Viscosity Stabilized Plug-and-Play Reconstruction

Arghya Sinha; Kunal N. Chaudhury; Trishit Mukherjee

REVIEW 3 major objections 1 minor 58 references

Viscosity Stabilized Plug-and-Play Reconstruction

T0 review · 3 major / 1 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Plug-and-play reconstruction can be stabilized by adaptively blending the denoiser with a contractive operator.

desk verdict Plausible PnP stabilization idea; the full text I received is corrupted, so the math and experiments are unverified – worth sending to review if the actual PDF is intact. read the letter →

arxiv 2508.01441 v1 pith:KQJLKAW3 submitted 2025-08-02 eess.IV

Arghya Sinha , Trishit Mukherjee , Kunal N. Chaudhury This is my paper · ORCID

classification eess.IV MSC 68U1094A08

keywords plug-and-playreconstructionimagedeepdenoiserviscosityregularizationcontractiveoperatorproximalalgorithmstability

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

Plug-and-play (PnP) reconstruction replaces half of a proximal algorithm with a pretrained denoiser, which lets one denoiser serve many imaging tasks. This paper identifies a common failure mode: the iterates improve for a while, then oscillate and drift off, because black-box networks need not make the iteration contractive. The proposed fix is a data-driven viscosity schedule that blends the PnP update with a contractive image-reconstruction operator, with the contractive term taking over as iterations proceed. The paper's claim is that this stabilizes late iterations across different proximal algorithms, denoiser architectures (CNNs, diffusion models, transformers), and imaging tasks, so off-the-shelf denoisers can be used without restrictive constraints.

What carries the argument

The central object is the viscosity-stabilized operator $V_k = (1-\lambda_k)T + \lambda_k C$, a convex combination of the potentially noncontractive PnP operator $T$ and a contractive operator $C$ from the image-reconstruction model. The coefficient $\lambda_k \in [0,1]$ is data-driven and increases with the iteration index, so early iterates keep the denoiser's full effect while late iterates lean on the contractive map. The mechanism is what carries the argument: the contractive term prevents the iteration from running away.

What would settle it

On a fixed linear inverse problem (for example, Gaussian deblurring with a known noise level), run the same pretrained denoiser in PnP with and without the viscosity schedule over many iterations; the central claim is falsified if the stabilized sequence does not converge to a fixed point, or if its final PSNR is no better than the best iterate of the unstabilized run.

Watch

Extended reading notes

Core claim

The authors propose viscosity-stabilized PnP: the standard update $x_{k+1} = T(x_k)$ formed with a pretrained denoiser is replaced by $x_{k+1} = (1-\lambda_k) T(x_k) + \lambda_k C(x_k)$, where $C$ is a contractive operator tied to the imaging model and $\lambda_k$ is an adaptive, data-driven coefficient that grows with $k$. The contractive component acts as viscosity, damping oscillations in later iterations so that the PSNR no longer peaks early and then degrades. The authors maintain that this is a general stabilization mechanism, not a new denoiser, and they validate it across proximal algorithms, denoising architectures, and imaging tasks.

Load-bearing premise

The load-bearing premise is that a contractive operator (a map that always shrinks distances) exists for the specific imaging task and that the data-driven blending schedule transfers beyond the training setting; if that premise fails, the stabilization will either over-smooth the output or fail to prevent divergence.

Editorial extensions

If this is right

Freely trained single-step denoisers can be dropped into PnP pipelines without architectural changes or retraining, since stability is supplied by the schedule rather than by the denoiser.
The late-iteration PSNR and visual-quality drop found across CNN, diffusion, and transformer denoisers should be suppressed or eliminated.
The same stabilization recipe transfers to different proximal algorithms and different imaging tasks, so PnP becomes more portable across applications.
The schedule provides a built-in trade-off between data consistency and regularization, with the contractive operator dominating at convergence.

