pith. sign in

arxiv: 2606.21099 · v1 · pith:OWHWVJP3new · submitted 2026-06-19 · 💻 cs.CV

ShuffleFlow: Scalable Posterior Inference for Bayesian Inverse Imaging

Tianao Li , Tjitske Starkenburg , Yu Sun , Emma Alexander This is my paper

Pith reviewed 2026-06-26 14:52 UTC · model grok-4.3

classification 💻 cs.CV
keywords variational inferencenormalizing flowsinverse imagingBayesian reconstructionneural fieldsposterior samplingscalable inference
0
0 comments X

The pith

ShuffleFlow scales variational inference for large image reconstruction by partitioning into sub-image stacks modeled with a shared conditional normalizing flow conditioned on neural field features.

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

The paper tries to solve the scalability barrier that prevents flow-based variational inference from handling large images in scientific inverse problems. Standard flow networks become too expensive in memory and compute as pixel count grows, limiting their use for principled posterior sampling. ShuffleFlow splits each image into a stack of sub-images via pixel-unshuffling, encodes the corresponding coordinates with a neural field, and routes the stack through one shared conditional normalizing flow whose latent variables are tied across channels. A sympathetic reader would care because this would let practitioners draw thousands of posterior samples quickly enough to quantify uncertainty in real imaging tasks where diffusion-based alternatives remain slow.

Core claim

By partitioning an image into a stack of sub-images with pixel-unshuffling, embedding the sample locations via a neural field, and modeling the joint distribution of the stack with a single conditional normalizing flow whose latent variable is shared across channels, the framework produces a scalable posterior estimator that works for both linear and nonlinear Bayesian inverse imaging problems and generates high-sample-count posteriors faster than diffusion samplers.

What carries the argument

Pixel-unshuffling into a sub-image stack together with a neural field feature encoder and a shared conditional normalizing flow with shared latent variables.

If this is right

  • The approach applies directly to both linear and nonlinear imaging inverse problems.
  • It produces high-sample-count posteriors more rapidly than diffusion samplers.
  • It can incorporate score-based or classic priors without changing the core architecture.
  • It extends the practical size of images for which full posterior inference remains feasible.

Where Pith is reading between the lines

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

  • The same unshuffling-plus-shared-flow pattern could be tested on video or volumetric data where full joint flows are likewise intractable.
  • Coordinate-based conditioning may prove useful in other high-dimensional generative tasks that currently rely on full-image networks.
  • Direct comparison against exact posteriors on toy problems would give a quantitative bound on the approximation error introduced by the factorization.

Load-bearing premise

That modeling the joint distribution of the unshuffled sub-image stack with one shared flow conditioned only on neural field features will retain enough spatial structure and channel correlations to represent the true posterior accurately.

What would settle it

On a small inverse problem whose exact posterior is known analytically or by exhaustive computation, draw samples from ShuffleFlow and test whether their empirical distribution matches the true posterior within a chosen statistical distance such as maximum mean discrepancy or total variation.

Figures

Figures reproduced from arXiv: 2606.21099 by Emma Alexander, Tianao Li, Tjitske Starkenburg, Yu Sun.

Figure 1
Figure 1. Figure 1: Overview of ShuffleFlow. (a) Overall pipeline of ShuffleFlow. The full N × N image is partitioned into S 2 sub-images with shape N S × N S using pixel-unshuffling with a pre-defined downsample factor S, and feature embeddings {hi}i=1,...,S2 are calculated from the sampling offset positions using a coordinate-based MLP (NF encoder), implemented with SIREN [21]. We then draw a base sample z0 with dimension o… view at source ↗
Figure 2
Figure 2. Figure 2: Scalability comparison. (a) VI methods scale dramatically better than diffusion samplers for high-resolution posterior estimation. (b,c,d) Flow network size, optimization time, and VRAM vs. image dimension across VI methods. When the computation is dominated by the flow network (TV prior, dashed lines), our method (purple) shows the theoretically predicted growth trend compared to DPI (black). Even when an… view at source ↗
Figure 3
Figure 3. Figure 3: Gaussian toy example with analytical posterior. We show that the posterior learned by ShuffleFlow matches the analytical posterior on the level of mean and standard deviation (left) as well as pixel value distributions (right). Vertical dashed lines show the mean pixel values [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual samples on linear inverse problems with score prior: motion deblurring (a) and compressed sensing MRI (b). We show the mean image computed from 128 posterior samples in row one and the scatter plot of absolute pixel errors vs. standard deviations in row two. Our method shows superior performance on image quality and uncertainty among VI methods and is comparable to the SOTA diffusion sampler DAPS [8… view at source ↗
Figure 5
Figure 5. Figure 5: Results on nonlinear Fourier phase retrieval. (a) We draw 128 samples from each method and filter the samples into different modes if a bimodal posterior exists. We show the mean image from each mode, and PSNR, SSIM, and LPIPS are calculated with respect to the ground truth of each mode. Our method correctly captures the bimodal posterior among VI methods with the best image quality. DAPS [8] stands out wi… view at source ↗
Figure 4
Figure 4. Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
read the original abstract

