainting via structure-aware appearance flow, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 181–190

    2019

  44. [52]

    Z. Cai, J. Tang, S. Mukherjee, J. Li, C.-B. Schonlieb, X. Zhang, Nf-ula: Langevin monte carlo with normalizing flow prior for imaging inverse problems, ArXiv abs/2304.08342 (2023)

    2023 arXiv

  45. [53]

    J. Ho, A. Jain, P. Abbeel, Denoising diffusion probabilistic models, ArXiv abs/2006.11239 (2020)

    2020 arXiv

  46. [54]

    Dhariwal, A

    P. Dhariwal, A. Nichol, Diffusion models beat gans on image synthesis, ArXiv abs/2105.05233 (2021)

    2021 arXiv

  47. [55]

    Y. Song, J. N. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, B.Poole, Score-basedgenerativemodelingthroughstochasticdifferential equations, ArXiv abs/2011.13456 (2020)

    2020 arXiv

  48. [56]

    J. Song, C. Meng, S. Ermon, Denoising diffusion implicit models, ArXiv abs/2010.02502 (2020). 39

    2020 arXiv

  49. [57]

    R. S. Roman, E. Nachmani, L. Wolf, Noise estimation for generative diffusion models, ArXiv abs/2104.02600 (2021)

    2021 arXiv

  50. [58]

    Bansal, E

    A. Bansal, E. Borgnia, H.-M. Chu, J. Li, H. Kazemi, F. Huang, M. Gold- blum, J. Geiping, T. Goldstein, Cold diffusion: Inverting arbitrary image transforms without noise, ArXiv abs/2208.09392 (2022)

    2022 arXiv

  51. [59]

    S. Chen, P. Sun, Y. Song, P. Luo, Diffusiondet: Diffusion model for ob- ject detection, 2023 IEEE/CVF International Conference on Computer Vision (ICCV) (2022) 19773–19786

    2022

  52. [60]

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sjölund, T. B. Schön, Image restoration with mean-reverting stochastic differential equations, in: International Conference on Machine Learning, 2023

    2023

  53. [61]

    J. Wu, H. Wu, G. Yuan, Detail-aware image denoising via structure preserved network and residual diffusion model, The Visual Computer (2024)

    2024

  54. [62]

    C. Yang, C. Wang, L. Liang, Z. Su, Real-world image denoising via efficient diffusion model with controllable noise generation, J. Electronic Imaging 33 (2024)

    2024

  55. [63]

    B. Xia, Y. Zhang, S. Wang, Y. Wang, X. Wu, Y. Tian, W. Yang, L. V. Gool, Diffir: Efficient diffusion model for image restoration, 2023 IEEE/CVF International Conference on Computer Vision (ICCV) (2023) 13049–13059

    2023

  56. [64]

    Z. Yue, J. Wang, C. C. Loy, Efficient diffusion model for image restora- tion by residual shifting, IEEE Transactions on Pattern Analysis and Machine Intelligence 47 (2024) 116–130

    2024

  57. [65]

    J. Song, D. Huang, X. Huang, M. Ruan, H. Zeng, Torch-advent- civilization-evolution: Accelerating diffusion model for image restora- tion, IEEE Transactions on Circuits and Systems for Video Technology (2024)

    2024

  58. [66]

    R. Wu, L. Sun, Z. Ma, L. Zhang, One-step effective diffusion network for real-world image super-resolution, ArXiv abs/2406.08177 (2024). 40

    2024 arXiv

  59. [67]

    Zheng, X.-M

    D. Zheng, X.-M. Wu, S. Yang, J. Zhang, J. Hu, W.-S. Zheng, Selective hourglass mapping for universal image restoration based on diffusion model, 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2024) 25445–25455

    2024

  60. [68]

    A. Gu, K. Goel, C. R’e, Efficiently modeling long sequences with structured state spaces, ArXiv abs/2111.00396 (2021)

    2021 arXiv

  61. [69]

    A. Gu, I. Johnson, K. Goel, K. K. Saab, T. Dao, A. Rudra, C. R’e, Combining recurrent, convolutional, and continuous-time models with linear state-space layers, in: Neural Information Processing Systems, 2021

    2021

  62. [70]

    Smith, A

    J. Smith, A. Warrington, S. W. Linderman, Simplified state space layers for sequence modeling, ArXiv abs/2208.04933 (2022)

