pith. sign in

arxiv: 2602.21361 · v3 · submitted 2026-02-24 · ⚛️ physics.optics · cs.AI· cs.CV· cs.LG· physics.comp-ph

Towards single-shot coherent imaging via overlap-free ptychography

Oliver Hoidn , Albert Vong , Aashwin Mishra , Steven Henke , Matthew Seaberg This is my paper

Pith reviewed 2026-05-15 19:40 UTC · model grok-4.3

classification ⚛️ physics.optics cs.AIcs.CVcs.LGphysics.comp-ph
keywords coherent diffractive imagingptychographysingle-shot imagingFresnel geometryoverlap-free reconstructiondifferentiable modelPoisson likelihoodneural reconstruction
0
0 comments X

The pith

A differentiable scattering model enables accurate single-shot ptychography without requiring probe overlap.

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

The paper shows that real-space overlap need not act as a hard constraint for ptychographic reconstruction of extended samples. By embedding overlap as a tunable parameter inside a neural forward model that incorporates Poisson statistics for photon detection, the method produces reliable images from single diffraction exposures in Fresnel geometry. A sympathetic reader would care because current synchrotron and XFEL measurements demand dense overlapping scans that limit speed and raise radiation dose. Experimental tests reach an amplitude SSIM of 0.904 without overlaps, compared with 0.968 when overlaps are enforced, while also processing conventional data approximately 40 times faster on GPU. The same framework generalizes to new illumination profiles using far less training data than supervised alternatives.

Core claim

The framework couples a differentiable forward model of coherent scattering with a Poisson photon-counting likelihood; real-space overlap enters as a tunable parameter via coordinate-based grouping rather than a hard requirement. This delivers overlap-free, single-shot reconstructions in Fresnel CDI geometry that remain accurate at low photon counts around 10,000 per frame, achieve experimental amplitude SSIM of 0.904 versus 0.968 for overlap-constrained cases, and generalize to unseen probes, while providing 40 times higher GPU throughput for conventional multi-shot ptychography.

What carries the argument

Differentiable forward model of coherent scattering with Poisson likelihood, where overlap is introduced as a tunable parameter through coordinate-based grouping instead of a hard constraint.

If this is right

  • Reconstructions stay accurate at photon counts around 10,000 per frame on synthetic benchmarks.
  • Overlap-free single-shot cases reach experimental amplitude SSIM of 0.904.
  • GPU throughput improves by a factor of approximately 40 over least-squares maximum-likelihood methods at 128 by 128 resolution.
  • The method exceeds the performance of a supervised model trained on 16 times more data while using only 1,024 images and generalizes to unseen illumination profiles.
  • Single-exposure Fresnel CDI and overlapped ptychography are handled inside the same framework.

Where Pith is reading between the lines

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

  • Fewer total exposures may become feasible for radiation-sensitive samples, lowering overall dose.
  • Scan times at light sources could shorten substantially for extended objects by removing the overlap requirement.
  • The tunable-overlap idea may transfer to other coherent optics inverse problems that currently treat overlap or similar constraints as mandatory.
  • Direct comparison against ground-truth structures at progressively lower photon fluxes would test robustness beyond the reported conditions.

Load-bearing premise

The physical model of coherent scattering and photon counting accurately describes the measured data even without enforced overlaps.

What would settle it

A single non-overlapping experimental reconstruction that fails to recover known sample features or deviates markedly from an overlapped reference at similar photon levels would show the accuracy claim does not hold.

Figures

Figures reproduced from arXiv: 2602.21361 by Aashwin Mishra, Albert Vong, Matthew Seaberg, Oliver Hoidn, Steven Henke.

Figure 1
Figure 1. Figure 1: Reconstruction comparison across probe types and acquisition modes. Rows: [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of reconstruction quality with different numbers of diffraction [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Photon-limited performance for two self-supervised PtychoPINN variants [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Structural similarity of PtychoPINN and the supervised baseline as a function [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of methods for an in-distribution LCLS control (train LCLS XPP, [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
read the original abstract

