Pith. sign in

REVIEW 5 major objections 5 minor 64 references

Neural Object Detection for 4D STEM: High-Throughput Sub-Pixel Electron Diffraction Pattern Recognition

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A neural object detector maps lattice strain from 4D-STEM diffraction data with sub-pixel accuracy, at over 100 frames per second.

desk verdict A useful, honest engineering paper on applying YOLOv8 to 4D-STEM Bragg disc localization, with a real speed claim and a precision claim that is not yet an accuracy claim; the paper is worth refereeing despite the missing validation. read the letter →

arxiv 2506.04477 v2 pith:WNLMCESV submitted 2025-06-04 cond-mat.mtrl-sci

Arda Genc , Ravit Silverstein This is my paper
classification cond-mat.mtrl-sci
keywords 4D-STEMobjectdetectionYOLOv8Braggdisclocalizationstrainmappingsub-pixelprecisionnanobeamelectrondiffractionphaseidentification
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

This paper tries to establish that neural object detection, a tool from everyday computer vision, can replace the sequential peak-fitting pipelines normally used to analyze four-dimensional scanning transmission electron microscopy (4D-STEM) diffraction data. The proposed NDPD workflow, built on the YOLOv8n detector, predicts bounding boxes around Bragg discs and reads each disc center from the box coordinates with sub-pixel accuracy through a distribution focal loss soft-argmax operation. On a Si/SiGe multilayer standard, the authors report strain precision of $5\times10^{-4}$, measured as the standard deviation of strain in the nominally strain-free Si substrate, and throughput above 100 frames per second on a desktop GPU. The same network also separates coexisting phases in a multiphase Ti-80Nb alloy by identifying the lattice spacings of each phase. If correct, this means a single learned model can turn raw 4D-STEM datasets into strain and phase maps in real time, without per-dataset parameter tuning.

What carries the argument

The machinery is a YOLOv8 Nano (YOLOv8n) single-stage anchor-free detector: a fully convolutional backbone-neck-head network that predicts, for each grid cell, offsets to nearby object centers and bounding-box width and height. Sub-pixel localization comes from distribution focal loss (DFL), where each of the four box sides is decoded as the expected value of a softmax distribution over discrete bins, and the center coordinate is the mean of the left/right and top/bottom box sides at the appropriate stride. An autocorrelation step based on the Wiener-Khinchin theorem, the inverse Fourier transform of the squared Fourier magnitude of the diffraction pattern, is used to make localization less sensitive to thickness and tilt effects. Asynchronous non-blocking function calling and automatic mixed precision carry the throughput.

What would settle it

Acquire or simulate 4D-STEM datasets with known ground-truth disc positions over a range of camera lengths, convergence angles, foil thicknesses, and noise levels, and compare NDPD's predicted centers to those positions; the central claim fails if the mean absolute center error exceeds a small fraction of a pixel, for example 0.1 pixels, or if the strain standard deviation in a truly strain-free region rises above $5\times10^{-4}$ under any of these conditions.

Watch

Extended reading notes

Core claim

The central discovery is that a single-stage anchor-free object detector, fine-tuned on 42 manually annotated diffraction patterns (augmented to 90), learns to locate Bragg disc centers well enough to resolve interatomic displacement fields at a precision of $5\times10^{-4}$ strain. The claim is that center localization is not merely pixel-level but sub-pixel, because distribution focal loss represents each bounding-box side as a discrete probability distribution over bins and the final coordinate is the expected value of that distribution, a soft-argmax. The authors further show that the same model detects double diffraction discs from overlapping crystal structures in a lamellar Ti-80Nb alloy and maps the $\alpha$-hcp lamellae, $\beta$-bcc matrix, coherent bcc precipitates, and the oxide phase Ti$_3$O. They attribute the practical speed to an asynchronous, non-blocking inference workflow combined with automatic mixed precision, reaching 110 fps on 3,880 patterns and 121 fps on 16,384 patterns.

Load-bearing premise

The load-bearing premise is that 42 manually annotated diffraction patterns from a small set of materials, augmented to 90 training images, capture enough of the real variation in disc size, sample thickness, camera length, and signal-to-noise ratio that the network's center predictions stay unbiased on unseen datasets; if the labels are biased or the set is unrepresentative, the reported strain precision and phase maps inherit that bias.

Editorial extensions

If this is right

  • A single end-to-end neural model can replace multi-step peak fitting for 4D-STEM, removing manual parameter tuning from strain mapping.
  • Strain precision reaches $5\times10^{-4}$ on a Si/SiGe standard, putting learned detection in the range of established nanobeam diffraction analysis pipelines.
  • Throughput above 100 fps on a desktop GPU makes real-time feedback during acquisition feasible.
  • The same network can identify overlapping phases in multiphase alloys by detecting double diffraction discs from superimposed crystal structures.
  • Phase maps can be reconstructed from interatomic lattice distances rather than diffraction intensities, analogous to composition-based XEDS phase mapping.

