{"id":"91357362-36e8-4572-bdac-752cc434f6e2","arxiv_id":"2608.01757","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A detector-calibrated confidence map is embedded into the ptychographic amplitude update, yielding improved SNR and near-Rayleigh resolution.","lead":"Ptychographic imaging usually treats the detector as a perfect recording plane. This paper adds a pixel-by-pixel reliability weight, learned from detector calibration, into the reconstruction loop, and reports sharper, less noisy images in three experimental settings.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The quantitative support for the central claim rests on a nonstandard SNR metric and an underspecified FRC protocol; if the FRC is not computed from independent reconstructions, the claimed Rayleigh-limit resolution may reflect noise suppression rather than genuine signal.","rationale":"The paper proposes a reasonable extension of ptychographic reconstruction: weighting the amplitude projection by a calibrated detector reliability map is a natural generalization of weighted least-squares. I do not object to the method's basic architecture, and the qualitative improvements across three imaging modalities are plausible. However, the evidence that would establish the central quantitative claims is not yet strong enough. The SNR metric is nonstandard and appears to conflate background suppression with signal fidelity, and the FRC methodology is underspecified. The reader's weakest assumption focused on the accuracy and stability of the confidence map, which is a related but distinct concern; I identify the metric and validation protocol as the more immediate load-bearing gap. The Discussion's admission that photon-free calibration omits signal-dependent shot noise makes this gap concrete, because low-confidence weighting could suppress genuine weak high-frequency signal, making the resolution and SNR gains overstate the true information content. A synthetic experiment with known ground truth and independent half-dataset FRC would settle the concern: if the method preserves known weak signal and maintains the FRC crossing, the claim stands; if not, the quantitative conclusions should be revised. This does not change the conditional verdict, but it sharpens the condition that the FRC and SNR must be established with standard, independently validated metrics.","tokens_in":12795,"tokens_out":4683,"duration_ms":46410,"concrete_test":"Run a synthetic ptychography simulation with a known complex object, realistic Poisson plus readout noise, and the same w(q,t_exp) calibration pipeline. Reconstruct with plain mPIE, DFS, and the proposed method using two independent halves of the scan positions. Compute (i) standard SNR and PSNR of the reconstructions against the known ground truth and (ii) FRC between the two half-dataset reconstructions. If the proposed method's 1/2-bit FRC crossing falls below the claimed 0.65 x lambda/NA or its ground-truth PSNR is not better than DFS, then the reported twofold SNR and k-factor claims are metric artifacts rather than genuine resolution improvements.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that detector-informed weighting yields 'an approximately twofold SNR enhancement' and 'reconstruction approaching the Rayleigh limit' depends on two metrics that are not fully specified. The SNR reported in Sections 2.2 and 2.3 is defined as the ratio of the mean-square intensity of the full diffraction frame to that of a signal-free background region. This definition rewards any background suppression, regardless of whether weak high-frequency diffraction signal is preserved, and it is not a standard signal-to-noise ratio for reconstruction quality. The reported values (65.11 dB vs 33.27 dB for DFS, and 'twofold' in the Abstract) are not mutually consistent under a single metric. The FRC in Fig. 2d is the key evidence for the k-factor of 0.65, but the text does not state how two independent reconstructions were obtained for the FRC calculation. If the FRC is computed from reconstructions that share the same weighted projection or from the same dataset split in a way that is not independent, the 1/2-bit threshold can be biased by correlated suppression of noise. Additionally, the Discussion concedes that the photon-free calibration does not fully account for signal-dependent shot noise, so the confidence map may down-weight genuine weak high-frequency signal while suppressing detector background. This would inflate the apparent SNR and resolution. The method is plausible, but the evidence currently does not cleanly separate signal preservation from background suppression.