pith. sign in

arxiv: 2509.14566 · v1 · submitted 2025-09-18 · 💻 cs.CV

DICE: Diffusion Consensus Equilibrium for Sparse-view CT Reconstruction

Leon Suarez-Rodriguez , Roman Jacome , Romario Gualdron-Hurtado , Ana Mantilla-Dulcey , Henry Arguello This is my paper

Pith reviewed 2026-05-18 16:47 UTC · model grok-4.3

classification 💻 cs.CV
keywords sparse-view CT reconstructiondiffusion modelsconsensus equilibriumproximal operatorgenerative priorsmedical image reconstructioninverse problemsCT imaging
0
0 comments X

The pith

By alternating a data-consistency check with a diffusion model prior inside a consensus equilibrium, DICE produces sharper CT reconstructions from sparse X-ray views.

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

The paper tries to show that sparse-view CT reconstruction can be solved more effectively by embedding a two-agent consensus equilibrium directly into the sampling steps of a diffusion model. One agent enforces consistency with the available measurements through a proximal operator while the other uses the diffusion model to estimate a clean image, and these steps are balanced iteratively. A sympathetic reader would care because fewer views mean lower radiation exposure for patients, yet current methods still leave noticeable artifacts in complex anatomy. The experiments cover uniform and non-uniform patterns at 15, 30, and 60 views out of 180 and report clear gains over prior baselines.

Core claim

DICE integrates a two-agent consensus equilibrium into the sampling process of a diffusion model. It alternates between a data-consistency agent realized by a proximal operator that enforces measurement consistency and a prior agent realized by the diffusion model that performs clean-image estimation at each step. Balancing these agents iteratively combines the generative strength of the diffusion prior with exact fidelity to the measured data, yielding higher-quality reconstructions than existing methods under uniform and non-uniform sparse-view conditions of 15, 30, and 60 views.

What carries the argument

Diffusion Consensus Equilibrium (DICE), a framework that alternates between a proximal operator for data consistency and the diffusion model for prior estimation at every sampling step.

If this is right

  • Clear outperformance over state-of-the-art baselines on both uniform and non-uniform sparse-view CT at 15, 30, and 60 views.
  • Successful fusion of strong generative priors with exact measurement consistency in a single iterative loop.
  • Stable performance across different sparsity levels without requiring handcrafted regularization.
  • Direct applicability to reducing the number of projections while preserving diagnostic detail in medical CT.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same alternating-agent structure could be tested on other limited-data inverse problems such as limited-angle tomography or MRI with accelerated acquisitions.
  • Replacing the generic diffusion model with one trained specifically on CT anatomy distributions might further reduce residual artifacts at the lowest view counts.
  • Clinical translation would require checking whether the method preserves fine structures such as small lesions when the input measurements contain realistic detector noise.

Load-bearing premise

The two agents can be balanced iteratively inside the consensus equilibrium without introducing systematic bias or divergence and the diffusion model remains stable when repeatedly projected back onto the measurement-consistent subspace.

What would settle it

Reconstruction quality that stops improving or begins to introduce new artifacts after a modest number of consensus iterations on the same sparse-view data would show the balancing step does not hold.

Figures

Figures reproduced from arXiv: 2509.14566 by Ana Mantilla-Dulcey, Henry Arguello, Leon Suarez-Rodriguez, Roman Jacome, Romario Gualdron-Hurtado.

Figure 1
Figure 1. Figure 1: Visual comparison of CT reconstructions under uniform and non-uniform sparse-view sampling (15, 30, and 60 views), obtained [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of the parameters: Strength of agent [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Effect of parameters: Mann iterations K, and conjugate gradient descent steps P. Skipping time-steps: While DICE significantly improves reconstruction performance over existing methods, it introduces additional sampling time due to solving a CE problem at each diffusion sampling step. Therefore, we analyze the performance of DICE using a reduced number of sampling steps, T ∈ {50, 100, 200, 500, 1000} [PIT… view at source ↗
Figure 4
Figure 4. Figure 4: Effect of skipping time-steps IV. CONCLUSIONS & FUTURE WORK This work proposed DICE: Diffusion consensus equilibrium, a framework that casts the conditioning generative process of a DM as a consensus equilibrium problem between two agents: (i) the data-consistency agent, given by a proximal map, and (ii) the generative prior agent, given by a denoising step. Simulation with sparse-view CT with high undersa… view at source ↗
read the original abstract