Reading between the lines

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

  • A natural extension the authors do not pursue is using the DFL probability distributions as uncertainty estimates, so each strain value could carry a localization confidence.
  • Because the detector is scale-aware through its feature pyramid and path aggregation network, the same architecture may generalize to larger or overlapping discs if the training set is expanded, which is testable by adding simulated CBED patterns.
  • The demonstrated 110-121 fps throughput suggests that, with a faster detector or smaller input, the workflow bottleneck may shift from inference to I/O, making data streaming and storage the next limit rather than the neural network.
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

5 major / 5 minor

Summary. The paper introduces NDPD, a YOLOv8n-based object detection framework for 4D-STEM diffraction pattern analysis. The central claims are that a distribution-focal-loss soft-argmax (Eq. 1) provides sub-pixel accurate Bragg-disc center localization, that this enables strain precision of 5x10^-4 (standard deviation in a nominally strain-free Si substrate of a Si/SiGe standard), and that an asynchronous, mixed-precision inference workflow exceeds 100 fps on a desktop GPU. The authors demonstrate phase and strain mapping on a Si/SiGe multilayer sample and a multiphase Ti-80Nb alloy, and they report throughput of 110 fps on 3,880 patterns and 121 fps on 16,384 patterns.

Significance. If the sub-pixel accuracy claim were properly validated, the paper would be a useful contribution: replacing sequential peak-fitting pipelines with a single end-to-end detector could enable real-time 4D-STEM analysis. The paper has concrete strengths: it uses a traceable MAG*I*CAL standard, compares the measured mean strain with an independently calculated theoretical value, reports specific throughput benchmarks, and states that code and a demonstration video are available. However, the central accuracy claim is not currently supported by the evidence presented, and no detection metrics or baseline comparisons are reported. The significance is therefore conditional on additional validation experiments.

major comments (5)
  1. [Sections 3.2 and 4.2] The central claim of sub-pixel accurate center localization is not tested. Training targets are manually annotated bounding boxes (Section 3.2), which do not constitute sub-pixel ground truth; manual annotation carries at least integer-pixel uncertainty. The reported 5x10^-4 strain standard deviation in the Si substrate is a precision (repeatability) metric, not an accuracy metric: a constant or slowly varying per-disc center bias would not increase the standard deviation but would shift the derived strain. The manuscript needs a validation against simulated diffraction patterns with known sub-pixel centers, or an independent comparison with a fitting-based method on the same experimental data.
  2. [Section 4.2] The strain precision is computed on a hand-selected substrate region described only as 'away from the thinnest part of the TEM sample.' The paper does not specify the selection criteria, the number of patterns or discs used, or whether the region was independently confirmed to be strain-free. Without this information, and without an error propagation analysis from center localization to strain, the 5x10^-4 figure cannot be evaluated. Please report the selection procedure, sample sizes, and a baseline comparison (e.g., Gaussian fitting or cross-correlation) on the same dataset.
  3. [Section 3.1, Eq. (4)] The autocorrelation-based enhancement via the Wiener-Khinchin theorem is asserted but never connected to the network outputs. No description is given of how R(Δk) is incorporated into the detection head or the training pipeline, no ablation is provided with and without this enhancement, and no evidence shows that it affects the reported strain or throughput results. Please clarify the architectural integration and provide an ablation study.
  4. [Section 4.3] The phase and strain maps for the Ti-80Nb alloy are not validated against any independent measurement, such as XEDS phase mapping, EBSD, or conventional strain mapping. The qualitative agreement with expected phases is suggestive but does not quantify accuracy. Please add a quantitative validation or explicitly label these maps as demonstrative without quantitative claims.
  5. [Section 3.2] The training set is characterized only as 42 manually annotated diffraction patterns from 'a diverse range of materials,' augmented to 90 images. No information is given about the distribution of disc sizes, camera lengths, specimen thickness, or signal-to-noise ratios, and no detection metrics (mAP, precision/recall, center-error distribution) are reported for the validation set. These omissions make the generalization claim difficult to assess; please report validation metrics and a description of the training distribution.
minor comments (5)
  1. [Eq. (5)] The theoretical strain expression uses aGe, aSi, xSi, and ν, but the text does not state whether Vegard's law is assumed or how xSi relates to the measured 87 at.% Si composition; please clarify the notation and the assumption.
  2. [Eq. (1)] The summation index runs from k=0 to reg max, which is consistent with the statement that reg max=16 gives 17 bins, but p_n^k should be explicitly defined as the softmax probability over the DFL bins.
  3. [Eq. (4) and surrounding text] The variable Δk is called a displacement vector in 'quefrency space,' but the autocorrelation of the diffraction intensity is a real-space (or Patterson-like) function; please rephrase to avoid confusion.
  4. [Figure 4 caption] The caption refers to a '13x13 pixel grid' while the text discusses stride levels of 8, 16, and 32; please clarify whether the figure is illustrative or corresponds to an actual detection grid.
  5. [Section 6] The data availability statement names a repository but gives no URL or DOI; please provide a working link or accession identifier.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the strain measurement is compared with an independent composition-based estimate, and no fitted parameter is renamed as a prediction.