Reading between the lines

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

One extension the authors do not pursue is proving convergence: a natural next step is to ask whether the viscosity schedule can be tuned to make the composed operator contractive in a norm, which would turn the empirical stability into a guarantee.
Because the mechanism treats the denoiser as a black box, it should apply to very large pretrained or foundation-model denoisers whose internal structure cannot be constrained; testing that is a direct extrapolation of the paper's validation strategy.
The idea could be carried over to other fixed-point algorithms outside imaging, such as deep equilibrium models, wherever late-iteration divergence is the practical bottleneck.

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.

Desk Editor's Note

read the letter

First, the important thing: the full text you sent me is corrupted – mojibake – so I can only judge the abstract. If the actual arXiv PDF is intact, the assessment below is conditional.

What the paper is trying to do: PnP reconstruction is known to degrade after early iterations with black-box denoisers. The authors propose to adaptively average the PnP operator with a contractive inversion (IR) operator, calling it viscosity regularization. That is a reasonable and simple idea. It is architecture-agnostic and would be useful if it works. The claim that it suppresses oscillations and prevents divergence is exactly the sort of thing the community wants.

I can't verify any of the specifics. No equations, no convergence proof, no experimental numbers appear in what I could read. The stress-test raises the right questions: for the averaged operator to be contractive, the IR operator needs a known controllable contraction constant; and the adaptive schedule for the averaging coefficient needs to be specified. If the coefficient goes to 1, you converge to the IR fixed point and the denoiser asymptotically disappears; if it doesn't, you need a different convergence argument. The abstract doesn't answer these. That doesn't mean the paper is wrong; it means the abstract alone isn't enough to judge.

What the paper does well, based on the abstract: it identifies a real problem, proposes a simple fix, and says the fix was tested across several proximal algorithms, denoiser architectures, and imaging tasks. That's the right kind of empirical scope if the numbers back it up.

My overall take: this is a plausible incremental contribution to a recognized subfield problem. If the full paper contains the derivation of the schedule and contraction analysis, plus honest comparisons, it's a solid submission. If not, it's just a heuristic with some promising pictures. I can't tell which from what I have. The copy I saw is unusable, but I wouldn't desk-reject a paper just because the provided extraction failed. I'd ask the authors for a clean PDF and send it to a referee who knows PnP theory.

Recommendation: yes, send it to peer review, with a request to check the convergence proof and the experimental baselines.

Referee Report

3 major / 1 minor

Summary. The manuscript proposes a viscosity-stabilized plug-and-play (PnP) reconstruction method. The central idea is to adaptively average the PnP update operator with a contractive image-reconstruction (IR) operator, so that the contractive component increasingly dampens updates in later iterations. The abstract claims this suppresses oscillations, prevents divergence, and is validated across proximal algorithms, denoising architectures, and imaging tasks. However, the provided full text is almost entirely garbled and unreadable; no equations, algorithm descriptions, proofs, experimental setups, or numerical results are accessible. Thus the technical content and evidence for the central claims cannot be assessed from the available manuscript.

Significance. If the claimed mechanism works, it would be practically valuable: it could allow freely trained deep denoisers to be used in PnP algorithms without restrictive conditions on the denoiser, potentially stabilizing a wide class of iterators. The idea of blending with a contractive operator as a form of viscosity regularization is conceptually appealing and may merit attention. However, the manuscript as provided contains no verifiable derivation, no specification of the adaptive schedule or the contractive IR operator, and no numerical evidence. The potential significance is therefore entirely conditional on content that is not present in the available file.

major comments (3)