Sparse-view computed tomography (CT) reconstruction is fundamentally challenging due to undersampling, leading to an ill-posed inverse problem. Traditional iterative methods incorporate handcrafted or learned priors to regularize the solution but struggle to capture the complex structures present in medical images. In contrast, diffusion models (DMs) have recently emerged as powerful generative priors that can accurately model complex image distributions. In this work, we introduce Diffusion Consensus Equilibrium (DICE), a framework that integrates a two-agent consensus equilibrium into the sampling process of a DM. DICE alternates between: (i) a data-consistency agent, implemented through a proximal operator enforcing measurement consistency, and (ii) a prior agent, realized by a DM performing a clean image estimation at each sampling step. By balancing these two complementary agents iteratively, DICE effectively combines strong generative prior capabilities with measurement consistency. Experimental results show that DICE significantly outperforms state-of-the-art baselines in reconstructing high-quality CT images under uniform and non-uniform sparse-view settings of 15, 30, and 60 views (out of a total of 180), demonstrating both its effectiveness and robustness.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces Diffusion Consensus Equilibrium (DICE), a framework that integrates a two-agent consensus equilibrium into the sampling process of a diffusion model for sparse-view CT reconstruction. It alternates between (i) a data-consistency agent implemented via a proximal operator enforcing measurement consistency with sparse Radon measurements and (ii) a prior agent realized by the diffusion model performing clean-image estimation at each sampling step. The central claim is that this iterative balancing significantly outperforms state-of-the-art baselines for both uniform and non-uniform sparse-view settings using 15, 30, and 60 views out of 180 total views.

Significance. If the stability of the alternating procedure and the reported gains are rigorously validated, the work could advance the application of generative diffusion priors to ill-posed inverse problems in medical imaging by providing a principled mechanism for enforcing data fidelity without handcrafted regularization.

major comments (2)
  1. [Methods (DICE framework and consensus equilibrium)] The description of the DICE alternating procedure (data-consistency proximal operator with DM prior agent) provides no convergence analysis, contraction mapping argument, or empirical monitoring of the data-fidelity residual across iterations. This is load-bearing for the robustness claim across sparsity levels, especially the 15-view non-uniform case where repeated projection onto the measurement-consistent subspace risks driving samples outside the learned manifold.
  2. [Experiments] The experimental section reports outperformance on multiple sparsity levels but, consistent with the abstract, supplies no quantitative metrics (e.g., PSNR/SSIM values), baseline implementation details, or ablation studies on the number of equilibrium iterations or agent weighting. Without these, the central empirical claim cannot be assessed.
minor comments (1)
  1. [Methods] Notation for the proximal operator and the clean-image estimation step within the consensus equilibrium could be clarified with explicit update equations to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript introducing DICE for sparse-view CT reconstruction. We address each major comment below and outline the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [Methods (DICE framework and consensus equilibrium)] The description of the DICE alternating procedure (data-consistency proximal operator with DM prior agent) provides no convergence analysis, contraction mapping argument, or empirical monitoring of the data-fidelity residual across iterations. This is load-bearing for the robustness claim across sparsity levels, especially the 15-view non-uniform case where repeated projection onto the measurement-consistent subspace risks driving samples outside the learned manifold.

    Authors: We acknowledge that the manuscript lacks a formal convergence analysis or contraction mapping argument for the alternating procedure. The integration of the proximal data-consistency operator with the stochastic diffusion sampling process renders a standard fixed-point guarantee non-trivial to derive, as the diffusion steps introduce non-convexity and stochasticity. However, we will add empirical monitoring of the data-fidelity residual (||Ax - y||_2) across equilibrium iterations for all sparsity levels and sampling patterns in the revised manuscript. These plots will demonstrate stabilization of the residual without divergence, even in the 15-view non-uniform case, supporting that samples remain consistent with the learned manifold while satisfying measurements. We believe this provides practical validation of stability. revision: partial

  2. Referee: [Experiments] The experimental section reports outperformance on multiple sparsity levels but, consistent with the abstract, supplies no quantitative metrics (e.g., PSNR/SSIM values), baseline implementation details, or ablation studies on the number of equilibrium iterations or agent weighting. Without these, the central empirical claim cannot be assessed.

