Pith. sign in

REVIEW 2 major objections 6 minor 29 references

Harnessing Machine Learning for Single-Shot Measurement of Free Electron Laser Pulse Power

T0 review · 2 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Machine parameters recorded during lasing can predict the lasing-off electron power profile well enough to beat batch-calibration averages.

desk verdict Solid in-domain prediction result, but the advertised single-shot diagnostic rests on an independence assumption the paper's own setup contradicts. read the letter →

arxiv 2411.09468 v2 pith:5F45NVPM submitted 2024-11-14 cs.LG physics.acc-ph

classification cs.LGphysics.acc-ph
keywords free-electronlasersmachinelearningmultilayerperceptronelectronbunchdiagnosticspulsereconstructiontransversedeflectingcavityvirtuallasing-offprediction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Free-electron lasers cannot measure, for a single electron bunch, both the electron power profile with lasing and the profile it would have without lasing; the missing lasing-off profile is needed to reconstruct the photon pulse. The paper claims that this missing profile can be predicted from 22 machine parameters that are available while lasing is on, using a multilayer perceptron trained on 2,826 lasing-off bunches. On a held-out test set, the prediction error was lower than the error of the current batch-calibration baseline and lower than using neighboring shots, with the difference significant at p<0.01 after correction for multiple comparisons. If correct, this is the enabling step for a virtual pulse reconstruction diagnostic that would give per-pulse photon power profiles without interrupting lasing for calibration.

What carries the argument

The central object is a trained multilayer perceptron regression model that maps 22 pre-undulator machine parameters (bunch charges, bunch-arrival-time monitor readings, beam position, and energy) to a 567-point electron temporal power profile. Supporting machinery is a GPU-accelerated image-processing pipeline that converts longitudinal phase-space images into power profiles, removes temporal jitter by aligning peak locations, and crops the signal with threshold-based segmentation. In the virtual pulse reconstruction workflow, the predicted lasing-off profile is subtracted from the measured lasing-on profile to obtain the photon pulse power.

What would settle it

Compare the joint distribution of the 22 machine parameters recorded during lasing-on operation with the distribution recorded during lasing-off at the same nominal accelerator settings; if the two distributions differ beyond shot-to-shot noise, the model's lasing-off prediction from lasing-on parameters is biased and the photon-pulse reconstruction built on it is wrong.

Watch

Extended reading notes

Core claim

The paper's central claim is that a multilayer perceptron with 22 input nodes, one hidden layer of 294 nodes, and 567 output nodes predicts the temporal power profile of an electron bunch in the lasing-off regime from measured machine parameters, and that this prediction beats the state of the art. The authors validate the claim by comparing the model's median squared error (0.007) on 282 test bunches with the error of the mean training profile (0.009) and with neighboring-shot measurements (0.02); a non-parametric signed-rank test with multiple-comparison correction gives p<0.01. They further show that the predicted profile can serve as the subtrahend in the lasing-on minus lasing-off subtraction, turning the measured lasing-on electron power into an estimate of the photon pulse power for each individual shot.

Load-bearing premise

The model is trained entirely on lasing-off data but will be asked to predict lasing-off profiles from parameters recorded while lasing is on; the assumption that the 22 pre-undulator machine parameters are unchanged by lasing is stated and not experimentally tested.