[Full Text] The full text is corrupted to the point of unreadability: almost all sentences are garbled, and no equations, theorem statements, algorithm boxes, or experimental tables can be recovered. This is the most serious issue: the central claims of the paper cannot be checked against any technical content. The authors must provide a clean, readable version before any substantive review can occur.
[Abstract] The abstract asserts validation 'across different proximal algorithms, denoising architectures, and imaging tasks', but the visible text contains no numbers, baselines, error bars, or even the names of the algorithms, architectures, and tasks. This claim is unsupported by the available evidence and must be substantiated with concrete experimental reporting.
[Abstract] The stabilization mechanism relies on two unspecified components: a contractive IR operator for the given imaging task and a data-driven adaptive schedule for the averaging coefficient. Neither the construction of this operator nor the form of the schedule is given in the visible text. Without a schedule formula and a contraction analysis, the claim that divergence is prevented cannot be evaluated; if the averaging coefficient does not tend to zero sufficiently fast, the averaged operator may remain expansive.

minor comments (1)

[Abstract] A brief statement of the reported quantitative improvement (e.g., PSNR gain or iteration count) would make the abstract more informative, though this is secondary to the missing full text.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identifiable from the available abstract; the proposed viscosity-stabilized PnP averaging scheme is a new algorithmic construction rather than a restatement of its inputs.

full rationale

The only readable portion of the manuscript is the abstract, which describes a data-driven stabilization mechanism that adaptively averages a potentially unstable plug-and-play operator with a contractive image-reconstruction operator. This is presented as a new algorithmic construction, not as a prediction derived from fitted parameters or from prior results by the same authors. There is no visible equation, fitted parameter, or self-citation chain that makes the claimed stabilization equivalent to its inputs by construction. The abstract explicitly contrasts the method with prior stability approaches that impose restrictive constraints on the denoiser, and the contribution is the adaptive averaging mechanism itself. Because the full text is unreadable in the provided material, no specific circular step can be quoted or exhibited, and per the hard rules, circularity must not be inferred from absence of detail. The correct finding is therefore no significant circularity: the available evidence does not show any load-bearing step reducing to its own inputs, any fitted input being renamed as a prediction, or any self-citation being used as the sole justification for the central claim.

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

Only the abstract is readable. The adaptive coefficient is a free parameter, and the availability of a contractive operator is a domain assumption. No new physical entities are introduced.

free parameters (1)

adaptive averaging coefficient
The abstract says the averaging is adaptive, implying a learned or scheduled coefficient. Its selection rule and values are not specified anywhere in the abstract.

assumptions (1)

domain assumption A contractive IR operator is available for every considered imaging task.
The method relies on averaging the PnP operator with a contractive IR operator, but the abstract does not explain how such an operator is constructed or chosen for each task.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Viscosity Stabilized Plug-and-Play Reconstruction." pith.science (2026). https://pith.science/paper/KQJLKAW3

@misc{pith2026250801441,
  author       = {Pith},
  title        = {Pith review of: Viscosity Stabilized Plug-and-Play Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQJLKAW3}},
  note         = {Machine review of arXiv:2508.01441}
}

read the original abstract

The plug-and-play (PnP) method uses a deep denoiser within a proximal algorithm for model-based image reconstruction (IR). Unlike end-to-end IR, PnP allows the same pretrained denoiser to be used across different imaging tasks, without the need for retraining. However, black-box networks can make the iterative process in PnP unstable. A common issue observed across architectures like CNNs, diffusion models, and transformers is that the visual quality and PSNR often improve initially but then degrade in later iterations. Previous attempts to ensure stability usually impose restrictive constraints on the denoiser. However, standard denoisers, which are freely trained for single-step noise removal, need not satisfy such constraints. We propose a simple data-driven stabilization mechanism that adaptively averages the potentially unstable PnP operator with a contractive IR operator. This acts as a form of viscosity regularization, where the contractive component progressively dampens updates in later iterations, helping to suppress oscillations and prevent divergence. We validate the effectiveness of our stabilization mechanism across different proximal algorithms, denoising architectures, and imaging tasks.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages

[1]

11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEauthordefaulttextstyle \@IEEEcompsocnotconfonly \@IEEEauthor...

work page
[2]