Ptychographic imaging at synchrotron and XFEL sources requires dense overlapping scans, limiting throughput and increasing dose. Extending coherent diffractive imaging to overlap-free operation on extended samples remains an open problem. Here, we extend PtychoPINN (O. Hoidn \emph{et al.}, \emph{Scientific Reports} \textbf{13}, 22789, 2023) to deliver \emph{overlap-free, single-shot} reconstructions in a Fresnel coherent diffraction imaging (CDI) geometry while also accelerating conventional multi-shot ptychography. The framework couples a differentiable forward model of coherent scattering with a Poisson photon-counting likelihood; real-space overlap enters as a tunable parameter via coordinate-based grouping rather than a hard requirement. On synthetic benchmarks, reconstructions remain accurate at low counts ($\sim\!10^4$ photons/frame), and overlap-free single-shot reconstruction with an experimental probe reaches amplitude structural similarity (SSIM) 0.904, compared with 0.968 for overlap-constrained reconstruction. Against a data-saturated supervised model with the same backbone (16,384 training images), PtychoPINN achieves higher SSIM with only 1,024 images and generalizes to unseen illumination profiles. Per-graphics processing unit (GPU) throughput is approximately $40\times$ that of least-squares maximum-likelihood (LSQ-ML) reconstruction at matched $128\times128$ resolution. These results, validated on experimental data from the Advanced Photon Source and the Linac Coherent Light Source, unify single-exposure Fresnel CDI and overlapped ptychography within one framework, supporting dose-efficient, high-throughput imaging at modern light sources.

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 extends PtychoPINN to enable overlap-free single-shot reconstructions in Fresnel CDI geometry by treating real-space overlap as a tunable grouping parameter in a differentiable coherent-scattering forward model coupled to a Poisson likelihood. It reports accurate low-count synthetic reconstructions, experimental amplitude SSIM of 0.904 for overlap-free single-shot cases (vs. 0.968 overlap-constrained), superior SSIM to a supervised baseline trained on 16,384 images using only 1,024 images, generalization to unseen probes, and ~40x GPU throughput gain over LSQ-ML at 128x128 resolution, validated on APS and LCLS data.

Significance. If the central claims hold, the work could meaningfully increase throughput and reduce dose in ptychography at synchrotrons and XFELs by relaxing the overlap requirement while unifying single-shot Fresnel CDI with overlapped scanning. The explicit physics-informed model, direct comparison against a data-saturated supervised network, and experimental validation on real facility data are notable strengths that support potential impact in coherent imaging.

major comments (2)
  1. [Abstract and experimental validation] Abstract and experimental results: the overlap-free single-shot claim depends on the Poisson forward model supplying sufficient regularization once overlap is removed as a hard constraint. The SSIM drop from 0.968 to 0.904 on experimental data is reported, but no residual maps, error decomposition, or ablation isolating forward-model mismatch (non-Poisson noise, Fresnel propagation inaccuracies, probe aberrations) from network capacity is provided; this directly affects whether the observed fidelity is attributable to the physics model.
  2. [Methods (forward model)] Methods section on forward model: the assumption that the differentiable coherent-scattering operator plus Poisson likelihood remains faithful for extended objects without overlap is load-bearing for uniqueness. The manuscript does not report a controlled test (e.g., synthetic data with known ground truth and deliberately mismatched noise model) to quantify degradation when overlap is set to zero.
minor comments (2)
  1. [Abstract] Abstract: specify the precise resolution and photon-count regime for the 40x throughput comparison to LSQ-ML to allow direct replication.
  2. [Results figures] Figure captions and text: ensure all SSIM values are accompanied by the number of independent realizations or error bars when comparing overlap-free vs. overlapped cases.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and valuable feedback on our work extending PtychoPINN to overlap-free single-shot Fresnel CDI. We address the major comments point by point below, proposing specific revisions to enhance the manuscript.

read point-by-point responses

  1. Referee: [Abstract and experimental validation] Abstract and experimental results: the overlap-free single-shot claim depends on the Poisson forward model supplying sufficient regularization once overlap is removed as a hard constraint. The SSIM drop from 0.968 to 0.904 on experimental data is reported, but no residual maps, error decomposition, or ablation isolating forward-model mismatch (non-Poisson noise, Fresnel propagation inaccuracies, probe aberrations) from network capacity is provided; this directly affects whether the observed fidelity is attributable to the physics model.

    Authors: We agree that providing residual maps and an error decomposition would strengthen the interpretation of the experimental results. In the revised version, we will add residual maps comparing the reconstructed and measured diffraction patterns for both overlap-constrained and overlap-free cases. Additionally, we will include an ablation study that varies the network capacity while keeping the forward model fixed to help isolate the contributions to the SSIM difference. revision: yes

  2. Referee: [Methods (forward model)] Methods section on forward model: the assumption that the differentiable coherent-scattering operator plus Poisson likelihood remains faithful for extended objects without overlap is load-bearing for uniqueness. The manuscript does not report a controlled test (e.g., synthetic data with known ground truth and deliberately mismatched noise model) to quantify degradation when overlap is set to zero.

