Pith. sign in

REVIEW 6 major objections 5 minor 52 references

4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion

T0 review · 6 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read 4D-MISR fuses sub-pixel-shifted virtual bright-field views from a single 4D-STEM scan and claims atomic-scale structure at 200 electrons per square angstrom, where ptychography fails below 500.

desk verdict A creative fusion of multi-view VBF images for low-dose 4D-STEM, undercut by an undefined train/test split and an architecture description that doesn't match the claims. read the letter →

arxiv 2507.09953 v3 pith:L2SG24PK submitted 2025-07-14 cs.CV

Zifei Wang , Zian Mao , Xiaoya He , Xi Huang , Haoran Zhang , Chun Cheng , Shufen Chu , Tingzheng Hou
show 2 more authors
Xiaoqin Zeng Yujun Xie
This is my paper · ORCID
classification cs.CV
keywords 4D-STEMlow-doseelectronmicroscopymulti-imagesuper-resolutionbeam-sensitivematerialsdeeplearningreconstructionvirtualbright-fieldimagingptychographyfeaturefusion
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

The paper proposes 4D-MISR, a deep-learning reconstruction method that turns the many low-resolution virtual bright-field views already present in a single 4D-STEM scan into one high-resolution atomic-scale image. A dual-path attention network fuses these sub-pixel-shifted views and up-samples them, and the training targets are infinite-dose ptychographic reconstructions of simulated specimens. The authors claim that on crystalline, semi-crystalline, and amorphous test materials this recovers atomic features at doses as low as 200 electrons per square angstrom, where conventional ptychography cannot resolve structure below about 500 electrons per square angstrom, and that it reaches equivalent structural recovery with roughly 60% fewer incident electrons. A sympathetic reader would care because beam-sensitive materials such as proteins, porous frameworks, and battery cathodes currently cannot be imaged at atomic resolution without radiation damage.

What carries the argument

The load-bearing mechanism is the reciprocity identity of 4D-STEM: each pixel of the detector within the central unscattered disc corresponds to illuminating the specimen with a tilted plane wave, so a single scan contains many low-resolution views of the same object with sub-pixel shifts. The network is an Attention U-Net encoder-decoder with attention gates that weight each virtual view per pixel, followed by a PixelShuffle sub-pixel convolution that upsamples the fused feature map by a factor $r=3$; a $1\times1$ convolution suppresses checkerboard artifacts. Training uses simulated multi-slice 4D-STEM data with on-the-fly dose-dependent noise injection and infinite-dose ptychographic reconstructions as ground truth, with a composite loss combining $L_1$, perceptual similarity, and MS-SSIM. The paper's argument is that this learned feature fusion recovers high-frequency components that shot-noise-limited Fourier inversion loses.

What would settle it

Retrain 4D-MISR on a set of simulated structures, then test it on a structure whose atomic model was never shown during training, using the same 200-electron-per-square-angstrom conditions; if atomic features appear only for structures resembling the training set and disappear for the unseen structure, the claimed generalization and dose advantage are not established. Reporting the exact training/test partition would settle whether the current figures come from generalization or memorization.

Watch

Extended reading notes

Core claim

The paper's central claim is that the information needed for atomic-scale super-resolution is already present in an ultra-low-dose 4D-STEM measurement, provided it is viewed as a multi-view image stack rather than as a single phase-retrieval problem. By integrating different detector pixels inside the central diffraction disc, one scan yields many virtual bright-field images that are mutually sub-pixel-shifted, and the network learns to fuse them into a sharper image. On Li2CoO3, apoferritin, and Cu-Zr metallic glass, the authors report visible atomic or molecular features at 200 electrons per square angstrom with 4D-MISR while ptychography from the same data fails below 500, and a greater than threefold contrast-to-noise improvement at low dose. The paper frames this as pushing beyond the electron-dose limit of conventional electron microscopy.

Load-bearing premise

The evaluation assumes that the test materials were held out from training, but the paper never defines a train/test split and generates training data from the same classes of structures, so the low-dose reconstructions could reflect memorization of target images rather than extraction of new information from the data.