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a detector-informed confidence weighting for ptychographic reconstruction. A pixelwise confidence map w(q,t_exp), calibrated from photon-free detector noise measurements, is incorporated into the mPIE amplitude projection, replacing the standard hard amplitude constraint with a weighted residual update: Psi'_j(q) = Psi_j(q) + w(q,t_exp) r_j(q) Psi_j(q)/|Psi_j(q)|. The manuscript reports experiments in transmission (USAF target), reflection (semiconductor sample), and weak biological phase imaging, comparing against dark-frame subtraction, minimization-based background removal, and in-loop Poisson-Gaussian maximum likelihood. Claims include an approximately twofold SNR enhancement, reconstruction approaching the Rayleigh limit with a k-factor of about 0.65, reduced anisotropic artifacts in reflection geometry, improved phase contrast in biological tissue, and faster convergence.","tokens_in":13220,"tokens_out":5482,"duration_ms":51871,"significance":"The proposed modification is simple, physically motivated, and potentially useful: it replaces the implicit assumption of a uniform detector with a calibrated pixelwise reliability map and applies the weighting inside the iterative loop rather than as preprocessing. The manuscript includes three distinct experimental modalities and comparisons with several baselines, which is a strength. However, the load-bearing quantitative evidence is not yet rigorous. The headline SNR metric directly rewards background suppression, which the method implements by construction; the Fourier ring correlation protocol is underspecified; and none of the quantitative comparisons carry uncertainty estimates or significance tests. The central idea is plausible and deserves further scrutiny, but the evidence currently does not cleanly separate genuine signal preservation from noise suppression.","major_comments":[{"comment":"The SNR metric is defined as the ratio of the mean-square intensity of the full diffraction frame to that of a selected signal-free background region. Because the confidence-weighted projection explicitly down-weights residuals in low-confidence pixels, any reduction of background in the estimated diffraction frame directly improves this metric; the 'twofold SNR enhancement' claim therefore does not provide independent evidence of improved reconstruction fidelity. In addition, the reported values (11.56 dB raw, 33.27 dB DFS, 65.11 dB proposed) correspond to linear improvements of many orders of magnitude under the standard 10 log10 definition, not 'approximately twofold'; the abstract's wording is inconsistent with the reported dB numbers and must be reconciled.","section":"Section 2.2 (SNR definition) and Abstract"},{"comment":"The FRC curve and the resulting 745 nm resolution / k-factor of 0.65 require two independent reconstructions. The manuscript does not state how the two reconstructions were obtained: for example, whether they came from independent halves of the scan, different random initializations, or separate noise realizations, and whether the same confidence weighting was used in both. Without this information, the 1/2-bit crossing may reflect correlated suppression of background rather than a genuine resolution limit. Please specify the FRC protocol and, if needed, recompute the FRC from genuinely independent reconstructions.","section":"Section 2.2 and Fig. 2d (FRC protocol)"},{"comment":"The update equation Psi'_j(q) = Psi_j(q) + w(q,t_exp) r_j(q) Psi_j(q)/|Psi_j(q)| applies the same weight to the entire amplitude residual. The interpretation then decomposes r_j into r_j^diff and r_j^det and assumes both components are down-weighted identically, but this decomposition is not observable and no evidence is given that low-confidence pixels do not contain genuine high-frequency diffraction signal. The Discussion's admission that the photon-free calibration does not fully account for signal-dependent shot noise makes it plausible that weak real signal is attenuated together with detector background, which would directly affect the resolution and phase-contrast claims.","section":"Section 4.2 (residual decomposition and weighting)"},{"comment":"In the reflection experiment, the exposure times exceed the calibrated range, so the confidence map is partly extrapolated. The claimed reduction of anisotropic artifacts therefore rests on an unvalidated extrapolation of the detector-response model. Please either perform the calibration at the experimental exposure settings or provide a sensitivity analysis showing that the resolution and contrast results are robust to plausible miscalibration of w(q,t_exp).","section":"Section 2.3 and Section 3 (calibration extrapolation)"},{"comment":"All quantitative comparisons, including the Michelson modulation in Fig. 2g, the CNR and structural contrast in Fig. 3h, the PSNR/SSIM values in Fig. 4f/h, and the phase CNR in Fig. 6e, are based on single reconstructions without uncertainty estimates or significance tests. Given that some reported improvements are modest, the reader cannot determine whether the differences are within run-to-run variability. At minimum, the headline claims (SNR