full rationale

Walking the derivation chain, the network is trained on manually annotated diffraction patterns and outputs bounding-box centers via the DFL soft-argmax (Eqs. 1-3); strain is then computed from relative center displacements referenced to a region in the Si substrate. The comparison value for the SiGe layer strain is not taken from the network: it is estimated from XEDS composition using Eq. (5), attributed to Munshi et al., with independently measured lattice parameters and Poisson ratio. The reported 5e-4 figure is the standard deviation of strain in the nominally strain-free substrate, a repeatability metric rather than an accuracy metric; this is a validation limitation, not a circular reduction. The paper does not disclose whether the Si/SiGe test patterns overlap with the 42-pattern training set, which would be a generalization or data-leakage concern, but no quoted step shows a fitted parameter or a defined quantity being equivalent to the claimed result. Self-citations [13] and [14] appear only as related-work context and are not load-bearing in the derivation. The autocorrelation/Wiener-Khinchin discussion and the YOLOv8n/DFL components are adopted from external sources and are not used to define the outcome into existence. No specific equation, definition, or citation chain can be exhibited that reduces the predicted strain or phase maps to the model's training targets by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical: a neural network was trained on a small annotated set and applied to two materials. It does not rest on a new derivation, so the ledger mainly records hand-chosen hyperparameters and unverified domain assumptions. No new physical entities are introduced.

free parameters (3)
  • reg max (number of DFL bins) = 16
    Chosen as a model hyperparameter in Section 3.1; it sets the resolution of the soft-argmax center localization, but its effect on precision is not analyzed in the paper.
  • Input resolution = 640x640 train, 512x512 inference
    Hand-chosen image sizes for training and inference; affects both localization granularity and the reported throughput, but no sensitivity analysis is provided.
  • Inference batch size for Ti-80Nb benchmark = 64
    Selected operational parameter for the 121 fps measurement; different batch sizes would change the throughput number.
assumptions (5)
  • domain assumption Manual annotations of 42 diffraction patterns provide accurate ground truth for disc centers
    Section 3.2 describes the training data as manually annotated; any bias in the annotations directly transfers to the model's center predictions and therefore to the strain maps.
  • ad hoc to paper Wiener-Khinchin autocorrelation (Eq. 4) improves center localization in the detection pipeline
    The theorem is standard, but the paper does not show any equation or experiment connecting R(Δk) to the predicted box centers; it functions as an asserted improvement rather than a demonstrated one.
  • domain assumption The selected Si substrate region is strain-free and representative
    Section 4.2 uses this region as the zero-strain reference; foil bending and thickness variations could introduce real strain variation that would inflate the reported precision metric.
  • domain assumption YOLOv8n pretrained on COCO is a valid initialization for electron diffraction patterns
    Transfer learning from natural images is common but not obviously suited to diffraction patterns; no experiments test alternative initializations.
  • standard math Equation (5) estimates the expected out-of-plane strain for the measured composition
    Taken from Munshi et al. [33]; used to compare the measured mean strain of ~0.8% with a theoretical maximum of ~0.85%.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Object Detection for 4D STEM: High-Throughput Sub-Pixel Electron Diffraction Pattern Recognition." pith.science (2026). https://pith.science/paper/WNLMCESV

@misc{pith2026250604477,
  author       = {Pith},
  title        = {Pith review of: Neural Object Detection for 4D STEM: High-Throughput Sub-Pixel Electron Diffraction Pattern Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNLMCESV}},
  note         = {Machine review of arXiv:2506.04477}
}
abstract

High-throughput analysis of multidimensional transmission electron microscopy (TEM) datasets remains a significant challenge, limiting the broader impact on strategic materials research. Conventional workflows typically involve sequential, modular processing steps that necessitate extensive manual intervention and offline parameter tuning. In this work, we introduce an end-to-end post-processing framework for large-scale four-dimensional scanning TEM (4D-STEM) datasets, built around a highly efficient neural network-based object detection model. Central to our method is a sub-pixel accurate object center localization algorithm, which serves as the foundation for high-precision and high-throughput analysis of electron diffraction patterns. We demonstrate a strain measurement precision of 5x$10^{-4}$, quantified by the standard deviation of strain values within the strain-free Si substrate of a Si/SiGe multilayer TEM sample. Furthermore, by implementing an asynchronous, non-blocking object detection workflow, we achieve speeds exceeding 100 frames per second (fps), substantially accelerating the crystallographic phase identification and strain mapping in complex multiphase metallic alloys.