Editorial extensions

If this is right

  • If the claim holds, the dose budget for atomic-scale imaging of radiation-sensitive materials drops by roughly 60%, making proteins, metal-organic frameworks, and battery interfaces accessible to 4D-STEM without cryogenic or low-dose compromises.
  • The method would become a software-only upgrade: standard 4D-STEM datasets, acquired without hardware modification, could be post-processed to super-resolution images.
  • The reported contrast-to-noise improvement of more than threefold at low dose suggests the practical limit is no longer the Poisson noise of the measurement but the quality and diversity of the simulated training data.
  • The strong central-region performance implies practical acquisitions may need overlapping scans or a confidence map to extend high-fidelity reconstruction across the full field of view.

Reading between the lines

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

  • A direct extension the paper leaves untested is cross-instrument transfer: because training is purely simulated, real-world use would likely require a per-instrument calibration of detector geometry, convergence angle, and noise statistics before the claimed gains appear.
  • The authors state that the exact codebase is unavailable for proprietary reasons and only the model code is released; independent replication of the training pipeline is therefore not yet possible from the paper alone.
  • If the 60% dose reduction is real, the freed dose budget could be spent on time-resolved or in-situ series rather than single static images, which the paper does not discuss.
  • Because the ground truth is infinite-dose ptychography, the network can at best inherit the fidelity of that phase-retrieval forward model; any systematic error in it would be baked into the training targets.
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

6 major / 5 minor

Summary. This manuscript proposes 4D-MISR, a deep-learning super-resolution method for low-dose 4D-STEM imaging. The method constructs multiple virtual bright-field (VBF) images from different detector regions of a 4D-STEM dataset, feeds them into an attention U-Net with sub-pixel upsampling, and is trained on simulated data whose targets are infinite-dose ptychographic reconstructions. The authors claim atomic-scale super-resolution at doses as low as 200 e-/Å^2 and a 60% reduction in incident electron dose relative to conventional ptychography, with improved contrast-to-noise ratio across crystalline, semi-crystalline, and amorphous samples. Validation is performed entirely by simulation on Li2CoO3, apoferritin, and Cu-Zr, with qualitative comparisons to ptychography and other methods.

Significance. If the claims were established, the approach would be valuable for beam-sensitive materials, since it addresses a real bottleneck in low-dose electron microscopy and offers a plausible strategy of adapting multi-image super-resolution to 4D-STEM. The authors make the 4D-MISR code available and describe the training pipeline in enough detail to be reproducible. However, the current evaluation does not substantiate the central claims: no train/test split is defined, the baseline is the same algorithm family as the training target, no quantitative CNR or resolution metrics are reported, and all data are simulated. As submitted, the paper is a proposal rather than a validated method, and the load-bearing evidence for the advertised dose reduction and resolution gains is missing.