enhancement, k-factor, phase CNR) should be accompanied by repeated acquisitions or bootstrap uncertainty estimates.","section":"Sections 2.2-2.5 (statistical support)"}],"minor_comments":[{"comment":"There is a grammatical error: 'the method construct a spatially resolved confidence map' should be 'the method constructs a spatially resolved confidence map'.","section":"Abstract"},{"comment":"'fourier ring correlation' should be capitalized as 'Fourier ring correlation'.","section":"Fig. 2d and text"},{"comment":"Several references appear mismatched with the cited claims: for example, Jagatap and Hegde (2019) is listed with a title about THz metamaterials and an IEEE Transactions on Information Theory venue, which does not match the ptychography context in which it is cited; Yang et al. (2022) is listed with a title about multi-scale exposure fusion, which likewise does not match the EUV imaging context. Please verify all citations against the bibliography.","section":"References"},{"comment":"The statement that 'the other three modes exhibit at least a two-fold improvement' in PSNR is ambiguous: if PSNR is quoted in dB, a twofold improvement would be an increase of about 3 dB, not the large increase implied by the text. Please state the units and the conversion used.","section":"Section 2.4"},{"comment":"The data availability statement says the datasets are not publicly available, and the code availability statement contains a grammatical error ('Codes used to post-process the diffraction data with in this paper'). Consider clarifying the code availability and whether the calibration data and reconstruction code can be shared to support reproducibility.","section":"Data Availability / Code Availability"}],"recommendation":"major_revision","confidential_remarks":"The paper's citation list contains several apparent mismatches between the cited work and the claims they support, which should be checked before any further consideration. The restrictive data and code availability statements may also be a concern for a computational imaging methods paper, since the central quantitative claims currently rest on underspecified metrics and single experimental runs. The core idea is simple and plausible, and the experimental coverage across transmission, reflection, and biological samples is commendable, but the quantitative evidence needs substantial strengthening."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the useful part: the paper takes a known trick—weighted amplitude projections in ptychography—and gives the weights a physical source: a pixelwise, exposure-dependent detector confidence map from photon-free calibration. That specific integration is new, and the experiments are admirably broad: transmission USAF, reflection semiconductor, and weak biological phase imaging. The qualitative improvements are consistent across all three, and the math is straightforward and correct. The authors also deserve credit for openly noting the two main limitations in the Discussion, namely shot noise not being fully captured and the reflection experiment extrapolating beyond the calibrated exposure range.\n\nThe soft spots are quantitative, not conceptual. The headline 'twofold SNR enhancement' does not survive contact with their own numbers: 33.27 dB to 65.11 dB is roughly a factor-of-40 change in power, not 2. The SNR metric itself is nonstandard—it measures background suppression in the diffraction frame, not reconstruction quality, so it rewards exactly what the weighting is designed to do. The FRC analysis is the real problem: the text never says how the two independent reconstructions for the FRC were obtained. If they share the same weighted projection or the same noise realization, the 1/2-bit crossing at 745 nm could reflect correlated noise suppression rather than genuine signal. No error bars or significance tests are given, and neither code nor data is public.\n\nNone of this kills the central idea. A weighted projection from calibrated detector reliability is plausible and the qualitative results support it. But the paper overstates the quantitative evidence, and the FRC protocol needs to be spelled out and ideally validated with independent scans. The right fix is a major revision, not rejection.