pith. sign in

arxiv: 2507.18654 · v2 · submitted 2025-07-22 · 💻 cs.LG · cs.CV

Diffusion Models for Solving Inverse Problems via Posterior Sampling with Piecewise Guidance

Saeed Mohseni-Sehdeh , Walid Saad , Kei Sakaguchi , Tao Yu This is my paper

Pith reviewed 2026-05-19 03:00 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords diffusion modelsinverse problemsposterior samplingpiecewise guidanceimage inpaintingsuper-resolutionconditional samplinggenerative models
0
0 comments X

The pith

A piecewise guidance scheme lets diffusion models solve inverse problems faster while preserving sample quality.

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

This paper proposes a diffusion-based framework for inverse problems that defines the guidance term as a piecewise function of the diffusion timestep. The design applies different approximations in high-noise and low-noise phases to improve efficiency without substantial accuracy loss. The approach is problem-agnostic, incorporates measurement noise directly, and requires no per-task retraining. Experiments on image inpainting and super-resolution show inference time reductions of 25 percent for inpainting and 23 to 24 percent for 4x and 8x super-resolution relative to a pseudoinverse-guided baseline, with only negligible drops in PSNR and SSIM. A reader would care because the method makes conditional sampling more practical for restoration tasks that arise in imaging and signal processing.

Core claim

The paper establishes that defining the guidance term as a piecewise function of the diffusion timestep allows the use of different approximations during high-noise versus low-noise phases. This balances computational efficiency against the accuracy of the resulting posterior samples. The resulting method applies to a variety of inverse problems, explicitly accounts for measurement noise, and delivers measurable speedups on image restoration benchmarks while maintaining reconstruction fidelity.

What carries the argument

The piecewise guidance scheme, which switches the form of the guidance approximation according to the current diffusion timestep to trade off cost and fidelity.

If this is right

  • The framework applies to multiple inverse problems without task-specific retraining.
  • Measurement noise is incorporated explicitly into the posterior sampling process.
  • Inference time is reduced by 25 percent on inpainting with random and center masks relative to the pseudoinverse-guided baseline.
  • Inference time is reduced by 23 percent and 24 percent on 4x and 8x super-resolution, respectively, with negligible loss in PSNR and SSIM.

Where Pith is reading between the lines

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

  • Optimizing the timestep threshold where the approximation switches could produce further speed gains on new tasks.
  • The same piecewise strategy may transfer to other conditional generation settings that rely on timestep-dependent guidance.

Load-bearing premise

The piecewise function of the diffusion timestep can safely use different approximations in high-noise versus low-noise phases without materially harming the accuracy of the posterior samples.

What would settle it

Re-running the reported inpainting and super-resolution experiments with a single non-piecewise guidance approximation throughout all timesteps and finding substantially larger quality degradation than the negligible loss claimed would falsify the efficiency-accuracy balance.

Figures

Figures reproduced from arXiv: 2507.18654 by Kei Sakaguchi, Saeed Mohseni-Sehdeh, Tao Yu, Walid Saad.

Figure 1
Figure 1. Figure 1: Restoration results on four inverse problems, inpainting with center [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Restoration results on four inverse problems, inpainting with center [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: Behavior of the coefficient 1−α¯t α¯t across different diffusion time-step values, shown on a logarithmic scale. Consistent with Theorems 1 and 2, the coefficient decreases rapidly at lower diffusion time steps, supporting the validity of the approximation in this regime. Proof. Similar to the proof of Theorem 1, for the true condi￾tional distribution we have pt(y|xt) = N  − √ 1 − α¯t √ α¯t Cvt + 1 √ α¯t … view at source ↗
Figure 9
Figure 9. Figure 9: Average SSIM scores across various inverse problems as a function [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 7
Figure 7. Figure 7: Average PSNR scores across various inverse problems as a function [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Average SSIM scores across various inverse problems as a function [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 13
Figure 13. Figure 13: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p008_13.png] view at source ↗
Figure 11
Figure 11. Figure 11: Average LPIPS scores across various inverse problems as a function [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p009_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p009_15.png] view at source ↗
Figure 19
Figure 19. Figure 19: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p010_19.png] view at source ↗
Figure 17
Figure 17. Figure 17: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p010_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Average inference time per image as a function of the guidance [PITH_FULL_IMAGE:figures/full_fig_p010_18.png] view at source ↗
read the original abstract