Editorial extensions

If this is right

  • Each lasing shot gets its own reconstructed photon power profile, instead of a batch-averaged difference between lasing-on and lasing-off runs.
  • Neighboring-shot measurements are a poor predictor of the lasing-off profile, so relying on them in future longitudinal phase-space models will not match this approach.
  • The model's roughly 16 microseconds per prediction is fast enough to use after an experiment, and potentially for live monitoring once integrated into beamline hardware.
  • The workflow supplies the missing component for the virtual pulse reconstruction diagnostic: measured lasing-on phase images combined with predicted lasing-off profiles complete the subtraction.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A distribution-drift check on live lasing-on parameters could test, before deployment, whether the unmeasured lasing-on regime invalidates the predictions; this is an extension the paper leaves to future work.
  • The parameter-to-profile mapping might transfer to other longitudinal bunch diagnostics and other FEL beamlines, but no evidence outside this dataset is presented.
  • Comparing the multilayer perceptron against simpler regressors on the same data would separate nonlinearity from architectural choices; the paper does not include such a comparison.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper trains a multi-layer perceptron (MLP) to predict the temporal power profile of the electron bunch in the lasing-off regime from 22 machine parameters measured at the FLASH2 beamline. The model is evaluated on a held-out test set drawn from the same lasing-off dataset, and its mean squared error is compared against two baselines: the mean of the training dataset and the profile of a neighboring shot. The authors report that the MLP predictions are statistically significantly better than both baselines (Wilcoxon signed-rank test with Bonferroni correction). The stated motivation is a Virtual Pulse Reconstruction Diagnostic (VPRD): during lasing-on operation, the model would predict the unmeasurable lasing-off electron power profile, which would then be subtracted from the measured lasing-on profile to reconstruct individual photon pulses.

Significance. If the transfer to the lasing-on regime were valid, this would be a valuable step toward single-shot FEL pulse characterization. The in-domain result is a useful proof of concept: the training procedure is careful (Optuna hyperparameter search, dropout, early stopping, statistical testing), and the authors provide code and data. However, the central claim—that the model can be used during lasing-on operation—rests on an independence assumption that is not tested and is internally contradicted by the paper's own description of the input features. If that assumption fails, the paper reduces to a demonstration of lasing-off profile prediction, which is still of interest but not 'single-shot measurement of FEL pulse power.' The contradiction and missing validation must be addressed before the application claim can be accepted.

major comments (2)
  1. [Section 4.1, Table 1, Appendix A.1] The central claim of the paper requires that the 22 input parameters are independent of whether lasing is on or off. Section 4.1 asserts this by stating that 'all parameters used for training are measured before the undulator magnet that induces lasing.' However, Table 1 includes 'BPM x, y electron beam position before TDS' as inputs, and Appendix A.1 states that the PolariX TDS is installed 'downstream of the FLASH2 undulator line.' A BPM located before that TDS is therefore after the undulator, not before it. The electron beam's position at such a BPM depends on the beam energy and energy spread, both of which are modified by FEL lasing, so the input distribution can shift between the training regime (lasing-off) and the deployment regime (lasing-on). This directly undermines the single-shot VPRD transfer: the predicted lasing-off profile for a lasing-on shot would be biased, and the subtraction in Figure 1 would inherit that bias. The manuscript should either (i) remove post-undulator inputs and retrain, (ii) empirically demonstrate (e.g., with archived lasing-on data) that the BPM distributions do not depend on lasing state, or (iii) restrict the claims to lasing-off prediction and describe the lasing-on application as future work. The in-domain result is not affected by this issue.
  2. [Section 3, Section 4.1] The evaluation is entirely in the lasing-off regime. The test set is randomly split from the same 2826-sample lasing-off dataset, and no lasing-on data are used either for validation or for checking the independence assumption. The paper's own Limitations section admits that 'this assumption needs to be tested in future experiments.' Because the abstract and introduction frame the contribution as a 'critical element' for single-shot measurement of FEL pulse power during lasing, the manuscript needs at least a feasibility argument—ideally a comparison of the input feature distributions between lasing-on and lasing-off shots, or a targeted physics argument for each of the 22 features—before the transfer claim can be accepted. Without such evidence, the central application claim is unsupported, even though the in-domain regression result itself is well supported.
minor comments (6)
  1. [Figure 2b] The caption states a dropout of 0.43, while Section 2.1 states 0.45; please reconcile these numbers.
  2. [Figure 2c] The reported prediction MSE '0.007(0.055 − 0.0101)' appears to contain a typo: the upper interquartile bound 0.055 is inconsistent with the median 0.007 and with the other reported values; likely the intended range is 0.0055–0.0101.
  3. [Equation (1)] The loss described as penalizing regression to the mean subtracts α times the squared deviations of predictions from the label mean, which encourages spread rather than penalizing it; please clarify the intended effect and the sign convention.
  4. [Section 4.1 vs. Table 1] The sentence claiming all training parameters are measured before the undulator should be reconciled with Table 1, which lists BPM positions before the TDS (a post-undulator device); at minimum, specify the exact location of each BPM relative to the undulator.
  5. [Section A.2.1, Figure 1] The de-jittering step aligns all training profiles to a common time grid; for deployment, the predicted lasing-off profile would need to be mapped back to the time axis of the measured lasing-on phase image. The paper should describe how this alignment is performed in the VPRD workflow.
  6. [Section 4.1] The phrase 'field programmable array' should be 'field-programmable gate array' (FPGA).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MLP's target labels are measured phase-space power profiles, not derived from the 22 input parameters, and the held-out test evaluation is self-contained.