\n\nWho is this for? Anyone doing ptychography with imperfect detectors, especially in low-flux regimes. I would not cite it yet, but I would bring it to a reading group and would definitely send it to peer review.","headline":"A practically useful but quantitatively overclaimed weighted-projection extension for ptychography; the idea is sound, the evidence is uneven.","tokens_in":13601,"tokens_out":1899,"would_cite":false,"duration_ms":17259,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Embedding a detector-calibrated confidence map into the ptychographic amplitude projection allows reconstructions to approach the Rayleigh diffraction limit (measured $k$-factor about 0.65) and roughly doubles SNR on imperfect detectors.","keywords":["ptychography","phase retrieval","computational imaging","detector calibration","confidence map","noise suppression","Rayleigh resolution limit","signal-to-noise ratio"],"falsifier":"Acquire a calibrated resolution target at several exposure times inside and outside the confidence map's calibration range, reconstruct with the confidence-weighted update and with standard mPIE under identical conditions, and compare Fourier ring correlation and line-pair contrast: if the weighted update's advantage over uniform weighting disappears or reverses outside the calibrated range, or if injecting known shot noise causes the weighted update to lose high-frequency features that the unweighted update retains, the claim that the map separates detector noise from signal is falsified.","tokens_in":12588,"feed_emoji":"🔬","tokens_out":15126,"duration_ms":110564,"temperature":0.7,"pith_summary":"Ptychographic imaging reconstructs a complex optical field from overlapping diffraction patterns, but its resolution is capped by detectors that do not record every pixel with equal reliability. This paper claims that the detector's own calibration can be used inside the reconstruction loop: a per-pixel confidence map weights the amplitude constraint, so unreliable detector residuals contribute less to the iterative correction while trustworthy diffraction information is preserved. In transmission, reflection, and weak biological phase imaging, the weighted update suppresses detector background, raises diffraction signal-to-noise ratio by roughly a factor of two, and resolves features near the Rayleigh criterion with a measured $k$-factor of about 0.65. The takeaway is that non-ideal detection hardware need not be treated as a bottleneck to be removed by preprocessing; calibrated detector reliability can serve as a physical constraint that improves the inverse problem itself.","feed_headline":"Calibrated detector map pushes ptychography near Rayleigh limit","feed_subtitle":"Weighting detector pixels by calibrated reliability doubles SNR and recovers weak high-frequency signal.","key_machinery":"The central object is the detector confidence map $w(\\mathbf{q}, t_{\\mathrm{exp}})$: a spatially resolved, exposure-dependent map of pixelwise detector reliability derived from a photon-free calibration of the camera noise. It is embedded in the amplitude-projection step of the mPIE (momentum-accelerated ptychographic iterative engine) algorithm by modifying the detector-plane wavefield as above, or equivalently by writing the amplitude-consistency loss as $\\mathcal{L}_{\\mathrm{amp}} = \\frac{1}{2}\\sum_{j,\\mathbf{q}} w(\\mathbf{q},t_{\\mathrm{exp}})(|\\Psi_j(\\mathbf{q})|-\\sqrt{I_j^{\\mathrm{mea}}(\\mathbf{q})})^2$. The map plays the role of a pixelwise reliability coefficient (analogous to $1/\\sigma_i^2$ in weighted least squares), but it is not prescribed from a global noise law; it is constructed from calibrated sensor-response characteristics and then applied to the residual within the iterative update, which is what allows low-confidence detector regions to contribute less to the correction of the object and probe.","core_discovery":"The central claim is that replacing the uniform amplitude projection in a ptychographic engine with a confidence-weighted projection—$\\Psi_j'(\\mathbf{q}) = [(1-w(\\mathbf{q},t_{\\mathrm{exp}}))|\\Psi_j(\\mathbf{q})| + w(\\mathbf{q},t_{\\mathrm{exp}})\\sqrt{I_j^{\\mathrm{mea}}(\\mathbf{q})}]\\,\\Psi_j(\\mathbf{q})/(|\\Psi_j(\\mathbf{q})|+\\varepsilon)$—improves reconstruction quality when the detector is imperfect. The weight $w(\\mathbf{q},t_{\\mathrm{exp}})$ is a spatially resolved, exposure-dependent map built from a photon-free, pixelwise camera-noise calibration; it enters the update as $w r_j$, down-weighting the amplitude residual $r_j = \\sqrt{I_j^{\\mathrm{mea}}} - |\\Psi_j|$ in low-confidence pixels. With this change the paper reports a background-based SNR improvement from 11.56 dB (raw) and 33.27 dB (dark-frame subtraction) to 65.11 dB in transmission USAF imaging, a Fourier ring correlation resolution of 745 nm corresponding to a $k$-factor of about 0.65 (Rayleigh value 0.61), reduced anisotropic artifacts in reflection geometry, higher tissue-background phase contrast on a weakly scattering biological sample, and near-converged reconstructions after about 300 rather than 500 iterations. The claim is distinct from both out-of-loop data cleaning and prescribed noise models: the detector is still imperfect, but its measured reliability becomes part of the physical constraint.","pith_inferences":["Beyond the paper: the same weighted-residual construction should transfer to any intensity-based coherent imaging method—Fourier ptychography, coherent diffraction imaging, inline holography—because the only change is in how measured amplitudes enter the data-fidelity