    Authors: We agree that the experimental presentation would be strengthened by explicit quantitative metrics and additional studies. In the revised version, we will include a comprehensive table reporting PSNR and SSIM values for DICE and all baselines (including implementation details such as network architectures and training protocols) across uniform and non-uniform settings with 15, 30, and 60 views. We will also add ablation experiments varying the number of equilibrium iterations (e.g., 1, 5, 10) and the relative weighting between the data-consistency and prior agents, with corresponding PSNR/SSIM results. These additions will make the performance claims fully assessable. revision: yes

Circularity Check

0 steps flagged

DICE alternating equilibrium is an independent algorithmic construction

full rationale

The paper defines DICE as an explicit two-agent consensus equilibrium that alternates a proximal data-consistency operator with a diffusion-model clean-image estimation step inside the sampling process. This procedure is introduced by construction as a new framework and is not obtained by fitting parameters to the target sparse-view reconstructions or by reducing to any self-citation chain. Experimental claims are presented as empirical validation on 15/30/60-view CT data rather than as a derived prediction that collapses to the inputs. No load-bearing step reduces to a fitted quantity renamed as a result or to an ansatz smuggled via prior work by the same authors. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The method rests on standard assumptions from optimization and generative modeling rather than new free parameters or invented entities.

axioms (2)
  • domain assumption Proximal operators can enforce data consistency for linear measurement models in inverse problems
    Invoked for the data-consistency agent in the consensus loop.
  • domain assumption Pre-trained diffusion models provide sufficiently accurate generative priors for medical CT images
    Central premise enabling the prior agent to produce clean image estimates.

pith-pipeline@v0.9.0 · 5747 in / 1319 out tokens · 36242 ms · 2026-05-18T16:47:09.683749+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Enhancing radiological diagnosis: A comprehensive review of image quality assessment and optimization strategies,

    A. P. Varghese, S. Naik, S. Asrar Up Haq Andrabi, A. Luharia, and S. Tivaskar, “Enhancing radiological diagnosis: A comprehensive review of image quality assessment and optimization strategies,”Cureus, vol. 16, no. 6, p. e63016, Jun. 2024

    work page 2024

  2. [2]

    An accurate iterative reconstruction algorithm for sparse objects: application to 3D blood vessel reconstruction from a limited number of projections,

    M. Li, H. Yang, and H. Kudo, “An accurate iterative reconstruction algorithm for sparse objects: application to 3D blood vessel reconstruction from a limited number of projections,”Phys. Med. Biol., vol. 47, no. 15, pp. 2599–2609, Aug. 2002

    work page 2002

  3. [3]

    Image reconstruction for sparse-view CT and interior CT-introduction to compressed sensing and differentiated backprojection,

    H. Kudo, T. Suzuki, and E. A. Rashed, “Image reconstruction for sparse-view CT and interior CT-introduction to compressed sensing and differentiated backprojection,”Quant. Imaging Med. Surg., vol. 3, no. 3, pp. 147–161, Jun. 2013

    work page 2013

  4. [4]

    Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,

    E. Y . Sidky and X. Pan, “Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization,” Phys. Med. Biol., vol. 53, no. 17, pp. 4777–4807, Sep. 2008

    work page 2008

  5. [5]

    An iterative thresholding algorithm for linear inverse problems with a sparsity constraint,

    I. Daubechies, M. Defrise, and C. De Mol, “An iterative thresholding algorithm for linear inverse problems with a sparsity constraint,” Commun. Pure Appl. Math., vol. 57, no. 11, pp. 1413–1457, Nov. 2004

    work page 2004

  6. [6]

    Plug-and-play priors for model based reconstruction,

    S. V . Venkatakrishnan, C. A. Bouman, and B. Wohlberg, “Plug-and-play priors for model based reconstruction,” in2013 IEEE Global Conference on Signal and Information Processing, 2013, pp. 945–948

    work page 2013

  7. [7]

    Neural-network-based regularization methods for inverse problems in imaging,

    A. Habring and M. Holler, “Neural-network-based regularization methods for inverse problems in imaging,”GAMM-Mitteilungen, vol. 47, no. 4, Nov. 2024

    work page 2024

  8. [8]

    Robust compressed sensing mri with deep generative priors,

    A. Jalal, M. Arvinte, G. Daras, E. Price, A. G. Dimakis, and J. Tamir, “Robust compressed sensing mri with deep generative priors,” inAdvances in Neural Information Processing Systems, M. Ranzato, A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan, Eds., vol. 34. Curran Associates, Inc., 2021, pp. 14 938–14 954. [Online]. Available: https://proceedi...

    work page 2021

  9. [9]

    Plug-and-play admm for image restoration: Fixed-point convergence and applications,

    S. H. Chan, X. Wang, and O. A. Elgendy, “Plug-and-play admm for image restoration: Fixed-point convergence and applications,” IEEE Transactions on Computational Imaging, vol. 3, no. 1, pp. 84–98, 2017

    work page 2017

  10. [10]