full rationale

The paper's derivation chain is not circular. The training labels are temporal electron power profiles computed from measured longitudinal phase-space images (Appendix A.2), while the model inputs are 22 machine parameters listed in Table 1. These quantities are measured independently, and no target label is reconstructed from the inputs by definition. The MLP is a standard supervised regressor trained with MSE loss on a random split, and the reported prediction errors are computed on a held-out test set against measured profiles, not against values used in training. The mean and neighbor baselines are evaluation comparators, not components of the model, the loss, or the predictive formula, so comparing against them is not a circular reduction. The only self-references are to the authors' code and data repositories and to PolariX TDS hardware papers; these are not load-bearing for the claim that the MLP predicts lasing-off power profiles. The paper's own Section 4.1 limitation states that the intended lasing-on application assumes the input parameters are independent of lasing state because they are measured before the undulator; however, Appendix A.1 and Table 1 include BPM positions 'before TDS', which sit downstream of the FLASH2 undulator. This is a real domain-shift/correctness risk for the intended application, but it is not circularity: it does not make the in-domain prediction equal to its inputs. Thus no circular step is present.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated. The Virtual Pulse Reconstruction Diagnostic (VPRD) is a proposed software workflow built from existing measurements, not an invented physical quantity. The central claim rests on measured data, standard ML machinery, and one explicit untested domain assumption about lasing-state independence.

free parameters (6)
  • Hidden layer size = 294
    Chosen by Optuna hyperparameter search; affects model capacity and prediction quality.
  • Dropout fraction = 0.45
    Chosen by Optuna; Figure 2 caption later says 0.43, a minor inconsistency.
  • Initial learning rate = 0.005
    Chosen by Optuna; part of the Adam optimizer setup.
  • Jitter compensation Gaussian radius = 10 pixels
    Chosen for smoothing power profiles to find peak locations; affects alignment and label quality.
  • Crop padding = 10 pixels
    Chosen for bounding-box crop of the signal region; determines output vector width (567 nodes).
  • Alternative loss penalty alpha = 0 (not used)
    Explored in Equation 1; models with alpha > 0.05 performed worse, so standard MSE was used.
assumptions (4)
  • domain assumption Machine parameters are independent of whether lasing is on or off
    Stated in Section 4.1 Limitations; all parameters are measured before the undulator, but this is untested.
  • domain assumption The temporal power profile computed from the longitudinal phase space image is an accurate ground truth
    Appendix A.2: the power profile is the energy-weighted projection of the phase image, relying on TDS calibration (1.13 fs/mm and 21 keV/mm).
  • domain assumption The 2826-sample dataset is representative of the relevant operating conditions
    Data were collected at a single machine setting (12 nm, 200 pC, 875 MeV); generalizability to other settings is unknown.
  • standard math The statistical test (Wilcoxon signed-rank test) is appropriate for the paired MSE comparisons
    The test makes no normality assumption and is applied to paired per-sample errors, with Bonferroni correction for three comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Harnessing Machine Learning for Single-Shot Measurement of Free Electron Laser Pulse Power." pith.science (2026). https://pith.science/paper/5F45NVPM

@misc{pith2026241109468,
  author       = {Pith},
  title        = {Pith review of: Harnessing Machine Learning for Single-Shot Measurement of Free Electron Laser Pulse Power},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5F45NVPM}},
  note         = {Machine review of arXiv:2411.09468}
}
read the original abstract