Variational inference (VI) is a powerful method for principled posterior inference for scientific inverse imaging. VI learns the posterior distribution, often with a flow-based network, which can cheaply generate posterior samples upon optimization, and can flexibly incorporate score-based or classic priors. However, its application to large-scale image reconstruction is severely hindered by the poor scalability of the flow-based networks. In this work, we introduce ShuffleFlow, a scalable VI framework to address this challenge. Our method breaks down the problem into three parts: a pixel-unshuffling-based image coordinate sampler, a neural field as feature encoder, and a conditional normalizing flow (CNF) as posterior estimator. Specifically, our framework partitions an image into a stack of sub-images with pixel-unshuffling and uses a shared CNF to model the joint distribution of the sub-image stack. We condition the CNF on the output of a neural field, which embeds feature vectors corresponding to pixel-unshuffling sample locations to capture spatial structures, and share the flow's latent variable across the channels to model their correlations. We demonstrate our method's effectiveness and efficiency on both linear and nonlinear imaging inverse problems, and show its ability to more rapidly generate a high-sample-count posterior than diffusion samplers.

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 / 0 minor

Summary. The manuscript introduces ShuffleFlow, a scalable variational inference framework for Bayesian inverse imaging. It decomposes the problem via pixel-unshuffling to form a stack of sub-images, encodes spatial features with a neural field, and employs a single conditional normalizing flow (CNF) applied to the stack, conditioned on the neural-field outputs and using a shared latent variable across channels to capture correlations. The authors claim the method is effective and efficient on both linear and nonlinear inverse problems and generates high-sample-count posteriors more rapidly than diffusion samplers.

Significance. If the central architectural assumptions hold and the method recovers accurate posteriors, it would address a key scalability barrier in flow-based VI for large images and offer a practical alternative to diffusion sampling when many posterior draws are needed. The combination of unshuffling, neural fields, and shared-latent CNF is a concrete proposal for factoring the modeling problem.

major comments (2)
  1. [Abstract] Abstract: the effectiveness and efficiency claims for both linear and nonlinear imaging inverse problems are stated without any equations, experimental results, error bars, baselines, or validation details, so the claims cannot be assessed from the provided text.
  2. [Abstract] Abstract (method description): the claim that the joint posterior is accurately recovered by applying a shared CNF to the unshuffled sub-image stack, conditioned only on neural-field features at sample locations and a single shared latent z, requires justification for nonlinear forward operators. Inter-pixel and inter-channel posterior dependencies are typically non-stationary and high-order; the implicit factorization via translation-equivariant unshuffling, shared flow, and broadcast z may lose expressivity and produce biased posteriors even if the ELBO is optimized, directly affecting both the effectiveness claim and the comparison to diffusion baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful comments on our manuscript. We address each major comment below, providing clarifications and indicating where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the effectiveness and efficiency claims for both linear and nonlinear imaging inverse problems are stated without any equations, experimental results, error bars, baselines, or validation details, so the claims cannot be assessed from the provided text.