major comments (6)
  1. [Sec. 2.3 / Sec. 3.2] The manuscript never defines a train/test split; the test materials Li2CoO3 (mp-1173879), apoferritin (8RQB), and Cu-Zr are drawn from the same sources (Materials Project and PDB) used to generate training data in Sec. 2.3, and no exclusion of these exact entries is stated. If those structures were present in the training set, the low-dose 'reconstructions' in Fig. 4 could be memorized outputs rather than generalized inference, so the claimed super-resolution and 60% dose reduction are not established. Please state explicitly whether these entries were held out and provide evaluation on structures that are verifiably absent from training.
  2. [Sec. 2.1.2 / Sec. 3.2 / Fig. 4] The training targets are infinite-dose ptychographic reconstructions and the baseline is a low-dose ptychographic reconstruction obtained with the same algorithm family (py4DSTEM). The comparison is therefore not independent of the training target; at best it demonstrates that the network can regress the target ptychography output from low-dose inputs. A meaningful evaluation needs ground-truth atomic models or another reference independent of the ptychography pipeline, together with quantitative error metrics such as PSNR or column-position error against the known structure.
  3. [Sec. 3.2 / Introduction] The abstract and introduction claim a '>3x improvement in contrast-to-noise ratio' and Sec. 3.2 claims 'equivalent structural recovery with 60% fewer incident electrons', but no CNR values, resolution estimates, error bars, or sample sizes are reported anywhere in the results. The line scans in Fig. 4 are qualitative and the power-spectrum analysis is only referenced as Supplementary Figure 3 without numeric resolution criteria. These central quantitative claims are therefore unsupported as stated.
  4. [Sec. 2.1.1 / Sec. 2.2 / Conclusion] The formal architecture description in Sec. 2.1.1 is a single Attention U-Net followed by PixelShuffle and a 1x1 convolution, while the Abstract, Sec. 2.2, and Conclusion repeatedly describe a 'dual-path' or 'dual-branch' network with an inter-view branch that fuses angular and spatial features. No such dual-path module appears in the equations or layer-by-layer description, so the published method does not match its central design claim. This discrepancy must be resolved by providing the actual architecture or explicitly retracting the dual-path description.
  5. [Sec. 3 / Data availability] All validation is performed on simulated 4D-STEM data; no experimental low-dose dataset is shown, yet the manuscript claims generalizability to real-world EM applications and radiation-vulnerable specimens (Secs. 2.1.2, 3.3). At minimum, the authors should clearly restrict their claims to the simulation domain and, ideally, include one experimental demonstration or a quantitative discussion of the domain gap between simulated VBF images and real detector data.
  6. [Sec. 2.1.2] Low-dose electron counting statistics are Poissonian, especially at the claimed 100-200 e-/Å^2 per probe, but the training procedure simulates low-dose conditions by adding Gaussian noise to infinite-dose 4D-STEM data. The authors do not justify this approximation or compare against a Poisson-based noise model, so the training distribution may not match the target low-dose regime. This is a load-bearing assumption for the claimed low-dose performance and should be validated or corrected.
minor comments (5)
  1. [Abstract / Sec. 4] There are several typos and awkward phrasings, e.g., 'one-stem expands' in the abstract, 'Often partially crystalline' in Sec. 3.1, and the author-contribution section lists 'Zhengting Hou' while the author list gives 'Tingzheng Hou'.
  2. [Sec. 3.2 / Fig. 4 caption] The caption sentence 'The electron doses used for the low-dose reconstructions are 300 e-/Å^2 for (d), (f) and (e)' is unclear; it should specify the dose for each panel and the intended sample labels.
  3. [Sec. 3.2] The phrase '100-infinite e-/Å^2' is not a well-defined dose range; please write the actual tested dose values and clarify whether 'infinite dose' is a simulation construct.
  4. [Eqs. (3)-(4)] The loss function switches based on 'step size < 1 Å' versus 'step size > 1 Å', but the equality case is not addressed and the choice of threshold is not motivated in the text.
  5. [Data availability] The data-availability statement is internally contradictory: it says the exact codebase is unavailable due to proprietary reasons but then states that the 4D-MISR code is available on GitHub. Please clarify which parts are public and which are proprietary.

Circularity Check

1 steps flagged · score 6.0 of 10

The low-dose 'super-resolution' evaluation may reduce to memorized ptychographic targets because the paper never defines a train/test split and the three test samples come from the same Materials Project and PDB databases used to generate training data.

  1. fitted input called prediction [Sec. 2.1.2, Sec. 2.3, Sec. 3.2 (Model Training; Atomic Model Generation; Super-Resolution under Low Electron Dose)]
    "the network inputs are the low-dose VBF images, while the targets are the high-fidelity ptychographic reconstructions that would be obtained under an infinite-dose assumption. ... The simulated dataset was generated using ... atomic models derived from Crystallographic Information File (CIF) sourced from the Materials Project [36] and Program Database File (PDB) obtained from the Protein Data Bank [37]. ... we conducted tests on Li2CoO3 (2 nm thickness, Materials Project ID: mp-1173879), apoferritin (16.1 nm thickness, PDB ID: 8RQB) and Cu-Zr metallic glass."