Electron beam accelerators are essential in many scientific and technological fields. Their operation relies heavily on the stability and precision of the electron beam. Traditional diagnostic techniques encounter difficulties in addressing the complex and dynamic nature of electron beams. Particularly in the context of free-electron lasers (FELs), it is fundamentally impossible to measure the lasing-on and lasingoff electron power profiles for a single electron bunch. This is a crucial hurdle in the exact reconstruction of the photon pulse profile. To overcome this hurdle, we developed a machine learning model that predicts the temporal power profile of the electron bunch in the lasing-off regime using machine parameters that can be obtained when lasing is on. The model was statistically validated and showed superior predictions compared to the state-of-the-art batch calibrations. The work we present here is a critical element for a virtual pulse reconstruction diagnostic (VPRD) tool designed to reconstruct the power profile of individual photon pulses without requiring repeated measurements in the lasing-off regime. This promises to significantly enhance the diagnostic capabilities in FELs at large.

Figures

Figures reproduced from arXiv: 2411.09468 by the authors.

Figure 1
Figure 1. Process workflow. During beamline operation, we obtain measured [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. MLP model training performance. (a) Training and validation loss. The validation loss is [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Schematic layout of the FLASH facility at Deutsches Elektronen-Synchrotron (DESY), [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: De-jittering. (a) Phase image. (b) Temporal power profile created by weighing the phase [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 12 canonical work pages

  1. [1]

    Introduction to Machine Learning for Accelerator Physics, June 2020

    Daniel Ratner. Introduction to Machine Learning for Accelerator Physics, June 2020. URL http://arxiv.org/abs/2006.09913. arXiv:2006.09913

  2. [2]

    Reinforcement learning-trained optimisers and bayesian optimisation for online particle accelerator tuning

    Jan Kaiser, Chenran Xu, Annika Eichler, Andrea Santamaria Garcia, Oliver Stein, Erik Brün- dermann, Willi Kuropka, Hannes Dinter, Frank Mayet, Thomas Vinatier, Florian Burkart, and Holger Schlarb. Reinforcement learning-trained optimisers and bayesian optimisation for online particle accelerator tuning. Scientific Reports, 14(1):15733, Jul 2024. ISSN 2045...

  3. [3]

    Bridging the gap between machine learning and particle accelerator physics with high-speed, differentiable simu- lations

    Jan Kaiser, Chenran Xu, Annika Eichler, and Andrea Santamaria Garcia. Bridging the gap between machine learning and particle accelerator physics with high-speed, differentiable simu- lations. Phys. Rev. Accel. Beams, 27:054601, May 2024. doi: 10.1103/PhysRevAccelBeams.27. 054601. URL https://link.aps.org/doi/10.1103/PhysRevAccelBeams.27.054601

  4. [4]

    Physics-informed neural network method for space charge effect in particle accelerators

    Kazuhiro Fujita. Physics-informed neural network method for space charge effect in particle accelerators. IEEE Access, 9:164017–164025, 2021. doi: 10.1109/ACCESS.2021.3132942. URL https://doi.org/10.1109/ACCESS.2021.3132942

  5. [5]

    Jhalani, Jacek Krzywinski, Timothy J

    Florian Christie, Alberto Andrea Lutman, Yuantao Ding, Zhirong Huang, Vatsal A. Jhalani, Jacek Krzywinski, Timothy J. Maxwell, Daniel Ratner, Juliane Rönsch-Schulenburg, and Mathias V ogt. Temporal x-ray reconstruction using temporal and spectral measurements at lcls. Scientific Reports , 10(1):9799, 2020. doi: 10.1038/s41598-020-66220-5. URL https://doi....

  6. [6]

    Yuantao Ding, Carsten Behrens, Paul Emma, Jérôme Frisch, Zhirong Huang, H. Loos, P. Krejcik, and M-H. Wang. Femtosecond x-ray pulse temporal characterization in free-electron lasers using a transverse deflector. Physical Review Special Topics - Accelerators and Beams, 14(12): 120701, December 2011. doi: 10.1103/PhysRevSTAB.14.120701. URL https://link.aps....