Figures

Figures reproduced from arXiv: 2506.04477 by the authors.

Figure 1
Figure 1. Top left to right: diagram of the neural diffraction pattern [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the NDPD architecture, featuring the backbone (C1-C5), neck (FPN (F3-F5) and PANet (P5-P3)) components, and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The flowchart illustrates the bounding box regression [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Comparing CPU and GPU performance in Synchronous [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: Illustration of anchor point-based center detection using [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: Top left to right: the virtual DF image of Si/SiGe multi [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Top left to right: example electron diffraction patterns [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 63 canonical work pages

  1. [1]

    Fast grain mapping with sub-nanometer resolution using 4d-stem with grain classification by princi- pal component analysis and non-negative matrix factoriza- tion

    Frances I Allen, Thomas C Pekin, Arun Persaud, Steven J Rozeveld, Gregory F Meyers, Jim Ciston, Colin Ophus, and Andrew M Minor. Fast grain mapping with sub-nanometer resolution using 4d-stem with grain classification by princi- pal component analysis and non-negative matrix factoriza- tion. Microscopy and Microanalysis, 27(4):794–803, 2021. 3

    work page 2021

  2. [2]

    Baccouche et al

    A. Baccouche et al. Early detection and classification of abnormality in prior mammograms using image-to-image translation and yolo techniques.Computer Methods and Pro- grams in Biomedicine, 221:106888, 2022. 3

    work page 2022

  3. [3]

    A B ´ech´e, JL Rouvi `ere, JP Barnes, and D Cooper. Strain measurement at the nanoscale: Comparison between conver- gent beam electron diffraction, nano-beam electron diffrac- tion, high resolution imaging and dark field electron holog- raphy. Ultramicroscopy, 131:10–23, 2013. 2, 6

    work page 2013

  4. [4]

    van Heijst, Jeroen J

    Maarten Bolhuis, Sabrya E. van Heijst, Jeroen J. M. Sangers, and Sonia Conesa-Boj. 4d-stem nanoscale strain analysis in van der waals materials: Advancing beyond planar configu- rations. Small Science, 4(3):2300249, 2024. 2, 3, 6

    work page 2024

  5. [5]

    Clough and A.I

    R. Clough and A.I. Kirkland. Direct digital electron detec- tors. In Advances in Imaging and Electron Physics , pages 1–42. Elsevier, 2016. 2

    work page 2016

  6. [6]

    Strain mapping of semicon- ductor specimens with nm-scale resolution in a transmission electron microscope

    David Cooper, Thibaud Denneulin, Nicolas Bernier, Armand B´ech´e, and Jean-Luc Rouvi`ere. Strain mapping of semicon- ductor specimens with nm-scale resolution in a transmission electron microscope. Micron, 80:145–165, 2016. 2, 6

    work page 2016

  7. [7]

    Phillip Crout, Dipanwita Chatterjee, Ingeborg Nævra Prestholdt, Tor Inge Thorsen, P. A. Midgley, and Antonius T. J. van Helvoort. Two-dimensional strain mapping with scanning precession electron diffraction: An investigation into data analysis routines, 2023. 2

    work page 2023

  8. [8]

    Centernet: Keypoint triplets for object detection

    Kaiwen Duan, Song Bai, Lingxi Xie, Honggang Qi, Qing- ming Huang, and Qi Tian. Centernet: Keypoint triplets for object detection. arXiv, 1904:08189v1, 2019. 3

    work page 1904

Show all 64 references
  1. [9]

    A novel automated method to measure strain at the nano scale

    B Freitag, J Stanley, E Sourty, J Ringnalda, and D Hubert. A novel automated method to measure strain at the nano scale. Microscopy and Microanalysis, 13(S02):834–835, 2007. 2

    2007

  2. [10]

    Phase object reconstruction for 4d-stem using deep learning

    Thomas Friedrich, Chu-Ping Yu, Johan Verbeeck, and San- dra Van Aert. Phase object reconstruction for 4d-stem using deep learning. Microscopy and Microanalysis , 29(1):395– 407, 2023. 2

    2023

  3. [11]

    Local and transient nanoscale strain map- ping during in situ deformation.Applied Physics Letters, 109 (8), 2016