Diffusion models are powerful tools for sampling from high-dimensional distributions by progressively transforming pure noise into structured data through a denoising process. When equipped with a guidance mechanism, these models can also generate samples from conditional distributions. In this paper, a novel diffusion-based framework is introduced for solving inverse problems using a piecewise guidance scheme. The guidance term is defined as a piecewise function of the diffusion timestep, facilitating the use of different approximations during high-noise and low-noise phases. This design is shown to effectively balance computational efficiency with the accuracy of the guidance term. Unlike task-specific approaches that require retraining for each problem, the proposed method is problem-agnostic and readily adaptable to a variety of inverse problems. Additionally, it explicitly incorporates measurement noise into the reconstruction process. The effectiveness of the proposed framework is demonstrated through extensive experiments on image restoration tasks, specifically image inpainting and super-resolution. Using a class conditional diffusion model for recovery, compared to the \blue{pseudoinverse-guided diffusion model (\textrm{\(\Pi\)}GDM) baseline}, the proposed framework achieves a reduction in inference time of \(25\%\) for inpainting with both random and center masks, and \(23\%\) and \(24\%\) for \(4\times\) and \(8\times\) super-resolution tasks, respectively, while incurring only negligible loss in PSNR and SSIM.

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 / 2 minor

Summary. The manuscript introduces a diffusion-based framework for solving inverse problems via posterior sampling with a piecewise guidance scheme. The guidance term is defined as a piecewise function of the diffusion timestep to apply different approximations in high-noise versus low-noise regimes, aiming to balance efficiency and accuracy. The method is presented as problem-agnostic, incorporates measurement noise explicitly, and is evaluated on image inpainting (random and center masks) and super-resolution (4× and 8×) tasks using a class-conditional diffusion model. It reports 25% inference-time reduction for inpainting and 23–24% for super-resolution relative to the ΠGDM baseline, with only negligible drops in PSNR and SSIM.

Significance. If the piecewise approximation is shown to preserve posterior accuracy, the framework would provide a practical, general-purpose acceleration technique for conditional diffusion sampling in inverse problems, avoiding task-specific retraining while explicitly handling measurement noise. The empirical speedups on standard image-restoration benchmarks would be a useful contribution to the efficiency literature in this area.

major comments (2)
  1. [§3] §3 (Proposed Method): The piecewise guidance definition and the specific approximations chosen for the high-noise and low-noise phases are not accompanied by any error analysis, bias bounds, or propagation study showing that residuals from the high-noise approximation are corrected in the low-noise phase. This is load-bearing for the central claim that the scheme achieves speedups with negligible quality loss.
  2. [§4] §4 (Experiments): The reported speedups (25% for inpainting, 23% and 24% for 4×/8× super-resolution) and the assertion of “negligible loss” in PSNR/SSIM are given without error bars, standard deviations across runs, or ablation on the timestep switch point, so the robustness of the quality-parity claim cannot be assessed from the presented data.
minor comments (2)
  1. Clarify the exact functional form of the piecewise guidance (including how the switch point is selected) and provide pseudocode or a small algorithmic box for reproducibility.
  2. The abstract and introduction should explicitly state the datasets, diffusion backbone, and number of function evaluations used for both the proposed method and the ΠGDM baseline.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and outline the revisions we will make to improve the theoretical grounding and experimental robustness of the work.

read point-by-point responses

  1. Referee: [§3] §3 (Proposed Method): The piecewise guidance definition and the specific approximations chosen for the high-noise and low-noise phases are not accompanied by any error analysis, bias bounds, or propagation study showing that residuals from the high-noise approximation are corrected in the low-noise phase. This is load-bearing for the central claim that the scheme achieves speedups with negligible quality loss.