  7. [7]

    Behrens, F.-J

    C. Behrens, F.-J. Decker, Yuantao Ding, V . A. Dolgashev, J. Frisch, Zhirong Huang, P. Krejcik, H. Loos, A. Lutman, T. J. Maxwell, J. Turner, J. Wang, M.-H. Wang, J. Welch, and J. Wu. Few- femtosecond time-resolved measurements of x-ray free-electron lasers.Nature Communications, 5(1):3762, Apr 2014. ISSN 2041-1723. doi: 10.1038/ncomms4762. URL https://do...

  8. [8]

    R. Akre, L. Bentson, P. Emma, and P. Krejcik. A transverse rf deflecting structure for bunch length and phase space diagnostics. In PACS2001. Proceedings of the 2001 Particle Accelerator Conference (Cat. No.01CH37268), volume 3, pages 2353–2355 vol.3, June 2001. doi: 10.1109/ PAC.2001.987379. URL https://ieeexplore.ieee.org/document/987379

Show all 29 references
  1. [9]

    Hanuka, C

    A. Hanuka, C. Emma, T. Maxwell, A. S. Fisher, B. Jacobson, M. J. Hogan, and Z. Huang. Accurate and confident prediction of electron beam longitudinal properties using spectral virtual diagnostics. Scientific Reports, 11(1):2945, Feb 2021. ISSN 2045-2322. doi: 10.1038/ s41598-0...

  2. [10]

    Optuna: A Next-generation Hyperparameter Optimization Framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A Next-generation Hyperparameter Optimization Framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’19, pages 2623–2631, New Y...

  3. [11]

    PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation

    Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael V oznesen- sky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, Geeta Chauhan, Anjali Chourdia, Will Constable, Alban Desmaison, Zachary DeVito, Elias Ellison, Will Feng, Jiong Gong, Michael Gsc...

  4. [12]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A Method for Stochastic Optimization, January

  5. [13]

    PyTorch Lightning, March 2019

    William Falcon and The PyTorch Lightning team. PyTorch Lightning, March 2019. URL https://github.com/Lightning-AI/lightning

  6. [14]

    Virtual Pulse Reconstruction Diagnostic, November 2024

    Till Korten, Peter Steinbach, and Najmeh Sadat Mirian. Virtual Pulse Reconstruction Diagnostic, November 2024. URL https://github.com/thawn/VPRD

  7. [15]

    harnessing machine learning for single-shot measurement of free electron laser pulse power

    Till Korten, Peter Steinbach, and Najmeh Sadat Mirian. Training data for "harnessing machine learning for single-shot measurement of free electron laser pulse power", November 2024. URL https://doi.org/10.14278/rodare.3253

  8. [16]

    J. Zhu, N. M. Lockmann, M. K. Czwalinna, and H. Schlarb. Mixed Diagnostics for Longitudinal Properties of Electron Bunches in a Free-Electron Laser. Frontiers in Physics, 10, July 2022. ISSN 2296-424X. doi: 10.3389/fphy.2022.903559. URL https://www.frontiersin.org/ articles/10...

  9. [17]

    Individual Comparisons by Ranking Methods

    Frank Wilcoxon. Individual Comparisons by Ranking Methods. Biometrics Bulletin, 1(6): 80–83, 1945. ISSN 0099-4987. doi: 10.2307/3001968. URL https://www.jstor.org/ stable/3001968. Publisher: [International Biometric Society, Wiley]

  10. [18]

    Teoria statistica delle classi e calcolo delle probabilita

    Carlo Bonferroni. Teoria statistica delle classi e calcolo delle probabilita. Pubblicazioni del R istituto superiore di scienze economiche e commericiali di firenze , 8:3–62, 1936. URL https://cir.nii.ac.jp/crid/1570009749360424576

  11. [19]

    Schneider, and Wilfried Wurth

    Jörg Rossbach, Jochen R. Schneider, and Wilfried Wurth. 10 years of pioneering X-ray science at the Free-Electron Laser FLASH at DESY. Physics Reports, 808:1–74, May 2019. ISSN 0370-1573. doi: 10.1016/j.physrep.2019.02.002. URL https://www.sciencedirect.com/ science/article/pi...