    C Gammer, J Kacher, C Czarnik, OL Warren, J Ciston, and Andrew M Minor. Local and transient nanoscale strain map- ping during in situ deformation.Applied Physics Letters, 109 (8), 2016. 2

    2016

  4. [12]

    Strain field around individual dislocations controls failure (small methods 12/2024)

    Christoph Gammer, Inas Issa, Andrew M Minor, Robert O Ritchie, and Daniel Kiener. Strain field around individual dislocations controls failure (small methods 12/2024). Small Methods, 8(12):2470075, 2024. 2

    2024

  5. [13]

    Arda Genc, Libor Kovarik, and Hamish L. Fraser. A deep learning approach for semantic segmentation of unbalanced data in electron tomography of catalytic materials. Scientific Reports, 12:16267, 2022. 3

    2022

  6. [14]

    A versatile ma- chine learning workflow for high-throughput analysis of supported metal catalyst particles

    Arda Genc, Justin Marlowe, Anika Jalil, Daniel Belzberg, Libor Kovarik, and Phillip Christopher. A versatile ma- chine learning workflow for high-throughput analysis of supported metal catalyst particles. Ultramicroscopy, 271: 114116, 2025. 3

    2025

  7. [15]

    Asynchronous llm function calling

    In Gim, Seung-seob Lee, and Lin Zhong. Asynchronous llm function calling. arXiv preprint arXiv:2412.07017, 2024. 3, 5

    2024 arXiv

  8. [16]

    Automated analysis of transmission electron micrographs of metallic nanoparticles by machine learning

    Nina Gumbiowski, Kateryna Loza, Marc Heggen, and Matthias Epple. Automated analysis of transmission electron micrographs of metallic nanoparticles by machine learning. Nanoscale Adv., 5:2318–2326, 2023. 3

    2023

  9. [17]

    Electron bessel beam diffraction for precise and accu- rate nanoscale strain mapping

    Giulio Guzzinati, Wannes Ghielens, Christoph Mahr, Ar- mand B´ech´e, Andreas Rosenauer, Toon Calders, and Jo Ver- beeck. Electron bessel beam diffraction for precise and accu- rate nanoscale strain mapping. Applied Physics Letters, 114 (24):243501, 2019. 4

    2019

  10. [18]

    Haifawi et al

    H. Haifawi et al. Drone detection & classification with surveillance ‘radar on-the-move’ and yolo. In Proceedings of the IEEE Radar Conference, pages 1–6. IEEE, 2023. 3

    2023

  11. [19]

    Horwath, Dmitri N

    James P. Horwath, Dmitri N. Zakharov, R ´emi M ´egret, and Eric A. Stach. Understanding important features of deep learning models for segmentation of high-resolution trans- mission electron microscopy images. npj Computational Materials, 6(1):108, 2020. 3

    2020

  12. [20]

    Jia et al

    X. Jia et al. Fast and accurate object detector for autonomous driving based on improved yolov5. Scientific Reports, 13: 9711, 2023. 3

    2023

  13. [21]

    Tate, Jiwoong Park, Sol M

    Yi Jiang, Zhen Chen, Yimo Han, Pratiti Deb, Hui Gao, Saien Xie, Prafull Purohit, Mark W. Tate, Jiwoong Park, Sol M. Gruner, Veit Elser, and David A. Muller. Electron ptychogra- phy of 2d materials to deep sub-˚angstr¨om resolution. Nature, 559(7714):343–349, 2018. 2

    2018

  14. [22]

    Yolo by ultralytics (version 8.0.0)

    Glenn Jocher, Ankit Chaurasia, and Jason Qiu. Yolo by ultralytics (version 8.0.0). https : / / github . com / ultralytics/ultralytics , 2023. Computer soft- ware. 2, 3

    2023

  15. [23]

    Korrelationstheorie der station ¨aren stochastischen prozesse

    Aleksandr Khinchin. Korrelationstheorie der station ¨aren stochastischen prozesse. Mathematische Annalen , 109(1): 604–615, 1934. 4

    1934

  16. [24]

    Atomic-scale strain analysis for advanced si/sige heterostructure by using transmission electron microscopy

    Lan Li, Ran Bi, Zuoyuan Dong, Changqing Ye, Jing Xie, Chaolun Wang, Xiaomei Li, Kin-Leong Pey, Ming Li, and Xing Wu. Atomic-scale strain analysis for advanced si/sige heterostructure by using transmission electron microscopy. Electron, 2(2):e32, 2024. 2, 6

    2024

  17. [25]

    Manifold learning of four-dimensional scanning transmission electron microscopy

    Xin Li, Ondrej E Dyck, Mark P Oxley, Andrew R Lupini, Leland McInnes, John Healy, Stephen Jesse, and Sergei V Kalinin. Manifold learning of four-dimensional scanning transmission electron microscopy. npj Computational Ma- terials, 5(1):5, 2019. 2