    Authors: We agree that the abstract, by design, offers a high-level summary without quantitative details or equations. The full manuscript presents the requested elements—including experimental results with error bars, baselines, and validation metrics for both linear and nonlinear problems—in the Experiments and Results sections. We will revise the abstract to include a brief reference to key quantitative findings to improve standalone readability. revision: partial

  2. Referee: [Abstract] Abstract (method description): the claim that the joint posterior is accurately recovered by applying a shared CNF to the unshuffled sub-image stack, conditioned only on neural-field features at sample locations and a single shared latent z, requires justification for nonlinear forward operators. Inter-pixel and inter-channel posterior dependencies are typically non-stationary and high-order; the implicit factorization via translation-equivariant unshuffling, shared flow, and broadcast z may lose expressivity and produce biased posteriors even if the ELBO is optimized, directly affecting both the effectiveness claim and the comparison to diffusion baselines.

    Authors: The manuscript justifies the approach through the neural field's spatially varying features, which adapt to non-stationary structures, combined with the shared latent variable that explicitly couples channels. Empirical results on nonlinear inverse problems (detailed in the Experiments section) show posterior accuracy comparable to diffusion baselines across multiple metrics, indicating that the factorization does not introduce substantial bias in the evaluated settings. We will expand the discussion of modeling assumptions and limitations for nonlinear operators in the revised manuscript. revision: partial

Circularity Check

0 steps flagged

No circularity: architectural construction with no derivation chain or self-referential reductions

full rationale

The paper introduces ShuffleFlow as an architectural framework combining pixel-unshuffling, a neural field feature encoder, and a shared conditional normalizing flow. No equations, derivations, or parameter-fitting steps are described that reduce a claimed prediction or result back to its own inputs by construction. Claims rest on the empirical performance of the proposed model rather than any mathematical identity or self-citation load-bearing uniqueness theorem. This is a standard non-circular presentation of a new VI architecture.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no information on free parameters, background axioms, or new postulated entities.

pith-pipeline@v0.9.1-grok · 5751 in / 1135 out tokens · 17522 ms · 2026-06-26T14:52:30.761756+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

91 extracted references · 4 linked inside Pith

  1. [1]

    Solving inverse problems in medical imaging with score-based generative models,

    Y. Song, L. Shen, L. Xing, and S. Ermon, “Solving inverse problems in medical imaging with score-based generative models,” in International Conference on Learning Representations, 2022. [Online]. Available: https://openreview.net/forum?id=vaRCHVj0uGI

    2022

  2. [2]

    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,”Advances in Neural Information Processing Systems, vol. 34, pp. 14 938–14 954, 2021

    2021

  3. [3]

    Denoising diffusion restoration models,

    B. Kawar, M. Elad, S. Ermon, and J. Song, “Denoising diffusion restoration models,”Advances in Neural Information Processing Sys- tems, vol. 35, pp. 23 593–23 606, 2022

    2022

  4. [4]

    Diffusion models as plug-and-play priors,

    A. Graikos, N. Malkin, N. Jojic, and D. Samaras, “Diffusion models as plug-and-play priors,”Advances in Neural Information Processing Systems, vol. 35, pp. 14 715–14 728, 2022

    2022

  5. [5]

    Dif- fusion posterior sampling for general noisy inverse problems,

    H. Chung, J. Kim, M. T. McCann, M. L. Klasky, and J. C. Ye, “Dif- fusion posterior sampling for general noisy inverse problems,” in 11th International Conference on Learning Representations, ICLR 2023, 2023

    2023

  6. [6]

    A variational perspective on solving inverse problems with diffusion models,

    M. Mardani, J. Song, J. Kautz, and A. Vahdat, “A variational perspective on solving inverse problems with diffusion models,” inThe Twelfth International Conference on Learning Representations,

  7. [7]

    Available: https://openreview.net/forum?id= 1YO4EE3SPB

    [Online]. Available: https://openreview.net/forum?id= 1YO4EE3SPB

  8. [8]

    Principled probabilistic imaging using diffusion models as plug- and-play priors,

    Z. Wu, Y. Sun, Y. Chen, B. Zhang, Y. Yue, and K. Bouman, “Principled probabilistic imaging using diffusion models as plug- and-play priors,”Advances in Neural Information Processing Systems, vol. 37, pp. 118 389–118 427, 2024

    2024

  9. [9]