    Authors: We acknowledge the value of a formal error analysis for supporting the central claim. In the revised manuscript we will add a dedicated paragraph in §3 that (i) characterizes the approximation error in the high-noise regime as being dominated by the large diffusion noise variance, (ii) notes that the low-noise regime reverts to the exact guidance term used by ΠGDM, and (iii) provides a short propagation argument showing that any residual bias introduced early is subsequently corrected by the accurate guidance steps. While deriving tight, non-asymptotic bias bounds for the full iterative sampler remains technically involved, we will include this qualitative analysis together with empirical intermediate-sample visualizations that illustrate the correction effect. These additions directly address the load-bearing concern. revision: partial

  2. Referee: [§4] §4 (Experiments): The reported speedups (25% for inpainting, 23% and 24% for 4×/8× super-resolution) and the assertion of “negligible loss” in PSNR/SSIM are given without error bars, standard deviations across runs, or ablation on the timestep switch point, so the robustness of the quality-parity claim cannot be assessed from the presented data.

    Authors: We agree that variability measures and ablation studies are necessary to substantiate the quality-parity claim. In the revised §4 we will (i) rerun all experiments over five independent random seeds and report mean ± standard deviation for PSNR and SSIM, (ii) attach error bars to the reported inference-time reductions, and (iii) add a new ablation table that varies the piecewise switch timestep (e.g., t = 400, 600, 800) while keeping all other settings fixed. These changes will allow readers to assess both statistical reliability and sensitivity to the design choice. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or claims

full rationale

The paper introduces a piecewise guidance scheme for diffusion-based posterior sampling in inverse problems and validates it empirically on inpainting and super-resolution tasks, reporting speedups versus the ΠGDM baseline with negligible PSNR/SSIM degradation. No equations, parameter-fitting procedures, or self-citation chains are present that would reduce any claimed prediction or result to an input by construction. The central efficiency and accuracy claims rest on experimental outcomes rather than tautological redefinitions or fitted quantities renamed as predictions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no visible free parameters, axioms, or invented entities; the method appears to rest on standard diffusion model assumptions not detailed here.

pith-pipeline@v0.9.0 · 5783 in / 1089 out tokens · 28768 ms · 2026-05-19T03:00:55.775908+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

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

  1. [1]

    Diffusion Models Beat GANs on Image Synthesis,

    P. Dhariwal and A. Nichol, “Diffusion Models Beat GANs on Image Synthesis,” in Advances in Neural Information Processing Systems, vol. 34. Curran Associates, Inc., 2021, pp. 8780–

    work page 2021

  2. [2]

    Available: https://proceedings.neurips.cc/paper/2021/ hash/49ad23d1ec9fa4bd8d77d02681df5cfa-Abstract.html

    [Online]. Available: https://proceedings.neurips.cc/paper/2021/ hash/49ad23d1ec9fa4bd8d77d02681df5cfa-Abstract.html

    work page 2021

  3. [3]

    Label-Efficient Semantic Segmentation with Diffusion Models,

    D. Baranchuk, A. V oynov, I. Rubachev, V . Khrulkov, and A. Babenko, “Label-Efficient Semantic Segmentation with Diffusion Models,” in Proc. of International Conference on Learning Representations , Oct. 2021. [Online]. Available: https://openreview.net/forum?id=SlxSY2UZQT

    work page 2021

  4. [4]

    SegDiff: Image Segmentation with Diffusion Probabilistic Models,

    T. Amit, T. Shaharbany, E. Nachmani, and L. Wolf, “SegDiff: Image Segmentation with Diffusion Probabilistic Models,” Sep. 2022, arXiv:2112.00390 [cs]. [Online]. Available: http://arxiv.org/abs/2112. 00390

    work page arXiv 2022

  5. [5]

    Structured Denoising Diffusion Models in Discrete State- Spaces,

    J. Austin, D. D. Johnson, J. Ho, D. Tarlow, and R. van den Berg, “Structured Denoising Diffusion Models in Discrete State- Spaces,” in Advances in Neural Information Processing Systems , vol. 34. Curran Associates, Inc., 2021, pp. 17 981–17 993. [Online]. Available: https://proceedings.neurips.cc/paper/2021/hash/ 958c530554f78bcd8e97125b70e6973d-Abstract.html

    work page 2021

  6. [6]