    Plug-and-play image restoration with deep denoiser prior,

    K. Zhang, Y . Li, W. Zuo, L. Zhang, L. Van Gool, and R. Timofte, “Plug-and-play image restoration with deep denoiser prior,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 10, pp. 6360–6376, 2022

    work page 2022

  11. [11]

    Deep convolutional neural network for inverse problems in imaging,

    K. H. Jin, M. T. McCann, E. Froustey, and M. Unser, “Deep convolutional neural network for inverse problems in imaging,”IEEE Transactions on Image Processing, vol. 26, no. 9, pp. 4509–4522, 2017

    work page 2017

  12. [12]

    Framing u-net via deep convolutional framelets: Application to sparse-view ct,

    Y . Han and J. C. Ye, “Framing u-net via deep convolutional framelets: Application to sparse-view ct,”IEEE Transactions on Medical Imaging, vol. 37, no. 6, pp. 1418–1429, 2018

    work page 2018

  13. [13]

    A sparse-view ct reconstruction method based on combination of densenet and deconvolution,

    Z. Zhang, X. Liang, X. Dong, Y . Xie, and G. Cao, “A sparse-view ct reconstruction method based on combination of densenet and deconvolution,”IEEE Transactions on Medical Imaging, vol. 37, no. 6, pp. 1407–1417, 2018

    work page 2018

  14. [14]

    Learn: Learned experts’ assessment- based reconstruction network for sparse-data ct,

    H. Chen, Y . Zhang, Y . Chen, J. Zhang, W. Zhang, H. Sun, Y . Lv, P. Liao, J. Zhou, and G. Wang, “Learn: Learned experts’ assessment- based reconstruction network for sparse-data ct,”IEEE Transactions on Medical Imaging, vol. 37, no. 6, pp. 1333–1347, 2018. 8

    work page 2018

  15. [15]

    Drone: Dual-domain residual-based optimization network for sparse- view ct reconstruction,

    W. Wu, D. Hu, C. Niu, H. Yu, V . Vardhanabhuti, and G. Wang, “Drone: Dual-domain residual-based optimization network for sparse- view ct reconstruction,”IEEE Transactions on Medical Imaging, vol. 40, no. 11, pp. 3002–3014, 2021

    work page 2021

  16. [16]

    Artifact removal using a hybrid-domain convolutional neural network for limited-angle computed tomography imaging,

    Q. Zhang, Z. Hu, C. Jiang, H. Zheng, Y . Ge, and D. Liang, “Artifact removal using a hybrid-domain convolutional neural network for limited-angle computed tomography imaging,”Physics in Medicine & Biology, vol. 65, no. 15, p. 155010, jul 2020. [Online]. Available: https://dx.doi.org/10.1088/1361-6560/ab9066

    work page doi:10.1088/1361-6560/ab9066 2020

  17. [17]

    Deep unsupervised learning using nonequilibrium thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep unsupervised learning using nonequilibrium thermodynamics,” inProceedings of the 32nd International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, F. Bach and D. Blei, Eds., vol. 37. Lille, France: PMLR, 07–09 Jul 2015, pp. 2256–2265. [Online]. Available...

    work page 2015

  18. [18]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” inAdvances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 6840–6851. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2020/file/4c5bcfec8584af0d967f1ab101...

    work page 2020

  19. [19]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” inInternational Conference on Learning Representations,

    work page

  20. [20]

    Available: https://openreview.net/forum?id=St1giarCHLP

    [Online]. Available: https://openreview.net/forum?id=St1giarCHLP

    work page

  21. [21]

    Score-based generative modeling through stochastic differential equations,

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differential equations,” inInternational Conference on Learning Representations, 2021. [Online]. Available: https://openreview.net/forum?id=PxTIG12RRHS

    work page 2021

  22. [22]

    Cddip: Constrained diffusion-driven deep image prior for seismic data reconstruction,

    P. Goyes-Pe ˜nafiel, U. S. Kamilov, and H. Arguello, “Cddip: Constrained diffusion-driven deep image prior for seismic data reconstruction,”IEEE Geoscience and Remote Sensing Letters, vol. 22, pp. 1–5, 2025

    work page 2025

  23. [23]

    Constrained diffusion for seismic data reconstruction using deep image and generative prior,