  12. [20]

    FLASH2020+ - upgrade of FLASH: concep- tual design report

    Martin Beye and Stephan Klumpp, editors. FLASH2020+ - upgrade of FLASH: concep- tual design report . Verlag Deutsches Elektronen-Synchrotron, Hamburg, 2020. ISBN 9783945931301. doi: 10.3204/PUBDB-2020-00465. URL https://bib-pubdb1.desy. de/record/434950

  13. [21]

    A PolariX TDS for the FLASH2 Beamline

    Florian Christie, Juliane Rönsch-Schulenburg, and Mathias V ogt. A PolariX TDS for the FLASH2 Beamline. In 39th International Free Electron Laser Conference, pages 328–331. JACOW Publishing, Geneva, Switzerland, November 2019. ISBN 978-3-95450-210-3. doi: 10.18429/JACoW-FEL201...

  14. [22]

    X-Band TDS Project

    Barbara Marchetti, Ralph Aßmann, Bolko Beutner, Markus Bopp, Julien Branlard, Hans- Heinrich Braun, Nuria Catalán Lasheras, Florian Christie, Paolo Craievich, Richard D’Arcy, Winfried Decking, Ulrich Dorda, Alexej Grudiev, Joerg Herrmann, Matthias Hoffmann, Markus Hüning, Olaf...

  15. [23]

    Status of the Polarix-TDS Project

    Paolo Craievich, Ralph Aßmann, Markus Bopp, Hans-Heinrich Braun, Nuria Catalán Lasheras, Florian Christie, Richard D’Arcy, Ulrich Dorda, Manon Foese, Romain Ganter, Pau González Caminal, Alexej Grudiev, Matthias Hoffmann, Markus Hüning, Rolf Jonas, Thomas Kleeb, Olaf Krebs, Sv...

  16. [24]

    Craievich, M

    P. Craievich, M. Bopp, H.-H. Braun, A. Citterio, R. Fortunati, R. Ganter, T. Kleeb, F. Marcellini, M. Pedrozzi, E. Prat, S. Reiche, K. Rolli, R. Sieber, A. Grudiev, W. L. Millar, N. Catalan- Lasheras, G. McMonagle, S. Pitman, V . del Pozo Romano, K. T. Szypula, W. Wuensch, B. ...

  17. [25]

    Lambert, Juan Nunez-Iglesias, Daniel P

    Robert Haase, Akanksha Jain, Stéphane Rigaud, Daniela V orkel, Pradeep Rajasekhar, Theresa Suckert, Talley J. Lambert, Juan Nunez-Iglesias, Daniel P. Poole, Pavel Tomancak, and Eu- gene W. Myers. Interactive design of GPU-accelerated Image Data Flow Graphs and cross- platform ...

  18. [26]

    clEsperanto/pyclesperanto, September 2024

    Stéphane Rigaud, Robert Haase, Cherif Latechre, Johannes Soltwedel, Mavin Albert, Pradeep Rajasekhar, and Graham Ross. clEsperanto/pyclesperanto, September 2024. URL https: //github.com/clEsperanto/pyclesperanto. original-date: 2022-03-31T11:00:44Z

  19. [27]

    A Threshold Selection Method from Gray-Level Histograms

    Nobuyuki Otsu. A Threshold Selection Method from Gray-Level Histograms. IEEE Trans- actions on Systems, Man, and Cybernetics , 9(1):62–66, January 1979. ISSN 2168-2909. doi: 10.1109/TSMC.1979.4310076. URL https://dspace.tul.cz/server/api/core/ bitstreams/36abcc1c-cd72-4569-90e...

  20. [2017]

    arXiv:1412.6980 [cs]

    URL http://arxiv.org/abs/1412.6980. arXiv:1412.6980 [cs]

  21. [3811]

    ISBN 978-3-95450-184-7

    JACOW Publishing, Geneva, Switzerland, June 2018. ISBN 978-3-95450-184-7. doi: 10.18429/JACoW-IPAC2018-THPAL068. URL https://accelconf.web.cern.ch/ ipac2018/doi/JACoW-IPAC2018-THPAL068.html

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.