    Argmax Flows and Multinomial Diffusion: Learning Categorical Distributions,

    E. Hoogeboom, D. Nielsen, P. Jaini, P. Forr ´e, and M. Welling, “Argmax Flows and Multinomial Diffusion: Learning Categorical Distributions,” in Advances in Neural Information Processing Systems , vol. 34, 2021, pp. 12 454–12 465. [Online]. Available: https://proceedings.neurips.cc/ paper/2021/hash/67d96d458abdef21792e6d8e590244e7-Abstract.html

    work page 2021

  7. [7]

    DiffWave: A Versatile Diffusion Model for Audio Synthesis

    Z. Kong, W. Ping, J. Huang, K. Zhao, and B. Catanzaro, “DiffWave: A Versatile Diffusion Model for Audio Synthesis,” Mar. 2021. [Online]. Available: http://arxiv.org/abs/2009.09761

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  8. [8]

    MR Image Denoising and Super- Resolution Using Regularized Reverse Diffusion,

    H. Chung, E. S. Lee, and J. C. Ye, “MR Image Denoising and Super- Resolution Using Regularized Reverse Diffusion,” IEEE Transactions on Medical Imaging , vol. 42, no. 4, pp. 922–934, Apr. 2023. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/9941138

    work page arXiv 2023

  9. [9]

    High-Frequency Space Diffusion Model for Accelerated MRI,

    C. Cao, Z.-X. Cui, Y . Wang, S. Liu, T. Chen, H. Zheng, D. Liang, and Y . Zhu, “High-Frequency Space Diffusion Model for Accelerated MRI,” IEEE Transactions on Medical Imaging , vol. 43, no. 5, pp. 1853–1865, May 2024. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10385176 11

    work page arXiv 2024

  10. [10]

    Diffusion Models: A Comprehensive Survey of Methods and Applications,

    L. Yang, Z. Zhang, Y . Song, S. Hong, R. Xu, Y . Zhao, W. Zhang, B. Cui, and M.-H. Yang, “Diffusion Models: A Comprehensive Survey of Methods and Applications,” ACM Computing Surveys , vol. 56, no. 4, pp. 1–39, Nov. 2023

    work page 2023

  11. [11]

    Denoising Diffusion Restoration Models,

    B. Kawar, M. Elad, S. Ermon, and J. Song, “Denoising Diffusion Restoration Models,” Advances in Neural Information Processing Systems , vol. 35, pp. 23 593–23 606, Dec. 2022. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2022/hash/ 95504595b6169131b6ed6cd72eb05616-Abstract-Conference.html

    work page 2022

  12. [12]

    The Perception-Distortion Tradeoff,

    Y . Blau and T. Michaeli, “The Perception-Distortion Tradeoff,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun. 2018, pp. 6228–6237. [Online]. Available: https: //ieeexplore.ieee.org/document/8578750/authors

    work page arXiv 2018

  13. [13]

    Pseudoinverse-Guided Diffusion Models for Inverse Problems,

    J. Song, A. Vahdat, M. Mardani, and J. Kautz, “Pseudoinverse-Guided Diffusion Models for Inverse Problems,” in Proc. of International Conference on Learning Representations , Sep. 2022. [Online]. Available: https://openreview.net/forum?id=9 gsMA8MRKQ

    work page 2022

  14. [14]

    Solving inverse problems in medical imaging with score-based generative models.arXiv preprint arXiv:2111.08005, 2021

    Y . Song, L. Shen, L. Xing, and S. Ermon, “Solving Inverse Problems in Medical Imaging with Score-Based Generative Models,” Jun. 2022, arXiv:2111.08005 [eess]. [Online]. Available: http://arxiv.org/abs/2111. 08005

    work page arXiv 2022

  15. [15]

    Deep Learning Techniques for Inverse Problems in Imaging,

    G. Ongie, A. Jalal, C. A. Metzler, R. G. Baraniuk, A. G. Dimakis, and R. Willett, “Deep Learning Techniques for Inverse Problems in Imaging,” IEEE Journal on Selected Areas in Information Theory, vol. 1, no. 1, pp. 39–56, May 2020. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/9084378

    work page arXiv 2020

  16. [16]