    Improving diffusion inverse problem solving with de- coupled noise annealing,

    B. Zhang, W. Chu, J. Berner, C. Meng, A. Anandkumar, and Y. Song, “Improving diffusion inverse problem solving with de- coupled noise annealing,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 20 895–20 905

    2025

  10. [10]

    Posterior samples of source galaxies in strong gravitational lenses with score-based priors,

    A. Adam, A. Coogan, N. Malkin, R. Legin, L. Perreault-Levasseur, Y. Hezaveh, and Y. Bengio, “Posterior samples of source galaxies in strong gravitational lenses with score-based priors,”arXiv preprint arXiv:2211.03812, 2022

    arXiv 2022

  11. [11]

    Strong- lensing source reconstruction with denoising diffusion restoration models,

    K. Karchev, N. A. Montel, A. Coogan, and C. Weniger, “Strong- lensing source reconstruction with denoising diffusion restoration models,”arXiv preprint arXiv:2211.04365, 2022

    arXiv 2022

  12. [12]

    Event-horizon-scale imaging of m87* under different assumptions via deep generative image priors,

    B. T. Feng, K. L. Bouman, and W. T. Freeman, “Event-horizon-scale imaging of m87* under different assumptions via deep generative image priors,”arXiv preprint arXiv:2406.02785, 2024

    arXiv 2024

  13. [13]

    Blind strong gravitational lensing inversion: Joint inference of source and lens mass with score-based models,

    G. M. Barco, R. Legin, C. Stone, Y. Hezaveh, and L. Perreault- Levasseur, “Blind strong gravitational lensing inversion: Joint inference of source and lens mass with score-based models,”arXiv preprint arXiv:2511.04792, 2025

    arXiv 2025

  14. [14]

    Deep probabilistic imaging: Uncer- tainty quantification and multi-modal solution characterization for computational imaging,

    H. Sun and K. L. Bouman, “Deep probabilistic imaging: Uncer- tainty quantification and multi-modal solution characterization for computational imaging,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 3, 2021, pp. 2628–2637

    2021

  15. [15]

    Score-based diffusion models as principled priors for inverse imaging,

    B. T. Feng, J. Smith, M. Rubinstein, H. Chang, K. L. Bouman, and W. T. Freeman, “Score-based diffusion models as principled priors for inverse imaging,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 10 520–10 531

    2023

  16. [16]

    Variational bayesian imaging with an efficient surrogate score-based prior,

    B. Feng and K. Bouman, “Variational bayesian imaging with an efficient surrogate score-based prior,”Transactions on Machine Learning Research, 2024. [Online]. Available: https: //openreview.net/forum?id=db2pFKVcm1

    2024

  17. [17]

    α-deep probabilistic inference (α-dpi): efficient uncertainty quan- tification from exoplanet astrometry to black hole feature extrac- tion,

    H. Sun, K. L. Bouman, P . Tiede, J. J. Wang, S. Blunt, and D. Mawet, “α-deep probabilistic inference (α-dpi): efficient uncertainty quan- tification from exoplanet astrometry to black hole feature extrac- tion,”The Astrophysical Journal, vol. 932, no. 2, p. 99, 2022

    2022

  18. [18]

    On the robustness of normal- izing flows for inverse problems in imaging,

    S. Hong, I. Park, and S. Y. Chun, “On the robustness of normal- izing flows for inverse problems in imaging,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 10 745–10 755

    2023

  19. [19]

    Gaussianization flows,

    C. Meng, Y. Song, J. Song, and S. Ermon, “Gaussianization flows,” inInternational Conference on Artificial Intelligence and Statistics. PMLR, 2020, pp. 4336–4345

    2020

  20. [20]

    Stochastic neural radiance fields: Quantifying uncertainty in implicit 3d rep- resentations,

    J. Shen, A. Ruiz, A. Agudo, and F. Moreno-Noguer, “Stochastic neural radiance fields: Quantifying uncertainty in implicit 3d rep- resentations,” in2021 International Conference on 3D Vision (3DV). IEEE, 2021, pp. 972–981

    2021

  21. [21]

    Conditional- flow nerf: Accurate 3d modelling with reliable uncertainty quan- tification,