    2022 arXiv

  63. [71]

    A new approach to linear filtering and prediction problems, 2002

    2002

  64. [72]

    T. Dao, D. Y. Fu, K. K. Saab, A. W. Thomas, A. Rudra, C. Ré, Hungry hungry hippos: Towards language modeling with state space models, ArXiv abs/2212.14052 (2022)

    2022 arXiv

  65. [73]

    Mehta, A

    H. Mehta, A. Gupta, A. Cutkosky, B. Neyshabur, Long range language modeling via gated state spaces, ArXiv abs/2206.13947 (2022)

    2022 arXiv

  66. [74]

    Soydan, N

    T. Soydan, N. Zubic, N. Messikommer, S. Mishra, D. Scaramuzza, S7: Selective and simplified state space layers for sequence modeling, ArXiv abs/2410.03464 (2024)

    2024 arXiv

  67. [75]

    A. Gu, T. Dao, Mamba: Linear-time sequence modeling with selective state spaces, ArXiv abs/2312.00752 (2023)

    2023 arXiv

  68. [76]

    Y. Liu, Y. Tian, Y. Zhao, H. Yu, L. Xie, Y. Wang, Q. Ye, Y. Liu, Vmamba: Visual state space model, ArXiv abs/2401.10166 (2024)

    2024 arXiv

  69. [77]

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, X. Wang, Vision mamba: Efficient visual representation learning with bidirectional state space model, ArXiv abs/2401.09417 (2024)

    2024 arXiv

  70. [78]

    Y. Yue, Z. Li, Medmamba: Vision mamba for medical image classifica- tion, ArXiv abs/2403.03849 (2024). 41

    2024 arXiv

  71. [79]

    Nasiri-Sarvi, M

    A. Nasiri-Sarvi, M. S. Hosseini, H. Rivaz, Vision mamba for classification of breast ultrasound images, ArXiv abs/2407.03552 (2024)

    2024 arXiv

  72. [80]

    J. Ruan, S. Xiang, Vm-unet: Vision mamba unet for medical image segmentation, ArXiv abs/2402.02491 (2024)

    2024 arXiv

  73. [81]

    Y. Yang, Z. Xing, L. Zhu, Vivim: a video vision mamba for medical video segmentation, 2024

    2024

  74. [82]

    R. Wu, Y. Liu, P. Liang, Q. Chang, H-vmunet: High-order vision mamba unet for medical image segmentation, ArXiv abs/2403.13642 (2024)

    2024 arXiv

  75. [83]

    J. Wang, J. Chen, D. Z. Chen, J. Wu, Lkm-unet: Large kernel vision mamba unet for medical image segmentation, 2024

    2024

  76. [84]

    Zheng, C

    Z. Zheng, C. Wu, U-shaped vision mamba for single image dehazing, ArXiv abs/2402.04139 (2024)

    2024 arXiv

  77. [85]

    H. Zhou, X. Wu, H. Chen, X. Chen, X. He, Rsdehamba: Lightweight vision mamba for remote sensing satellite image dehazing, ArXiv abs/2405.10030 (2024)

    2024 arXiv

  78. [86]

    R. Deng, T. Gu, Cu-mamba: Selective state space models with channel learning for image restoration, 2024 IEEE 7th International Conference on Multimedia Information Processing and Retrieval (MIPR) (2024) 328–334

    2024

  79. [87]

    H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, S.-T. Xia, Mambair: A simple baseline for image restoration with state-space model, in: European Conference on Computer Vision, 2024

    2024

  80. [88]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, CoRR abs/1412.6980 (2014)

    2014 arXiv

  81. [89]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Sgdr: Stochastic gradient descent with warm restarts, arXiv: Learning (2016)

    2016

  82. [90]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, O. Wang, The un- reasonable effectiveness of deep features as a perceptual metric, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 586–595. 42

    2018

  83. [91]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, S. Hochreiter, Gans trained by a two time-scale update rule converge to a local nash equilibrium, in: Neural Information Processing Systems, 2017

    2017

  84. [92]

    Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli, Image quality assessment: from error visibility to structural similarity, IEEE Transac- tions on Image Processing 13 (2004) 600–612

    2004

  85. [93]

    Menon, A

    S. Menon, A. Damian, S. Hu, N. Ravi, C. Rudin, Pulse: Self-supervised photoupsamplingvialatentspaceexplorationofgenerativemodels, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020) 2434–2442