    Authors: The manuscript does include synthetic reconstructions with known ground truth under the Poisson model at low photon counts, demonstrating robustness. However, we recognize the value of testing under model mismatch. We will add a controlled experiment in the revised methods section using synthetic data with added Gaussian noise to simulate mismatch and report the resulting degradation in reconstruction quality for the overlap-free setting. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation grounded in explicit physics forward model

full rationale

The paper's central framework rests on a differentiable coherent-scattering operator combined with Poisson likelihood, which is a first-principles model rather than a data-driven fit to the target reconstructions. Overlap enters only as a tunable grouping parameter, and the overlap-free claims are supported by direct comparisons to overlap-constrained baselines, a data-saturated supervised network (16,384 images), and experimental data from APS/LCLS. The self-citation to the 2023 PtychoPINN paper supplies the base architecture but does not substitute for the new extension or the reported SSIM/throughput metrics. No equation or claim reduces by construction to its own fitted inputs or to a self-citation chain; the derivation chain remains externally falsifiable.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on a differentiable coherent-scattering forward model and Poisson photon-counting likelihood; no new physical entities are introduced.

free parameters (1)
  • overlap parameter
    Treated as tunable via coordinate-based grouping rather than fixed; exact fitting procedure not detailed in abstract.
axioms (2)
  • domain assumption Poisson photon-counting likelihood accurately models detector statistics
    Standard assumption for photon-limited coherent imaging; invoked in the loss function.
  • domain assumption Differentiable forward model of coherent scattering is faithful to the Fresnel geometry
    Core modeling choice that enables gradient-based training.

pith-pipeline@v0.9.0 · 5621 in / 1465 out tokens · 23209 ms · 2026-05-15T19:40:00.832385+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

25 extracted references · 25 canonical work pages

  1. [1]

    LCLS-II-HE: Design and Performance,

    SLAC National Accelerator Laboratory, “LCLS-II-HE: Design and Performance,” https://lcls.slac.stanford.edu/lcls- ii-he/design-and-performance (2023). Accessed: 2025-08-14

    work page 2023

  2. [2]

    Ptychography: A solution to the phase problem,

    M. Guizar-Sicairos and P. Thibault, “Ptychography: A solution to the phase problem,” Phys. Today74, 42–48 (2021)

    work page 2021

  3. [3]

    Influence of the overlap parameter on the convergence of the ptychographical iterative engine,

    O. Bunk, M. Dierolf, S. Kynde,et al., “Influence of the overlap parameter on the convergence of the ptychographical iterative engine,” Ultramicroscopy108, 481–487 (2008)

    work page 2008

  4. [4]

    An improved ptychographical phase retrieval algorithm for diffractive imaging,

    A. M. Maiden and J. M. Rodenburg, “An improved ptychographical phase retrieval algorithm for diffractive imaging,” Ultramicroscopy109, 1256–1262 (2009)

    work page 2009

  5. [5]

    Sharp: a distributed gpu-based ptychographic solver,

    S. Marchesini, H. Krishnan, B. J. Daurer,et al., “Sharp: a distributed gpu-based ptychographic solver,” J. Appl. Crystallogr.49, 1245–1252 (2016)

    work page 2016

  6. [6]

    Deeplearningattheedgeenablesreal-timestreamingptychographicimaging,

    A.V.Babu,T.Zhou,S.Kandel,etal.,“Deeplearningattheedgeenablesreal-timestreamingptychographicimaging,” Nat. Commun.14, 7059 (2023)

    work page 2023

  7. [7]

    Ai-enabledhigh-resolutionscanningcoherentdiffractionimaging,

    M.J.Cherukara, T.Zhou, Y.S.G.Nashed,etal., “Ai-enabledhigh-resolutionscanningcoherentdiffractionimaging,” Appl. Phys. Lett.117, 044103 (2020)

    work page 2020

  8. [8]

    prdeep: Robust phase retrieval with a flexible deep network,

    C. A. Metzler, P. Schniter, A. Veeraraghavan, and R. G. Baraniuk, “prdeep: Robust phase retrieval with a flexible deep network,” inProceedings of the 35th International Conference on Machine Learning,vol. 80 ofProceedings of Machine Learning Research(2018), pp. 3501–3510

    work page 2018

  9. [9]

    Accelerating iterative ptychography with an integrated neural network,

    A. R. C. McCray, S. M. Ribet, G. Varnavides, and C. Ophus, “Accelerating iterative ptychography with an integrated neural network,” J. Microsc.300, 180–190 (2025)

    work page 2025

  10. [10]

    Implicitneuralrepresentationswithperiodicactivationfunctions,

    V.Sitzmann,J.N.P.Martel,A.W.Bergman,etal.,“Implicitneuralrepresentationswithperiodicactivationfunctions,” inAdvances in Neural Information Processing Systems,vol. 33 (2020), pp. 7462–7473

    work page 2020

  11. [11]