C. A. Bouman, Foundations of Computational Imaging: A Model-Based Approach. 1em plus 0.5em minus 0.4em SIAM, 2022

work page 2022
[3]

L. I. Rudin, S. Osher, and E. Fatemi, ``Nonlinear total variation based noise removal algorithms,'' Phys. D: Nonlinear Phenom., vol. 60, no. 1-4, pp. 259--268, 1992

work page 1992
[4]

Geman and C

S. Geman and C. Graffigne, ``Markov R andom F ield image models and their applications to computer vision,'' Proc. ICM, vol. 1, p. 2, 1986

work page 1986
[5]

K. H. Jin, M. T. McCann, E. Froustey, and M. Unser, ``Deep convolutional neural network for inverse problems in imaging,'' IEEE Trans. Image Process., vol. 26, no. 9, pp. 4509--4522, 2017

work page 2017
[6]

H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, and S.-T. Xia, `` MambaIR : A simple baseline for image restoration with state-space model,'' Proc. ECCV, pp. 222--241, 2024

work page 2024
[7]

Liang, J

J. Liang, J. Cao, G. Sun, K. Zhang, L. Van Gool, and R. Timofte, `` SwinIR : Image restoration using swin transformer,'' Proc. ICCV Workshops, pp. 1833--1844, 2021

work page 2021
[8]

S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang, ``Restormer: E fficient transformer for high-resolution image restoration,'' Proc. CVPR, pp. 5718--5739, 2022

work page 2022

Show all 58 references

[9]

S. V. Venkatakrishnan, C. A. Bouman, and B. Wohlberg, ``Plug-and- Play priors for model based reconstruction,'' Proc. IEEE GlobalSIP, pp. 945--948, 2013

2013
[10]

Sreehari, S

S. Sreehari, S. V. Venkatakrishnan, B. Wohlberg, G. T. Buzzard, L. F. Drummy, J. P. Simmons, and C. A. Bouman, ``Plug-and-play priors for bright field electron tomography and sparse interpolation,'' IEEE Trans. Comput. Imaging, vol. 2, no. 4, pp. 408--423, 2016

2016
[11]

Romano, M

Y. Romano, M. Elad, and P. Milanfar, ``The little engine that could: Regularization by Denoising ( RED ),'' SIAM J. Imaging Sci., vol. 10, no. 4, pp. 1804--1844, 2017

2017
[12]

Zhang, Y

K. Zhang, Y. Li, W. Zuo, L. Zhang, L. Van Gool, and R. Timofte, ``Plug-and-play image restoration with deep denoiser prior,'' IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 10, pp. 6360--6376, 2021

2021
[13]

Hurault, A

S. Hurault, A. Leclaire, and N. Papadakis, ``Gradient step denoiser for convergent plug-and-play,'' Proc. ICLR, 2022

2022
[14]

Chung, J

H. Chung, J. Kim, M. T. Mccann, M. L. Klasky, and J. C. Ye, ``Diffusion posterior sampling for general noisy inverse problems,'' Proc. ICLR, 2022

2022
[15]

Kawar, M

B. Kawar, M. Elad, S. Ermon, and J. Song, ``Denoising diffusion restoration models,'' Proc. NeurIPS, pp. 23\,593--23\,606, 2022

2022
[16]

Y. Zhu, K. Zhang, J. Liang, J. Cao, B. Wen, R. Timofte, and L. V. Gool, ``Denoising diffusion models for plug-and-play image restoration,'' Proc. CVPR Workshops, pp. 1219--1229, 2023

2023
[17]

Lugmayr, M

A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool, `` RePaint : Inpainting using denoising diffusion probabilistic models,'' Proc. CVPR, pp. 11\,451--11\,461, 2022

2022
[18]

M. Elad, B. Kawar, and G. Vaksman, ``Image denoising: The deep learning revolution and beyond—a survey paper,'' SIAM J. Imaging Sci., vol. 16, no. 3, pp. 1594--1654, 2023