    2019

  18. [26]

    Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection

    Xiang Li, Wenhai Wang, Lijun Wu, Shuo Chen, Xiaolin Hu, Jun Li, Jinhui Tang, and Jian Yang. Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection. In Advances in Neural Information Pro- cessing Systems (NeurIPS), 2020. 3

    2020

  19. [27]

    Feature pyramid networks for object detection

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 936–944. IEEE Computer Society, 2017. 3 9

    2017

  20. [28]

    Path aggregation network for instance seg- mentation

    Shifeng Liu, Xiaohui Shen, Jingdong Yang, Xizhou Wang, and Jian Sun. Path aggregation network for instance seg- mentation. arXiv, 1803:01534v4, 2018. 3

    2018

  21. [29]

    Comparing different software packages for the mapping of strain from scanning precession diffraction data

    Ian MacLaren, Emma Devine, Hristo Gergov, Gary Paterson, K P Harikrishnan, Benjamin Savitzky, Colin Ophus, Ren- liang Yuan, Jian-Min Zuo, Kirsten Forster, Gaja Kobe, Eliz- abeth Koppany, Kirsten McClymont, Anjelo Narendran, and David Riley. Comparing different software package...

    2021

  22. [30]

    Krause, Dennis Zillmann, and Andreas Rosenauer

    Christoph Mahr, Knut M ¨uller-Caspary, Tim Grieb, Marco Schowalter, Thorsten Mehrtens, Florian F. Krause, Dennis Zillmann, and Andreas Rosenauer. Theoretical study of pre- cision and accuracy of strain analysis by nano-beam electron diffraction. Ultramicroscopy, 158:38–48, 2015. 2

    2015

  23. [31]

    Mixed precision training, 2018

    Paulius Micikevicius, Sharan Narang, Jonah Alben, Gregory Diamos, Erich Elsen, David Garcia, Boris Ginsburg, Michael Houston, Oleksii Kuchaiev, Ganesh Venkatesh, and Hao Wu. Mixed precision training, 2018. 5

    2018

  24. [32]

    Debangshu Mukherjee, Jocelyn T. L. Gamler, Sara E. Skra- balak, and Raymond R. Unocic. Lattice strain measurement of core@shell electrocatalysts with 4d scanning transmission electron microscopy nanobeam electron diffraction. ACS Catalysis, 10(10):5529–5541, 2020. 2

    2020

  25. [33]

    Disentangling multiple scattering with deep learning: application to strain mapping from electron diffrac- tion patterns

    Joydeep Munshi, Alexander Rakowski, Benjamin H Sav- itzky, Steven E Zeltmann, Jim Ciston, Matthew Henderson, Shreyas Cholia, Andrew M Minor, Maria KY Chan, and Colin Ophus. Disentangling multiple scattering with deep learning: application to strain mapping from electron diffra...

    2022

  26. [34]

    Strain measurement in semiconductor heterostructures by scanning transmission electron microscopy

    Knut M ¨uller, Andreas Rosenauer, Marco Schowalter, Josef Zweck, Rafael Fritz, and Kerstin V olz. Strain measurement in semiconductor heterostructures by scanning transmission electron microscopy. Microscopy and Microanalysis, 18(5): 995–1009, 2012. 2

    2012

  27. [35]

    Four-dimensional scanning transmission elec- tron microscopy (4d-stem): From scanning nanodiffraction to ptychography and beyond

    Colin Ophus. Four-dimensional scanning transmission elec- tron microscopy (4d-stem): From scanning nanodiffraction to ptychography and beyond. Microscopy and Microanaly- sis, 25(3):563–582, 2019. 2

    2019

  28. [36]

    Automated crystal orientation map- ping in py4dstem using sparse correlation matching

    Colin Ophus, Steven E Zeltmann, Alexandra Bruefach, Alexander Rakowski, Benjamin H Savitzky, Andrew M Mi- nor, and Mary C Scott. Automated crystal orientation map- ping in py4dstem using sparse correlation matching. Mi- croscopy and Microanalysis, 28(2):390–403, 2022. 2

    2022

  29. [37]

    Strain mapping at nanometer resolu- tion using advanced nano-beam electron diffraction

    VB Ozdol, C Gammer, XG Jin, P Ercius, C Ophus, J Cis- ton, and AM Minor. Strain mapping at nanometer resolu- tion using advanced nano-beam electron diffraction. Applied Physics Letters, 106(25), 2015. 2

    2015

  30. [38]

    The exit-wave power-cepstrum transform for scan- ning nanobeam electron diffraction: robust strain mapping at subnanometer resolution and subpicometer precision

    Elliot Padgett, Megan E Holtz, Paul Cueva, Yu-Tsun Shao, Eric Langenberg, Darrell G Schlom, and David A Muller. The exit-wave power-cepstrum transform for scan- ning nanobeam electron diffraction: robust strain mapping at subnanometer resolution and subpicometer precision. Ult...