    Sec. 2.3 creates training data from Materials Project CIFs and PDB entries, while Sec. 3.2 evaluates exactly identified entries from those two databases. The paper never states a train/test split or exclusion. If these entries were in the training corpus, the network's loss (Eqs. 3-7) was minimized against the infinite-dose ptychographic images of the same Li2CoO3, apoferritin, and Cu-Zr structures, so the reported low-dose reconstructions and the '60% fewer electrons' comparison are fitted recalls of memorized targets. Since the baseline is ptychography, the same algorithm family that produced the training labels, 'closely resemble the infinite-dose ground truth' is the training objective restated.

full rationale

The central result is not a first-principles derivation: 4D-MISR is a supervised CNN trained to map low-dose VBF images to infinite-dose ptychographic reconstructions. The paper's own equations define loss functions against those ptychographic targets, so the network is, by construction, a learned emulator of ptychography. In a properly split evaluation, that could still support a useful low-dose denoising or super-resolution claim. However, Sec. 2.3 states that training data are generated from Materials Project CIFs and PDB entries, and Sec. 3.2 evaluates Li2CoO3 (mp-1173879), apoferritin (8RQB), and Cu-Zr without stating that these entries were excluded from the training set. Since the training targets are the infinite-dose ptychographic reconstructions of exactly those atomistic models, inclusion of these test structures would make the reported 'equivalent structural recovery' and '60% fewer electrons' comparisons a measure of memorization rather than generalization. The comparison baseline is the same ptychographic algorithm family used to produce the labels, so 'closely resemble the infinite-dose ground truth' restates the training objective. No load-bearing self-citation chain appears in the paper. The missing train/test specification is therefore the key circularity risk; because the paper's text does not positively confirm leakage, the score is 6 (partial circularity by fitted-input-called-prediction) rather than 8-10.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on five unverified or unstated premises: the fidelity of abTEM simulations to real experiments, the adequacy of ptychographic reconstructions as ground truth, the physical validity of the virtual-bright-field-as-multi-view analogy, the equivalence of Gaussian training noise to Poisson shot noise, and the independence of the evaluation specimens from the training set. The first two are standard practices but not demonstrated here; the last three are load-bearing assumptions introduced for this paper. No new physical entities are introduced.

free parameters (2)
  • perceptual loss weight lambda = 0.006
    Set by hand following SRGAN (Ref. 32); controls the trade-off between pixel and perceptual losses in Eq. 3, so all reported reconstructions depend on this value.
  • super-resolution scale r = 3
    Chosen as the upsampling factor in the PixelShuffle module (Sec. 2.1.1); the claim of atomic-scale super-resolution is only demonstrated at this fixed factor and the scan step of 4 Angstrom.
assumptions (5)
  • domain assumption abTEM multi-slice simulations with the stated parameters (300 keV, 10 mrad, 4 Å scan step, defocus 1500 Å) faithfully represent real 4D-STEM scattering for the materials studied.
    All training and evaluation data are simulated (Sec. 2.3); there is no experimental 4D-STEM data, so the transferability to real instruments is assumed.
  • domain assumption Infinite-dose ptychographic reconstructions are a valid ground truth for atomic-scale structure.
    Used as training targets (Sec. 2.1.2). Ptychography itself is an approximate phase-retrieval method; treating its output as ground truth assumes the algorithm is correct and dose-independent.
  • ad hoc to paper Virtual bright-field images from different detector regions behave like sub-pixel-shifted multi-view observations, so MISR principles apply.
    This is the core premise of the method (Sec. 2.2), but no shift/parallax analysis or image registration is provided; the analogy is asserted.
  • ad hoc to paper Gaussian noise injection during training accurately simulates low-dose electron counting statistics.
    Sec. 2.1.2 says Gaussian noise per epoch is used, while Sec. 1 argues Poisson shot noise dominates low-dose imaging; no validation that Gaussian noise produces equivalent reconstructions.
  • ad hoc to paper The evaluation materials Li2CoO3, apoferritin, and Cu-Zr are independent of the training set.
    No train/test split is stated (Secs. 2.3, 3.2); if they were in training, the results measure memorization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion." pith.science (2026). https://pith.science/paper/L2SG24PK