2023
[19]

X. Yuan, Y. Liu, J. Suo, and Q. Dai, ``Plug-and-play algorithms for large-scale snapshot compressive imaging,'' Proc. CVPR, pp. 1444--1457, 2020

2020
[20]

K. Wei, A. Aviles-Rivero, J. Liang, Y. Fu, H. Huang, and C.-B. Sch\" o nlieb, ``Tuning-free plug-and-play proximal algorithm for inverse imaging problems,'' Proc. ICML, pp. 10\,158--10\,169, 2020

2020
[21]

U. S. Kamilov, C. A. Bouman, G. T. Buzzard, and B. Wohlberg, ``Plug-and-play methods for integrating physical and learned models in computational imaging: Theory, algorithms, and applications,'' IEEE Signal Process. Mag., vol. 40, no. 1, pp. 85--97, 2023

2023
[22]

Ahmad, C

R. Ahmad, C. A. Bouman, G. T. Buzzard, S. Chan, S. Liu, E. T. Reehorst, and P. Schniter, ``Plug-and-play methods for magnetic resonance imaging: Using denoisers for image recovery,'' IEEE Signal Process. Mag., vol. 37, no. 1, pp. 105--116, 2020

2020
[23]

Parikh and S

N. Parikh and S. Boyd, Proximal algorithms. 1em plus 0.5em minus 0.4em Now Publishers, Inc., 2014

2014
[24]

H. H. Bauschke and P. L. Combettes, Convex Analysis and Monotone Operator Theory in Hilbert Spaces. 1em plus 0.5em minus 0.4em Springer, 2011

2011
[25]

Beck, First-Order Methods in Optimization

A. Beck, First-Order Methods in Optimization. 1em plus 0.5em minus 0.4em SIAM , 2017

2017
[26]

Geman and C

D. Geman and C. Yang, ``Nonlinear image recovery with half-quadratic regularization,'' IEEE Trans. Image Process., vol. 4, no. 7, pp. 932--946, 1995

1995
[27]

Pesquet, A

J.-C. Pesquet, A. Repetti, M. Terris, and Y. Wiaux, ``Learning maximally monotone operators for image recovery,'' SIAM J. Imaging Sci., vol. 14, no. 3, pp. 1206--1237, 2021

2021
[28]

Hertrich, S

J. Hertrich, S. Neumayer, and G. Steidl, ``Convolutional proximal neural networks and plug-and-play algorithms,'' Linear Algebra Appl., vol. 631, pp. 203--234, 2021

2021
[29]

Cohen, M

R. Cohen, M. Elad, and P. Milanfar, ``Regularization by denoising via fixed-point projection ( RED - PRO ),'' SIAM J. Imaging Sci., vol. 14, no. 3, pp. 1374--1406, 2021

2021
[30]

Hurault, A

S. Hurault, A. Leclaire, and N. Papadakis, ``Proximal denoiser for convergent plug-and-play optimization with nonconvex regularization,'' Proc. ICML, pp. 9483--9505, 2022

2022
[31]

Nair and K

P. Nair and K. N. Chaudhury, ``Averaged deep denoisers for image regularization,'' J. Math. Imaging Vis., vol. 66, no. 3, p. 362–379, 2024

2024
[32]

Goujon, S

A. Goujon, S. Neumayer, P. Bohra, S. Ducotterd, and M. Unser, ``A neural-network-based convex regularizer for inverse problems,'' IEEE Trans. Comput. Imaging, vol. 9, pp. 781--795, 2023

2023
[33]

Goujon, S

A. Goujon, S. Neumayer, and M. Unser, ``Learning weakly convex regularizers for convergent image-reconstruction algorithms,'' SIAM J. Imaging Sci., vol. 17, no. 1, pp. 91--115, 2024

2024
[34]