    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,” in IEEE Global Conference on Signal and Information Processing , pp. 945–948. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/6737048

    work page arXiv

  17. [17]

    The Little Engine That Could: Regularization by Denoising (RED),

    Y . Romano, M. Elad, and P. Milanfar, “The Little Engine That Could: Regularization by Denoising (RED),” SIAM Journal on Imaging Sciences, vol. 10, no. 4, pp. 1804–1844, Jan. 2017. [Online]. Available: https://epubs.siam.org/doi/abs/10.1137/16M1102884

    work page doi:10.1137/16m1102884 2017

  18. [18]

    DeepRED: Deep Image Prior Powered by RED,

    G. Mataev, M. Elad, and P. Milanfar, “DeepRED: Deep Image Prior Powered by RED,” Oct. 2019, arXiv:1903.10176 [cs]. [Online]. Available: http://arxiv.org/abs/1903.10176

    work page arXiv 2019

  19. [19]

    Compressed Sensing using Generative Models,

    A. Bora, A. Jalal, E. Price, and A. G. Dimakis, “Compressed Sensing using Generative Models,” in Proceedings of the 34th International Conference on Machine Learning . PMLR, Jul. 2017, pp. 537–546. [Online]. Available: https://proceedings.mlr.press/v70/bora17a.html

    work page 2017

  20. [20]

    Intermediate Layer Optimization for Inverse Problems using Deep Generative Models,

    G. Daras, J. Dean, A. Jalal, and A. G. Dimakis, “Intermediate Layer Optimization for Inverse Problems using Deep Generative Models,” Feb. 2021, arXiv:2102.07364 [cs]. [Online]. Available: http://arxiv.org/abs/2102.07364

    work page arXiv 2021

  21. [21]

    PULSE: Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models,

    S. Menon, A. Damian, S. Hu, N. Ravi, and C. Rudin, “PULSE: Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models,” 2020, pp. 2437–2445. [Online]. Available: https://openaccess.thecvf.com/content CVPR 2020/html/ Menon PULSE Self-Supervised Photo Upsampling via Latent Space Exploration of Generative CVPR 2020 paper.html

    work page 2020

  22. [22]

    An Online Plug-and-Play Algorithm for Regularized Image Reconstruction,

    Y . Sun, B. Wohlberg, and U. S. Kamilov, “An Online Plug-and-Play Algorithm for Regularized Image Reconstruction,” IEEE Transactions on Computational Imaging , vol. 5, no. 3, pp. 395–408, Sep

    work page

  23. [23]

    Available: https://ieeexplore.ieee.org/abstract/document/ 8616843

    [Online]. Available: https://ieeexplore.ieee.org/abstract/document/ 8616843

    work page

  24. [24]

    Stochastic Solutions for Linear Inverse Problems using the Prior Implicit in a Denoiser,

    Z. Kadkhodaie and E. Simoncelli, “Stochastic Solutions for Linear Inverse Problems using the Prior Implicit in a Denoiser,” in Advances in Neural Information Processing Systems , vol. 34, 2021, pp. 13 242– 13 254. [Online]. Available: https://proceedings.neurips.cc/paper/2021/ hash/6e28943943dbed3c7f82fc05f269947a-Abstract.html

    work page 2021

  25. [25]

    Instance- Optimal Compressed Sensing via Posterior Sampling,

    A. Jalal, S. Karmalkar, A. G. Dimakis, and E. Price, “Instance- Optimal Compressed Sensing via Posterior Sampling,” Jun. 2021, arXiv:2106.11438 [cs]. [Online]. Available: http://arxiv.org/abs/2106. 11438

    work page arXiv 2021

  26. [26]

    SNIPS: Solving Noisy Inverse Problems Stochastically,

    B. Kawar, G. Vaksman, and M. Elad, “SNIPS: Solving Noisy Inverse Problems Stochastically,” in Advances in Neural Information Processing Systems , vol. 34, 2021, pp. 21 757– 21 769. [Online]. Available: https://proceedings.neurips.cc/paper files/ paper/2021/hash/b5c01503041b70d41d80e3dbe31bbd8c-Abstract.html

    work page 2021

  27. [27]