    Predicting ptychography probe positions using single-shot phase retrieval neural network,

    M. Du, T. Zhou, J. Deng,et al., “Predicting ptychography probe positions using single-shot phase retrieval neural network,” Opt. Express32, 36757–36780 (2024)

    work page 2024

  12. [12]

    Ptychodv: Visiontransformer-baseddeepunrollingnetworkforptychographic image reconstruction,

    W.Gan,Q.Zhai,M.T.McCann,etal.,“Ptychodv: Visiontransformer-baseddeepunrollingnetworkforptychographic image reconstruction,” IEEE Open J. Signal Process.5, 539–547 (2024)

    work page 2024

  13. [13]

    Autophasenn: unsupervised physics-aware deep learning of 3d nanoscale bragg coherent diffraction imaging,

    Y. Yao, H. Chan, S. K. R. S. Sankaranarayanan,et al., “Autophasenn: unsupervised physics-aware deep learning of 3d nanoscale bragg coherent diffraction imaging,” npj Comput. Mater.8, 124 (2022)

    work page 2022

  14. [14]

    Maximum-likelihood refinement for coherent diffractive imaging,

    P. Thibault and M. Guizar-Sicairos, “Maximum-likelihood refinement for coherent diffractive imaging,” New J. Phys. 14, 063004 (2012)

    work page 2012

  15. [15]

    Maximum-likelihood ptychography in the presence of poisson–gaussian noise,

    J. P. Seifert, Z. Chen, M.-J. Yoon,et al., “Maximum-likelihood ptychography in the presence of poisson–gaussian noise,” Opt. Lett.48, 4897–4900 (2023)

    work page 2023

  16. [16]

    Fresnel coherent diffractive imaging,

    G. J. Williams, H. M. Quiney, B. B. Dhal,et al., “Fresnel coherent diffractive imaging,” Phys. Rev. Lett.97, 025506 (2006)

    work page 2006

  17. [17]

    Near-field ptychography: phase retrieval for inline holography using a structured illumination,

    M. Stockmar, P. Cloetens, I. Zanette,et al., “Near-field ptychography: phase retrieval for inline holography using a structured illumination,” Sci. Reports3, 1927 (2013)

    work page 1927

  18. [18]

    Single-shot ptychography,

    P. Sidorenko and O. Cohen, “Single-shot ptychography,” Optica3, 9–14 (2016)

    work page 2016

  19. [19]

    Single-shot ptychography at a soft x-ray free-electron laser,

    K. Kharitonov, M. Mehrjoo, M. Ruiz-Lopez,et al., “Single-shot ptychography at a soft x-ray free-electron laser,” Sci. Reports12, 14430 (2022)

    work page 2022

  20. [20]

    Phase retrieval by coherent modulation imaging,

    F. Zhang, I. Peterson, J. Vila-Comamala,et al., “Phase retrieval by coherent modulation imaging,” Nat. Commun.7, 13367 (2016)

    work page 2016

  21. [21]

    Single shot multi-wavelength phase retrieval with coherent modulation imaging,

    X. Dong, X. Pan, C. Liu, and J. Zhu, “Single shot multi-wavelength phase retrieval with coherent modulation imaging,” Opt. Lett.43, 1762–1765 (2018)

    work page 2018

  22. [22]

    Physics constrained unsupervised deep learning for rapid, high resolution scanning coherent diffraction reconstruction,

    O. Hoidn, A. A. Mishra, and A. Mehta, “Physics constrained unsupervised deep learning for rapid, high resolution scanning coherent diffraction reconstruction,” Sci. Reports13, 22789 (2023)

    work page 2023

  23. [23]

    Towards generalizable deep ptychography neural networks,

    A. Vong, S. Henke, O. Hoidn,et al., “Towards generalizable deep ptychography neural networks,” arXiv abs/2509.25104, 1–1 (2025)

    work page arXiv 2025

  24. [24]

    Extending the methodology of x-ray crystallography to allow imaging of micrometre-sized non-crystalline specimens,

    J. Miao, P. Charalambous, J. Kirz, and D. Sayre, “Extending the methodology of x-ray crystallography to allow imaging of micrometre-sized non-crystalline specimens,” Nature400, 342–344 (1999)

    work page 1999

  25. [25]

    Pty-chi: A pytorch-based modern ptychographic data analysis package,

    M. Du, H. Ruth, S. Henke,et al., “Pty-chi: A pytorch-based modern ptychographic data analysis package,” arXiv abs/2510.20929, 1–1 (2025)

    work page arXiv 2025