Pourya, E

M. Pourya, E. Kobler, M. Unser, and S. Neumayer, `` DEAL ing with image reconstruction: Deep attentive least squares,'' Proc. ICML, 2025

2025
[35]

C. Dong, C. C. Loy, and X. Tang, ``Accelerating the super-resolution convolutional neural network,'' Proc. ECCV, pp. 391--407, 2016

2016
[36]

Cohen, Y

R. Cohen, Y. Blau, D. Freedman, and E. Rivlin, ``It has potential: G radient-driven denoisers for convergent solutions to inverse problems,'' Proc. NeurIPS, pp. 18\,152--18\,164, 2021

2021
[37]

E. T. Reehorst and P. Schniter, ``Regularization by denoising: Clarifications and new interpretations,'' IEEE Trans. Comput. Imaging, vol. 5, no. 1, pp. 52--67, 2018

2018
[38]

H. Y. Tan, S. Mukherjee, J. Tang, and C.-B. Sch\" o nlieb, ``Provably convergent plug-and-play quasi- N ewton methods,'' SIAM J. Imaging Sci., vol. 17, no. 2, pp. 785--819, 2024

2024
[39]

E. Ryu, J. Liu, S. Wang, X. Chen, Z. Wang, and W. Yin, ``Plug-and-play methods provably converge with properly trained denoisers,'' Proc. ICML, pp. 5546--5557, 2019

2019
[40]

C. D. Athalye, K. N. Chaudhury, and B. Kumar, ``On the contractivity of plug-and-play operators,'' IEEE Signal Process. Lett., vol. 30, pp. 1447--1451, 2023

2023
[41]

Sinha, B

A. Sinha, B. Kumar, C. D. Athalye, and K. N. Chaudhury, ``Linear convergence of plug-and-play algorithms with kernel denoisers,'' IEEE Trans. Signal Process., vol. 73, pp. 2646--2659, 2025

2025
[42]

Terris, T

M. Terris, T. Moreau, N. Pustelnik, and J. Tachella, ``Equivariant plug-and-play image reconstruction,'' Proc. CVPR, pp. 25\,255--25\,264, 2024

2024
[43]

Effland, E

A. Effland, E. Kobler, K. Kunisch, and T. Pock, ``Variational networks: An optimal control approach to early stopping variational methods for image restoration,'' J. Math. Imaging Vis., vol. 62, pp. 396--416, 2020

2020
[44]

Attouch, ``Viscosity solutions of minimization problems,'' SIAM J

H. Attouch, ``Viscosity solutions of minimization problems,'' SIAM J. Optim., vol. 6, no. 3, pp. 769--806, 1996

1996
[45]

Xu, ``Viscosity approximation methods for nonexpansive mappings,'' J

H.-K. Xu, ``Viscosity approximation methods for nonexpansive mappings,'' J. Math. Anal. Appl., vol. 298, no. 1, pp. 279--291, 2004

2004
[46]

Sabach and S

S. Sabach and S. Shtern, ``A first order method for solving convex bilevel optimization problems,'' SIAM J. Optim., vol. 27, no. 2, pp. 640--660, 2017

2017
[47]

H. H. Bauschke, R. S. Burachik, P. L. Combettes, V. Elser, D. R. Luke, and H. Wolkowicz, Eds., Fixed- Point Algorithms for Inverse Problems in Science and Engineering . 1em plus 0.5em minus 0.4em Springer International Publishing, 2011

2011
[48]

S. H. Chan, X. Wang, and O. A. Elgendy, ``Plug-and-play ADMM for image restoration: Fixed-point convergence and applications,'' IEEE Trans. Comput. Imaging, vol. 3, no. 1, pp. 84--98, 2017

2017
[49]

Zhang, W

K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, ``Beyond a G aussian denoiser: Residual learning of deep CNN for image denoising,'' IEEE Trans. Image Process., vol. 26, no. 7, pp. 3142--3155, 2017