    J. Shen, A. Agudo, F. Moreno-Noguer, and A. Ruiz, “Conditional- flow nerf: Accurate 3d modelling with reliable uncertainty quan- tification,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 540–557

    2022

  22. [22]

    Implicit neural representations with periodic activation func- tions,

    V . Sitzmann, J. Martel, A. Bergman, D. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation func- tions,”Advances in neural information processing systems, vol. 33, pp. 7462–7473, 2020

    2020

  23. [23]

    On the local behavior of spaces of natural images,

    G. Carlsson, T. Ishkhanov, V . De Silva, and A. Zomorodian, “On the local behavior of spaces of natural images,”International jour- nal of computer vision, vol. 76, no. 1, pp. 1–12, 2008

    2008

  24. [24]

    Testing the manifold hypothesis,

    C. Fefferman, S. Mitter, and H. Narayanan, “Testing the manifold hypothesis,”Journal of the American Mathematical Society, vol. 29, no. 4, pp. 983–1049, 2016

    2016

  25. [25]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network,

    W. Shi, J. Caballero, F. Husz ´ar, J. Totz, A. P . Aitken, R. Bishop, D. Rueckert, and Z. Wang, “Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 1874–1883

    2016

  26. [26]

    Frame-recurrent video super-resolution,

    M. S. Sajjadi, R. Vemulapalli, and M. Brown, “Frame-recurrent video super-resolution,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 6626–6634

    2018

  27. [27]

    Self-guided network for fast image denoising,

    S. Gu, Y. Li, L. V . Gool, and R. Timofte, “Self-guided network for fast image denoising,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 2511–2520

    2019

  28. [28]

    Nonlinear total variation based noise removal algorithms,

    L. I. Rudin, S. Osher, and E. Fatemi, “Nonlinear total variation based noise removal algorithms,”Physica D: nonlinear phenomena, vol. 60, no. 1-4, pp. 259–268, 1992

    1992

  29. [29]

    Sparsity and incoherence in compres- sive sampling,

    E. Candes and J. Romberg, “Sparsity and incoherence in compres- sive sampling,”Inverse problems, vol. 23, no. 3, p. 969, 2007

    2007

  30. [30]

    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. IEEE, 2013, pp. 945–948

    2013

  31. [31]

    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, 2016

    2016

  32. [32]

    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 Sci- ences, vol. 10, no. 4, pp. 1804–1844, 2017

    2017

  33. [33]

    Learned reconstructions for practical mask-based lens- less imaging,

    K. Monakhova, J. Yurtsever, G. Kuo, N. Antipa, K. Yanny, and L. Waller, “Learned reconstructions for practical mask-based lens- less imaging,”Optics express, vol. 27, no. 20, pp. 28 075–28 090, 2019

    2019

  34. [34]

    Scalable plug-and-play admm with convergence guarantees,

    Y. Sun, Z. Wu, X. Xu, B. Wohlberg, and U. S. Kamilov, “Scalable plug-and-play admm with convergence guarantees,”IEEE Trans- actions on Computational Imaging, vol. 7, pp. 849–863, 2021

    2021

  35. [35]

    Galaxy image deconvolution for weak gravitational lensing with unrolled plug-and-play admm,

    T. Li and E. Alexander, “Galaxy image deconvolution for weak gravitational lensing with unrolled plug-and-play admm,” Monthly Notices of the Royal Astronomical Society: Letters, vol. 522, no. 1, pp. L31–L35, 2023. 10

    2023

  36. [36]

    Generative modeling by estimating gra- dients of the data distribution,

    Y. Song and S. Ermon, “Generative modeling by estimating gra- dients of the data distribution,”Advances in neural information processing systems, vol. 32, 2019

    2019

  37. [37]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P . Abbeel, “Denoising diffusion probabilistic models,”Advances in neural information processing systems, vol. 33, pp. 6840–6851, 2020

    2020

  38. [38]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,”arXiv preprint arXiv:2010.02502, 2020

    Pith/arXiv arXiv 2010

  39. [39]

    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,”arXiv preprint arXiv:2011.13456, 2020

    Pith/arXiv arXiv 2011

  40. [40]