    2020

  86. [94]

    Z. Wang, A. C. Bovik, Mean squared error: Love it or leave it? a new look at signal fidelity measures, IEEE Signal Processing Magazine 26 (2009) 98–117

    2009

  87. [95]

    W. Yang, R. T. Tan, J. Feng, J. Liu, Z. Guo, S. Yan, Deep joint rain detection and removal from a single image, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016) 1685–1694

    2016

  88. [96]

    W. Yang, R. T. Tan, J. Feng, Z. Guo, S. Yan, J. Liu, Joint rain detection and removal from a single image with contextualized deep networks, IEEE Transactions on Pattern Analysis and Machine Intelligence 42 (2020) 1377–1393

    2020

  89. [97]

    Z. Tu, H. Talebi, H. Zhang, F. Yang, P. Milanfar, A. C. Bovik, Y. Li, Maxim: Multi-axis mlp for image processing, 2022 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR) (2022) 5759–5770

    2022

  90. [98]

    S. Nah, T. H. Kim, K. M. Lee, Deep multi-scale convolutional neu- ral network for dynamic scene deblurring, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016) 257–265

    2016

  91. [99]

    Kupyn, T

    O. Kupyn, T. Martyniuk, J. Wu, Z. Wang, Deblurgan-v2: Deblurring (orders-of-magnitude) faster and better, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 8877–8886. 43

    2019

  92. [100]

    Zhang, W

    K. Zhang, W. Luo, Y. Zhong, L. Ma, B. Stenger, W. Liu, H. Li, Deblur- ring by realistic blurring, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020) 2734–2743

    2020

  93. [101]

    Anwar, N

    S. Anwar, N. Barnes, Real image denoising with feature attention, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 3155–3164

    2019

  94. [102]

    Z. Yue, Q. Zhao, L. Zhang, D. Meng, Dual adversarial network: Toward real-world noise removal and noise generation, ArXiv abs/2007.05946 (2020)

    2020 arXiv

  95. [103]

    S. W. Zamir, A. Arora, S. H. Khan, M. Hayat, F. S. Khan, M.-H. Yang, L. Shao, Cycleisp: Real image restoration via improved data synthesis, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020) 2693–2702

    2020

  96. [104]

    Abdelhamed, S

    A. Abdelhamed, S. Lin, M. S. Brown, A high-quality denoising dataset for smartphone cameras, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 1692–1700

    2018

  97. [105]

    Agustsson, R

    E. Agustsson, R. Timofte, Ntire 2017 challenge on single image super- resolution: Dataset and study, 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (2017) 1122– 1131

    2017

  98. [106]

    Arbeláez, M

    P. Arbeláez, M. Maire, C. C. Fowlkes, J. Malik, Contour detection and hierarchical image segmentation, IEEE Transactions on Pattern Analysis and Machine Intelligence 33 (2011) 898–916

    2011

  99. [107]

    K. Ma, Z. Duanmu, Q. Wu, Z. Wang, H. Yong, H. Li, L. Zhang, Waterloo exploration database: New challenges for image quality assessment models, IEEE Transactions on Image Processing 26 (2017) 1004–1016

    2017

  100. [108]

    D. R. Martin, C. C. Fowlkes, D. Tal, J. Malik, A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics, Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001 2 (2001) 41...

    2001

  101. [109]

    Franzen, Kodak lossless true color image suite: Photocd pcd0992

    R. Franzen, Kodak lossless true color image suite: Photocd pcd0992. 44

  102. [110]

    Zhang, X

    L. Zhang, X. Wu, A. Buades, X. Li, Color demosaicking by local direc- tional interpolation and nonlocal adaptive thresholding, J. Electronic Imaging 20 (2011) 023016

    2011

  103. [111]

    Zhang, W

    K. Zhang, W. Zuo, S. Gu, L. Zhang, Learning deep cnn denoiser prior for image restoration, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017) 2808–2817

    2017

  104. [112]

    C. Tian, Y. Xu, Z. Li, W. Zuo, L. Fei, H. Liu, Attention-guided cnn for image denoising, Neural networks : the official journal of the International Neural Network Society 124 (2020) 117–129

    2020

  105. [113]

    Salimans, J

    T. Salimans, J. Ho, Progressive distillation for fast sampling of diffusion models (2022). arXiv:2202.00512. 45

    2022 arXiv

Pith tools

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