    2020

  31. [39]

    High-resolution mapping of strain partitioning and relaxation in ingan/gan nanowire heterostructures

    Bumsu Park, Ja Kyung Lee, Christoph T Koch, Martin W¨olz, Lutz Geelhaar, and Sang Ho Oh. High-resolution mapping of strain partitioning and relaxation in ingan/gan nanowire heterostructures. Advanced Science, 9(22):2200323, 2022. 2

    2022

  32. [40]

    A. L. Patterson. A fourier series method for the determina- tion of the components of interatomic distances in crystals. Phys. Rev., 46:372–376, 1934. 5

    1934

  33. [41]

    Pekin, Christoph Gammer, Jim Ciston, An- drew M

    Thomas C. Pekin, Christoph Gammer, Jim Ciston, An- drew M. Minor, and Colin Ophus. Optimizing disk regis- tration algorithms for nanobeam electron diffraction strain mapping. Ultramicroscopy, 176:170–176, 2017. 70th Birth- day of Robert Sinclair and 65th Birthday of Nestor J. ...

    2017

  34. [42]

    Yolo9000: Better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: Better, faster, stronger. arXiv, 1612:08242v1, 2016. 2, 3

    2016

  35. [43]

    U- Net: Convolutional Networks for Biomedical Image Seg- mentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- Net: Convolutional Networks for Biomedical Image Seg- mentation. In MICCAI, 2015. 3

    2015

  36. [44]

    Bekkers, Marleen Hoogendoorn, and Remco Duits

    Luca Sangalli, Erik J. Bekkers, Marleen Hoogendoorn, and Remco Duits. Scale-equivariant u-net. In Proceedings of the British Machine Vision Conference (BMVC), 2022. 3

    2022

  37. [45]

    Cepstral scanning transmission electron microscopy imaging of severe lattice distortions.Ul- tramicroscopy, 231:113252, 2021

    Yu-Tsun Shao, Renliang Yuan, Haw-Wen Hsiao, Qun Yang, Yang Hu, and Jian-Min Zuo. Cepstral scanning transmission electron microscopy imaging of severe lattice distortions.Ul- tramicroscopy, 231:113252, 2021. 2

    2021

  38. [46]

    Uncovering material deformations via machine learning combined with four-dimensional scanning transmis- sion electron microscopy

    Chuqiao Shi, Michael C Cao, Sarah M Rehn, Sang-Hoon Bae, Jeehwan Kim, Matthew R Jones, David A Muller, and Yimo Han. Uncovering material deformations via machine learning combined with four-dimensional scanning transmis- sion electron microscopy. npj Computational Materials , ...

    2022

  39. [47]

    Domain-dependent strain and stack- ing in two-dimensional van der waals ferroelectrics

    Chuqiao Shi, Nannan Mao, Kena Zhang, Tianyi Zhang, Ming-Hui Chiu, Kenna Ashen, Bo Wang, Xiuyu Tang, Galio Guo, Shiming Lei, et al. Domain-dependent strain and stack- ing in two-dimensional van der waals ferroelectrics. Nature communications, 14(1):7168, 2023. 2

    2023

  40. [48]

    David J. Smith. Nanocharacterisation. In Nanocharacteri- sation, pages 1–29. The Royal Society of Chemistry, Cam- bridge, UK, 2nd edition, 2015. 1

    2015

  41. [49]

    Using stem with quasi- parallel illumination and an automated peak-finding routine for strain analysis at the nanometre scale

    E Sourty, J Stanley, and B Freitag. Using stem with quasi- parallel illumination and an automated peak-finding routine for strain analysis at the nanometre scale. In 2009 16th IEEE International Symposium on the Physical and Failure Anal- ysis of Integrated Circuits, pages 479...

    2009

  42. [50]

    Sunde, C.D

    J.K. Sunde, C.D. Marioara, A.T.J. van Helvoort, and R. Holmestad. The evolution of precipitate crystal structures in an al-mg-si(-cu) alloy studied by a combined haadf-stem and sped approach. Materials Characterization, 142:458– 469, 2018. 3

    2018

  43. [51]

    Generalization across experimental parameters in neural net- work analysis of high-resolution transmission electron mi- croscopy datasets

    Katherine Sytwu, Luis Rangel DaCosta, and Mary C Scott. Generalization across experimental parameters in neural net- work analysis of high-resolution transmission electron mi- croscopy datasets. Microscopy and Microanalysis , 30(1): 85–95, 2024. 3

    2024

  44. [52]

    Tate, Prafull Purohit, Darol Chamberlain, Kayla X

    Mark W. Tate, Prafull Purohit, Darol Chamberlain, Kayla X. Nguyen, Robert Hovden, Celesta S. Chang, Pratiti Deb, Em- rah Turgut, John T. Heron, Darrell G. Schlom, Daniel C. Ralph, Gregory D. Fuchs, Katherine S. Shanks, Hugh T. 10 Philipp, David A. Muller, and Sol M. Gruner. Hi...