    Score-based diffusion models for acceler- ated mri,

    H. Chung and J. C. Ye, “Score-based diffusion models for acceler- ated mri,”Medical image analysis, vol. 80, p. 102479, 2022

    2022

  41. [41]

    Improving diffusion mod- els for inverse problems using manifold constraints,

    H. Chung, B. Sim, D. Ryu, and J. C. Ye, “Improving diffusion mod- els for inverse problems using manifold constraints,”Advances in Neural Information Processing Systems, vol. 35, pp. 25 683–25 696, 2022

    2022

  42. [42]

    Neural fields in visual computing and beyond,

    Y. Xie, T. Takikawa, S. Saito, O. Litany, S. Yan, N. Khan, F. Tombari, J. Tompkin, V . Sitzmann, and S. Sridhar, “Neural fields in visual computing and beyond,” inComputer Graphics Forum, vol. 41, no. 2. Wiley Online Library, 2022, pp. 641–676

    2022

  43. [43]

    Nerf in the wild: Neural radiance fields for unconstrained photo collections,

    R. Martin-Brualla, N. Radwan, M. S. Sajjadi, J. T. Barron, A. Doso- vitskiy, and D. Duckworth, “Nerf in the wild: Neural radiance fields for unconstrained photo collections,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 7210–7219

    2021

  44. [44]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P . P . Srinivasan, M. Tancik, J. T. Barron, R. Ra- mamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

    2021

  45. [45]

    Mip-nerf: A multiscale represen- tation for anti-aliasing neural radiance fields,

    J. T. Barron, B. Mildenhall, M. Tancik, P . Hedman, R. Martin- Brualla, and P . P . Srinivasan, “Mip-nerf: A multiscale represen- tation for anti-aliasing neural radiance fields,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 5855–5864

    2021

  46. [46]

    Coil: Coordinate-based internal learning for tomographic imaging,

    Y. Sun, J. Liu, M. Xie, B. Wohlberg, and U. S. Kamilov, “Coil: Coordinate-based internal learning for tomographic imaging,” IEEE Transactions on Computational Imaging, vol. 7, pp. 1400–1412, 2021

    2021

  47. [47]

    Intratomo: self-supervised learning-based tomography via sinogram synthe- sis and prediction,

    G. Zang, R. Idoughi, R. Li, P . Wonka, and W. Heidrich, “Intratomo: self-supervised learning-based tomography via sinogram synthe- sis and prediction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 1960–1970

    2021

  48. [48]

    Recovery of continuous 3d refractive index maps from discrete intensity-only measurements using neural fields,

    R. Liu, Y. Sun, J. Zhu, L. Tian, and U. S. Kamilov, “Recovery of continuous 3d refractive index maps from discrete intensity-only measurements using neural fields,”Nature Machine Intelligence, vol. 4, no. 9, pp. 781–791, 2022

    2022

  49. [49]

    Nerp: implicit neural representa- tion learning with prior embedding for sparsely sampled image reconstruction,

    L. Shen, J. Pauly, and L. Xing, “Nerp: implicit neural representa- tion learning with prior embedding for sparsely sampled image reconstruction,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 1, pp. 770–782, 2022

    2022

  50. [50]

    Implicit neural representation in medical imaging: A comparative survey,

    A. Molaei, A. Aminimehr, A. Tavakoli, A. Kazerouni, B. Azad, R. Azad, and D. Merhof, “Implicit neural representation in medical imaging: A comparative survey,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 2381–2391

    2023

  51. [51]

    Fourier ptychographic microscopy image stack re- construction using implicit neural representations,

    H. Zhou, B. Y. Feng, H. Guo, S. S. Lin, M. Liang, C. A. Metzler, and C. Yang, “Fourier ptychographic microscopy image stack re- construction using implicit neural representations,”Optica, vol. 10, no. 12, pp. 1679–1687, 2023

    2023

  52. [52]

    Nesvor: implicit neural representation for slice-to-volume reconstruction in mri,

    J. Xu, D. Moyer, B. Gagoski, J. E. Iglesias, P . E. Grant, P . Golland, and E. Adalsteinsson, “Nesvor: implicit neural representation for slice-to-volume reconstruction in mri,”IEEE transactions on medical imaging, vol. 42, no. 6, pp. 1707–1719, 2023