    Stochastic Image Denoising by Sampling From the Posterior Distribution,

    ——, “Stochastic Image Denoising by Sampling From the Posterior Distribution,” 2021, pp. 1866–1875. [Online]. Available: https://openaccess.thecvf.com/content/ICCV2021W/AIM/html/Kawar Stochastic Image Denoising by Sampling From the Posterior Distribution ICCVW 2021 paper.html

    work page 2021

  28. [28]

    Come-Closer-Diffuse-Faster: Accelerating Conditional Diffusion Models for Inverse Problems through Stochastic Contraction,

    H. Chung, B. Sim, and J. C. Ye, “Come-Closer-Diffuse-Faster: Accelerating Conditional Diffusion Models for Inverse Problems through Stochastic Contraction,” in Proc. of IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , Jun. 2022, pp. 12 403–12 412. [Online]. Available: https://ieeexplore.ieee.org/document/9879691/

    work page arXiv 2022

  29. [29]

    Palette: Image-to-Image Diffusion Models,

    C. Saharia, W. Chan, H. Chang, C. Lee, J. Ho, T. Salimans, D. Fleet, and M. Norouzi, “Palette: Image-to-Image Diffusion Models,” in ACM SIGGRAPH Conference Proceedings . Association for Computing Machinery, Jul. 2022, pp. 1–10. [Online]. Available: https://dl.acm.org/doi/10.1145/3528233.3530757

    work page doi:10.1145/3528233.3530757 2022

  30. [30]

    Image Super-Resolution via Iterative Refinement,

    C. Saharia, J. Ho, W. Chan, T. Salimans, D. J. Fleet, and M. Norouzi, “Image Super-Resolution via Iterative Refinement,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 4, pp. 4713–4726, Apr. 2023. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/9887996

    work page arXiv 2023

  31. [31]

    Bertero, P

    M. Bertero, P. Boccacci, and C. D. Mol, Introduction to Inverse Problems in Imaging , 2nd ed. Boca Raton: CRC Press, Dec. 2021

    work page 2021

  32. [32]

    Deep Learning and Inverse Problems,

    A. Mohammad-Djafari, N. Chu, L. Wang, and L. Yu, “Deep Learning and Inverse Problems,” Sep. 2023, arXiv:2309.00802 [cs]. [Online]. Available: http://arxiv.org/abs/2309.00802

    work page arXiv 2023

  33. [33]

    Inverse problems in astronomy,

    I. Craig and J. Brown, “Inverse problems in astronomy,” Jan. 1986, publisher: Adam Hilger Ltd.,Accord, MA. [Online]. Available: https://www.osti.gov/biblio/5734250

    work page arXiv 1986

  34. [34]

    Inverse problems: From regularization to Bayesian inference,

    D. Calvetti and E. Somersalo, “Inverse problems: From regularization to Bayesian inference,” WIREs Computational Statistics , vol. 10, no. 3, p. e1427, 2018

    work page 2018

  35. [35]

    Reverse-time diffusion equation models,

    B. D. O. Anderson, “Reverse-time diffusion equation models,” Stochastic Processes and their Applications , vol. 12, no. 3, pp. 313–326, May 1982. [Online]. Available: https://www.sciencedirect. com/science/article/pii/0304414982900515

    work page arXiv 1982

  36. [36]

    Approximating the Kullback Leibler Divergence Between Gaussian Mixture Models,

    J. R. Hershey and P. A. Olsen, “Approximating the Kullback Leibler Divergence Between Gaussian Mixture Models,” in Proc. of IEEE International Conference on Acoustics, Speech and Signal Processing, vol. 4, Apr. 2007, pp. IV–317–IV–320. [Online]. Available: https://ieeexplore.ieee.org/document/4218101

    work page arXiv 2007

  37. [37]

    ImageNet Large Scale Visual Recognition Challenge,

    O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei, “ImageNet Large Scale Visual Recognition Challenge,” International Journal of Computer Vision , vol. 115, no. 3, pp. 211–252, Dec. 2015. [Online]. Available: https://doi.org/10.1007/ s11263-015-0816-y

    work page 2015