  45. [53]

    Thronsen, T

    E. Thronsen, T. Bergh, T.I. Thorsen, E.F. Christiansen, J. Frafjord, P. Crout, A.T.J. van Helvoort, P.A. Midgley, and R. Holmestad. Scanning precession electron diffraction data analysis approaches for phase mapping of precipitates in alu- minium alloys. Ultramicroscopy, 255:1...

    2024

  46. [54]

    Fcos: Fully convolutional one-stage object detection

    Zhi Tian, Chunhua Shen, Hao Chen, and Tong He. Fcos: Fully convolutional one-stage object detection. arXiv, 1904: 01355v1, 2019. 3

    1904

  47. [55]

    K. P. Treder, C. Huang, C. G. Bell, B. M. Noko, Y . Han, W. Moritz, M. Michaelis, F. Meirer, M. Mavrikakis, B. Gault, B. Roldan Cuenya, M. M. Deegan, N. Marzari, and C. Schnei- der. nnpipe: a neural network pipeline for automated analysis of morphologically diverse catalyst sy...

    2023

  48. [56]

    Autodisk: Automated diffraction processing and strain map- ping in 4d-stem

    Sihan Wang, Tim B Eldred, Jacob G Smith, and Wenpei Gao. Autodisk: Automated diffraction processing and strain map- ping in 4d-stem. Ultramicroscopy, 236:113513, 2022. 2

    2022

  49. [57]

    Generalized harmonic analysis

    Norbert Wiener. Generalized harmonic analysis. Acta Math- ematica, 55(1):117–258, 1930. 4

    1930

  50. [58]

    Flexible for- mation of coherent probes on an aberration-corrected stem with three condensers

    Feng Yi, Peter Tiemeijer, and Paul M V oyles. Flexible for- mation of coherent probes on an aberration-corrected stem with three condensers. Journal of electron microscopy , 59 (S1):S15–S21, 2010. 7

    2010

  51. [59]

    Un- supervised machine learning and cepstral analysis with 4d- stem for characterizing complex microstructures of metallic alloys

    Timothy Yoo, Eitan Hershkovitz, Yang Yang, Fl ´avia da Cruz Gallo, Michele V Manuel, and Honggyu Kim. Un- supervised machine learning and cepstral analysis with 4d- stem for characterizing complex microstructures of metallic alloys. npj Computational Materials, 10(1):223, 2024. 2, 3

    2024

  52. [60]

    Lattice strain mapping using circular hough transform for electron diffraction disk detection

    Renliang Yuan, Jiong Zhang, and Jian-Min Zuo. Lattice strain mapping using circular hough transform for electron diffraction disk detection. Ultramicroscopy, 207:112837,

  53. [61]

    Training artificial neural networks for precision ori- entation and strain mapping using 4d electron diffraction datasets

    Renliang Yuan, Jiong Zhang, Lingfeng He, and Jian-Min Zuo. Training artificial neural networks for precision ori- entation and strain mapping using 4d electron diffraction datasets. Ultramicroscopy, 231:113256, 2021. 2, 3

    2021

  54. [62]

    Patterned probes for high precision 4d-stem bragg measurements

    Steven E Zeltmann, Alexander M ¨uller, Karen C Bustillo, Benjamin Savitzky, Lauren Hughes, Andrew M Minor, and Colin Ophus. Patterned probes for high precision 4d-stem bragg measurements. Ultramicroscopy, 209:112890, 2020. 2

    2020

  55. [63]

    J. M. Zuo, M. Gao, J. Tao, B. Q. Li, R. Twesten, and I. Petrov. Coherent nano-area electron diffraction. Microscopy Research and Technique, 64(5-6):347–355, 2004. 7

    2004

  56. [64]

    Crystal phase mapping by scanning preces- sion electron diffraction and machine learning decomposi- tion

    H ˚akon W ˚Anes, Ingrid Marie Andersen, and Antonius T J van Helvoort. Crystal phase mapping by scanning preces- sion electron diffraction and machine learning decomposi- tion. Microscopy and Microanalysis, 24(S1):586–587, 2018. 2 11

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.