    2023

  53. [53]

    Neural space–time model for dynamic multi-shot imaging,

    R. Cao, N. S. Divekar, J. K. Nu ˜nez, S. Upadhyayula, and L. Waller, “Neural space–time model for dynamic multi-shot imaging,”Na- ture Methods, pp. 1–6, 2024

    2024

  54. [54]

    Coordinate-based speed of sound recovery for aberration-corrected photoacoustic computed tomography,

    T. Li, M. Cui, C. Ma, and E. Alexander, “Coordinate-based speed of sound recovery for aberration-corrected photoacoustic computed tomography,” inProceedings of the IEEE/CVF International Confer- ence on Computer Vision, 2025, pp. 27 466–27 475

    2025

  55. [55]

    Variational inference with normal- izing flows,

    D. Rezende and S. Mohamed, “Variational inference with normal- izing flows,” inInternational conference on machine learning. PMLR, 2015, pp. 1530–1538

    2015

  56. [56]

    Normalizing flows for probabilistic mod- eling and inference,

    G. Papamakarios, E. Nalisnick, D. J. Rezende, S. Mohamed, and B. Lakshminarayanan, “Normalizing flows for probabilistic mod- eling and inference,”Journal of Machine Learning Research, vol. 22, no. 57, pp. 1–64, 2021

    2021

  57. [57]

    Pointflow: 3d point cloud generation with continuous normalizing flows,

    G. Yang, X. Huang, Z. Hao, M.-Y. Liu, S. Belongie, and B. Har- iharan, “Pointflow: 3d point cloud generation with continuous normalizing flows,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 4541–4550

    2019

  58. [58]

    C- flow: Conditional generative flow models for images and 3d point clouds,

    A. Pumarola, S. Popov, F. Moreno-Noguer, and V . Ferrari, “C- flow: Conditional generative flow models for images and 3d point clouds,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 7949–7958

    2020

  59. [59]

    Sliced iterative normalizing flows,

    B. Dai and U. Seljak, “Sliced iterative normalizing flows,” in International Conference on Machine Learning. PMLR, 2021, pp. 2352–2364

    2021

  60. [60]

    Mcmc-based image reconstruction with uncer- tainty quantification,

    J. M. Bardsley, “Mcmc-based image reconstruction with uncer- tainty quantification,”SIAM Journal on Scientific Computing, vol. 34, no. 3, pp. A1316–A1332, 2012

    2012

  61. [61]

    Bayesian imaging using plug & play priors: when langevin meets tweedie,

    R. Laumont, V . D. Bortoli, A. Almansa, J. Delon, A. Durmus, and M. Pereyra, “Bayesian imaging using plug & play priors: when langevin meets tweedie,”SIAM Journal on Imaging Sciences, vol. 15, no. 2, pp. 701–737, 2022

    2022

  62. [62]

    Monte carlo guided diffusion for bayesian linear inverse problems,

    G. Cardoso, Y. J. E. Idrissi, S. L. Corff, and E. Moulines, “Monte carlo guided diffusion for bayesian linear inverse problems,”arXiv preprint arXiv:2308.07983, 2023

    arXiv 2023

  63. [63]

    Provable probabilistic imaging using score-based generative priors,

    Y. Sun, Z. Wu, Y. Chen, B. T. Feng, and K. L. Bouman, “Provable probabilistic imaging using score-based generative priors,”IEEE Transactions on Computational Imaging, 2024

    2024

  64. [64]

    Scalable bayesian un- certainty quantification in imaging inverse problems via convex optimization,

    A. Repetti, M. Pereyra, and Y. Wiaux, “Scalable bayesian un- certainty quantification in imaging inverse problems via convex optimization,”SIAM Journal on Imaging Sciences, vol. 12, no. 1, pp. 87–118, 2019

    2019

  65. [65]

    Variational infer- ence: A review for statisticians,

    D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, “Variational infer- ence: A review for statisticians,”Journal of the American statistical Association, vol. 112, no. 518, pp. 859–877, 2017

    2017

  66. [66]

    R. M. Neal,Bayesian learning for neural networks. Springer Science & Business Media, 2012, vol. 118

    2012

  67. [67]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning,

    Y. Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Representing model uncertainty in deep learning,” ininternational conference on machine learning. PMLR, 2016, pp. 1050–1059

    2016

  68. [68]

    What uncertainties do we need in bayesian deep learning for computer vision?