    L. Rodr ´ıguez-L´opez, P. Goyes-Pe ˜nafiel, L. Galvis, and H. Arguello, “Constrained diffusion for seismic data reconstruction using deep image and generative prior,” vol. 2025, no. 1, pp. 1–5, 2025. [Online]. Available: https://www.earthdoc.org/content/papers/10.3997/ 2214-4609.2025101342

    work page arXiv 2025

  24. [24]

    Score-based diffusion models for accelerated mri,

    H. Chung and J. C. Ye, “Score-based diffusion models for accelerated mri,”Medical Image Analysis, vol. 80, p. 102479, 2022. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1361841522001268

    work page 2022

  25. [25]

    Diffusion Posterior Sampling for General Noisy Inverse Problems

    H. Chung, J. Kim, M. T. Mccann, M. L. Klasky, and J. C. Ye, “Diffusion posterior sampling for general noisy inverse problems,”arXiv preprint arXiv:2209.14687, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  26. [26]

    Denoising diffusion models for plug-and-play image restoration,

    Y . Zhu, K. Zhang, J. Liang, J. Cao, B. Wen, R. Timofte, and L. Van Gool, “Denoising diffusion models for plug-and-play image restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2023, pp. 1219–1229

    work page 2023

  27. [27]

    Denoising diffusion restoration models,

    B. Kawar, M. Elad, S. Ermon, and J. Song, “Denoising diffusion restoration models,” inAdvances in Neural Information Processing Systems, S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, Eds., vol. 35. Curran Associates, Inc., 2022, pp. 23 593–23 606. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2022/file/95504595...

    work page arXiv 2022

  28. [28]

    Solving inverse problems with latent diffusion models via hard data consistency,

    B. Song, S. M. Kwon, Z. Zhang, X. Hu, Q. Qu, and L. Shen, “Solving inverse problems with latent diffusion models via hard data consistency,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://openreview.net/forum?id=j8hdRqOUhN

    work page 2024

  29. [29]

    Deep data consistency: a fast and robust diffusion model-based solver for inverse problems,

    H. Chen, Z. Hao, and L. Xiao, “Deep data consistency: a fast and robust diffusion model-based solver for inverse problems,”arXiv preprint arXiv:2405.10748, 2024

    work page arXiv 2024

  30. [30]

    The perception-distortion tradeoff,

    Y . Blau and T. Michaeli, “The perception-distortion tradeoff,” in2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 6228–6237

    work page 2018

  31. [31]

    Plug-and-play unplugged: Optimization-free reconstruction using consensus equilibrium,

    G. T. Buzzard, S. H. Chan, S. Sreehari, and C. A. Bouman, “Plug-and-play unplugged: Optimization-free reconstruction using consensus equilibrium,”SIAM J. Imaging Sci., vol. 11, no. 3, pp. 2001–2020, Jan. 2018

    work page 2001

  32. [32]

    C. A. Bouman,F oundations of computational imaging, ser. Other Titles in Applied Mathematics. New York, NY: Society for Industrial & Applied Mathematics, Aug. 2022

    work page 2022

  33. [33]

    A consensus equilibrium approach for 3-d land seismic shots recovery,

    P. Goyes-Pe ˜nafiel, E. Vargas, C. V . Correa, W. Agudelo, B. Wohlberg, and H. Arguello, “A consensus equilibrium approach for 3-d land seismic shots recovery,”IEEE Geoscience and Remote Sensing Letters, vol. 19, pp. 1–5, 2022

    work page 2022

  34. [34]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” inInternational Conference on Learning Representations, 2019. [Online]. Available: https://openreview.net/forum?id=Bkg6RiCqY7

    work page 2019

  35. [35]

    LoDoPaB-CT, a benchmark dataset for low-dose computed tomography reconstruction,

    J. Leuschner, M. Schmidt, D. O. Baguer, and P. Maass, “LoDoPaB-CT, a benchmark dataset for low-dose computed tomography reconstruction,”Sci. Data, vol. 8, no. 1, p. 109, Apr. 2021

    work page 2021

  36. [36]

    Image reconstruction: Part 1–understanding filtered back projection, noise and image acquisition,

    R. Schofield, L. King, U. Tayal, I. Castellano, J. Stirrup, F. Pontana, J. Earls, and E. Nicol, “Image reconstruction: Part 1–understanding filtered back projection, noise and image acquisition,”Journal of cardiovascular computed tomography, vol. 14, no. 3, pp. 219–225, 2020

    work page 2020

  37. [37]

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

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, “Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,” IEEE Transactions on Image Processing, vol. 26, no. 7, pp. 3142–3155, 2017

    work page 2017