term, provided a per-pixel reliability map is available.","Beyond the paper: a natural testable extension is to make the confidence weights intensity-dependent, adding a shot-noise term to the photon-free calibration; this would directly address the paper's stated caveat and should improve low-flux performance.","Beyond the paper: if the reported $k$-factor of about 0.65 reproduces across sensor architectures, detector-informed weighting could become a standard component of ptychographic pipelines, letting laboratories improve resolution through calibration software rather than hardware upgrades."],"forward_implications":["Ptychographic systems with imperfect detectors can reach resolutions close to the diffraction limit without pre-cleaning the diffraction data, as long as a detector calibration is available.","A single pre-calibrated confidence map remains effective over months; the paper reports stable power spectra and intensity statistics when the same calibration is reused after a three-month interval.","The method suppresses detector-induced background while preserving mid- and high-frequency structural signal, whereas the Poisson–Gaussian maximum-likelihood baseline tends to smooth weak features in the comparisons.","Because the weighting acts inside the amplitude projection, convergence is accelerated: near-converged reconstructions appear at about 300 iterations rather than 500.","The improvement is not limited to one geometry: transmission, reflection, and weakly scattering biological phase imaging all benefit."],"supporting_citations":[{"why":"Supplies the photon-free camera-noise calibration from which the pixelwise confidence map $w(\\mathbf{q},t_{\\mathrm{exp}})$ is constructed.","marker":"(Xie, Zhou, et al. 2025)"},{"why":"Provides the mPIE iterative engine whose amplitude projection the confidence-weighted update modifies.","marker":"(Andrew Maiden et al. 2017)"},{"why":"Defines the maximum-likelihood refinement and Poisson–Gaussian noise-model approach that the proposed method contrasts with.","marker":"(Thibault and Guizar-Sicairos 2012)"},{"why":"Supplies the in-loop Poisson–Gaussian maximum-likelihood (PG-ML) baseline used in the experimental comparisons.","marker":"(Seifert et al. 2023)"},{"why":"Provides the background-noise removal preprocessing baseline for dark-frame and background-subtracted diffraction data.","marker":"(Wang et al. 2017)"},{"why":"Supplies the adaptive-filtering baseline for suppressing background noise before reconstruction.","marker":"(Qiao et al. 2023)"},{"why":"Gives the prior observation that suppressing background fluctuations is required for stable high-resolution ptychographic reconstruction.","marker":"(Xie, Lin, et al. 2025)"}],"fun_headline_variants":["Ptychography calibrates detector trust to double SNR","Confidence-weighted ptychography nears the Rayleigh limit","Detector reliability as constraint doubles ptychographic SNR","Pixelwise sensor trust map lifts ptychography to Rayleigh edge","In-loop detector calibration doubles SNR in ptychography"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a light-free detector calibration—which the paper itself says does not fully capture signal-dependent shot noise, and which is partly extrapolated beyond the calibrated exposure range in the reflection experiment—produces a dependable map of which pixels are reliable; if that map misranks pixels, the weights will suppress genuine weak signal rather than detector noise.","fun_headline_variants_meta":{"raw":{"variants":["Ptychography calibrates detector trust to double SNR","Confidence-weighted ptychography nears the Rayleigh limit","Detector reliability as constraint doubles ptychographic SNR","Pixelwise sensor trust map lifts ptychography to Rayleigh edge","In-loop detector calibration doubles SNR in ptychography"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000637,"raw_usage":{"total_tokens":2993,"prompt_tokens":1059,"completion_tokens":1934,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":675,"completion_tokens_details":{"reasoning_tokens":1865}},"tokens_in":675,"tokens_out":1934,"duration_ms":13388,"temperature":1.0,"reasoning_tokens":1865,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:04:15.368796+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Acquire a calibrated resolution target at several exposure times inside and outside the confidence map's calibration range, reconstruct with the confidence-weighted update and with standard mPIE under identical conditions, and compare Fourier ring correlation and line-pair contrast: if the weighted update's advantage over uniform weighting disappears or reverses outside the calibrated range, or if injecting known shot noise causes the weighted update to lose high-frequency features that the unweighted update retains, the claim that the map separates detector noise from signal is falsified.","supporting_citations":[],"review_version":2}