    A. Kendall and Y. Gal, “What uncertainties do we need in bayesian deep learning for computer vision?”Advances in neural information processing systems, vol. 30, 2017

    2017

  69. [69]

    Reliable deep-learning-based phase imaging with uncertainty quantification,

    Y. Xue, S. Cheng, Y. Li, and L. Tian, “Reliable deep-learning-based phase imaging with uncertainty quantification,”Optica, vol. 6, no. 5, pp. 618–629, 2019

    2019

  70. [70]

    Uncertainr: Uncer- tainty quantification of end-to-end implicit neural representations for computed tomography,

    F. Vasconcelos, B. He, N. Singh, and Y. W. Teh, “Uncertainr: Uncer- tainty quantification of end-to-end implicit neural representations for computed tomography,”arXiv preprint arXiv:2202.10847, 2022

    arXiv 2022

  71. [71]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” Advances in neural information processing systems, vol. 30, 2017

    2017

  72. [72]

    Equivariant bootstrapping for uncer- tainty quantification in imaging inverse problems,

    M. Pereyra and J. Tachella, “Equivariant bootstrapping for uncer- tainty quantification in imaging inverse problems,” inInternational Conference on Artificial Intelligence and Statistics. PMLR, 2024, pp. 4141–4149

    2024

  73. [73]

    Image-to-image regression with distribution-free uncertainty quantification and applications in imaging,

    A. N. Angelopoulos, A. P . Kohli, S. Bates, M. Jordan, J. Malik, T. Alshaabi, S. Upadhyayula, and Y. Romano, “Image-to-image regression with distribution-free uncertainty quantification and applications in imaging,” inInternational Conference on Machine Learning. PMLR, 2022, pp. 717–730

    2022

  74. [74]

    Learned, uncertainty-driven adaptive acqui- sition for photon-efficient scanning microscopy,

    C. T. Ye, J. Han, K. Liu, A. Angelopoulos, L. Griffith, K. Mon- akhova, and S. You, “Learned, uncertainty-driven adaptive acqui- sition for photon-efficient scanning microscopy,”Optics Express, vol. 33, no. 6, pp. 12 269–12 287, 2025

    2025

  75. [75]

    Density estimation using real nvp,

    L. Dinh, J. Sohl-Dickstein, and S. Bengio, “Density estimation using real nvp,”arXiv preprint arXiv:1605.08803, 2016

    Pith/arXiv arXiv 2016

  76. [76]

    On the intrinsic dimen- sionality of image representations,

    S. Gong, V . N. Boddeti, and A. K. Jain, “On the intrinsic dimen- sionality of image representations,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 3987–3996

    2019

  77. [77]

    The intrinsic dimension of images and its impact on learning,

    P . Pope, C. Zhu, A. Abdelkader, M. Goldblum, and T. Goldstein, “The intrinsic dimension of images and its impact on learning,” arXiv preprint arXiv:2104.08894, 2021

    arXiv 2021

  78. [78]

    Understanding representation dynamics of diffusion models via low-dimensional modeling,

    X. Li, Z. Zhang, X. Li, S. Chen, Z. Zhu, P . Wang, and Q. Qu, “Understanding representation dynamics of diffusion models via low-dimensional modeling,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. [Online]. Available: https://openreview.net/forum?id=BE6QmLdJqY

    2025

  79. [79]

    Neighbor2neighbor: Self- supervised denoising from single noisy images,

    T. Huang, S. Li, X. Jia, H. Lu, and J. Liu, “Neighbor2neighbor: Self- supervised denoising from single noisy images,” inProceedings of 11 the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 14 781–14 790

    2021

  80. [80]

    Zero-shot noise2noise: Efficient image denoising without any data,

    Y. Mansour and R. Heckel, “Zero-shot noise2noise: Efficient image denoising without any data,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2023, pp. 14 018– 14 027

    2023

Showing first 80 references.