2017
[50]

J. Choi, S. Kim, Y. Jeong, Y. Gwon, and S. Yoon, `` ILVR : Conditioning method for denoising diffusion probabilistic models,'' Proc. ICCV, pp. 14\,347--14\,356, 2021

2021
[51]

Virmaux and K

A. Virmaux and K. Scaman, ``Lipschitz regularity of deep neural networks: analysis and efficient estimation,'' Proc. NeurIPS, 2018

2018
[52]

H. Gouk, E. Frank, B. Pfahringer, and M. J. Cree, ``Regularisation of neural networks by enforcing lipschitz continuity,'' Machine Learning, vol. 110, no. 2, pp. 393--416, 2021

2021
[53]

Buades, B

A. Buades, B. Coll, and J.-M. Morel, ``A non-local algorithm for image denoising,'' Proc. CVPR, vol. 2, pp. 60--65, 2005

2005
[54]

Levin, Y

A. Levin, Y. Weiss, F. Durand, and W. T. Freeman, ``Understanding and evaluating blind deconvolution algorithms,'' Proc. CVPR, pp. 1964--1971, 2009

1964
[55]

Martin, C

D. Martin, C. Fowlkes, D. Tal, and J. Malik, ``A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics,'' Proc. ICCV, vol. 2, pp. 416--423, 2001

2001
[56]

Huang, A

J.-B. Huang, A. Singh, and N. Ahuja, ``Single image super-resolution from transformed self-exemplars,'' Proc. CVPR, June 2015

2015
[57]

Tachella, D

J. Tachella, D. Chen, S. Hurault, M. Terris, and A. Wang, `` D eep I nverse: A deep learning framework for inverse problems in imaging,'' URL: https://deepinv.github.io/deepinv, 2023

2023
[58]

Zhang, Y

K. Zhang, Y. Li, J. Liang, J. Cao, Y. Zhang, H. Tang, D.-P. Fan, R. Timofte, and L. V. Gool, ``Practical blind image denoising via S win- C onv- UN et and data synthesis,'' Mach. Intell. Res., vol. 20, no. 6, pp. 822--836, 2023

2023

Pith tools

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

[1] [1]

11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEcompsocconfonly \@IEEEauthordefaulttextstyle \@IEEEcompsocnotconfonly \@IEEEauthor...

work page

[2] [2]

C. A. Bouman, Foundations of Computational Imaging: A Model-Based Approach. 1em plus 0.5em minus 0.4em SIAM, 2022

work page 2022

[3] [3]

L. I. Rudin, S. Osher, and E. Fatemi, ``Nonlinear total variation based noise removal algorithms,'' Phys. D: Nonlinear Phenom., vol. 60, no. 1-4, pp. 259--268, 1992

work page 1992

[4] [4]

Geman and C

S. Geman and C. Graffigne, ``Markov R andom F ield image models and their applications to computer vision,'' Proc. ICM, vol. 1, p. 2, 1986

work page 1986

[5] [5]

K. H. Jin, M. T. McCann, E. Froustey, and M. Unser, ``Deep convolutional neural network for inverse problems in imaging,'' IEEE Trans. Image Process., vol. 26, no. 9, pp. 4509--4522, 2017

work page 2017

[6] [6]

H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, and S.-T. Xia, `` MambaIR : A simple baseline for image restoration with state-space model,'' Proc. ECCV, pp. 222--241, 2024

work page 2024

[7] [7]

Liang, J

J. Liang, J. Cao, G. Sun, K. Zhang, L. Van Gool, and R. Timofte, `` SwinIR : Image restoration using swin transformer,'' Proc. ICCV Workshops, pp. 1833--1844, 2021

work page 2021

[8] [8]

S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang, ``Restormer: E fficient transformer for high-resolution image restoration,'' Proc. CVPR, pp. 5718--5739, 2022

work page 2022