@misc{pith2026250709953,
  author       = {Pith},
  title        = {Pith review of: 4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L2SG24PK}},
  note         = {Machine review of arXiv:2507.09953}
}
read the original abstract

While electron microscopy offers crucial atomic-resolution insights into structure-property relationships, radiation damage severely limits its use on beam-sensitive materials like proteins and 2D materials. To overcome this challenge, we push beyond the electron dose limits of conventional electron microscopy by adapting principles from multi-image super-resolution (MISR) that have been widely used in remote sensing. Our method fuses multiple low-resolution, sub-pixel-shifted views and enhances the reconstruction with a convolutional neural network (CNN) that integrates features from synthetic, multi-angle observations. We developed a dual-path, attention-guided network for 4D-STEM that achieves atomic-scale super-resolution from ultra-low-dose data. This provides robust atomic-scale visualization across amorphous, semi-crystalline, and crystalline beam-sensitive specimens. Systematic evaluations on representative materials demonstrate comparable spatial resolution to conventional ptychography under ultra-low-dose conditions. Our work expands the capabilities of 4D-STEM, offering a new and generalizable method for the structural analysis of radiation-vulnerable materials.

Figures

Figures reproduced from arXiv: 2507.09953 by the authors.

Figure 1
Figure 1. (a) Multi-view electron optical configurations for 4D-STEM with defocus. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Overview of proposed workflow. The pipeline integrates simulation-based training data generation from atomic models, experimental data calibration with domain interconversion mapping, and a 4D-MISR deep learning model that performs intra-view feature extraction and inter-view information fusion to reconstruct high-fidelity super￾resolution images across a range of electron dose conditions [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 3
Figure 3. 4D-MISR reconstruction results under low electron dose for materials with different structural orders. (a) Crystalline materials (MoS2, MOF, NCM, CaTiO3) show clear lattice periodicity. (b) Semi-crystalline biomolecules (apoferritin, α-synuclein truncation, neurotensin, eukaryotic protein) retain overall morphology. (c) Amorphous materials (PEO, Cu-Zr, LiPON, Li5La3Nb2O12) display disordered atomic arrangements. Lef… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparison of 4D-MISR and ptychography reconstructions for various samples under different electron dose conditions. (a) Li2CoO3, (b) Cu-Zr metallic glass, and (c) apoferritin reconstructions are shown for electron doses ranging from 100 to 1000 e−/˚A2 . For each sampl…
Figure 5
Figure 5. Figure 5: Qualitative comparison of 4D-MISR with ptychography, bright-field [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 49 canonical work pages

  1. [1]

    Lv J, Zhang H, Zhang D, Liu L and Han Y 2022 Acc. Mater. Res.3 552–564

    work page 2022

  2. [2]

    Liu L, Zhang D, Zhu Y and Han Y 2020 Commun. Chem. 3 99

    work page 2020

  3. [3]

    Zhou Y, Dong Z, Terasaki O and Ma Y 2021 Acc. Mater. Res.3 110–121

    work page 2021

  4. [4]

    Microanal

    Ghosh S, Kumar P, Conrad S, Tsapatsis M and Mkhoyan K A 2019 Microsc. Microanal. 25 1704–1705

    work page 2019

  5. [5]

    2024 Nat

    K¨ u¸ c¨ uko˘ glu B, Mohammed I, Guerrero-Ferreira R C, Ribet S M, Varnavides G, Leidl M L, Lau K, Nazarov S, Myasnikov A, Kube M et al. 2024 Nat. Commun. 15 8062

    work page 2024

  6. [6]

    2025 Nat

    Li G, Xu M, Tang W Q, Liu Y, Chen C, Zhang D, Liu L, Ning S, Zhang H, Gu Z Y et al. 2025 Nat. Commun. 16 914

    work page 2025

  7. [7]

    2020 Nat

    Zhou L, Song J, Kim J S, Pei X, Huang C, Boyce M, Mendon¸ ca L, Clare D, Siebert A, Allen C S et al. 2020 Nat. Commun. 11 2773

    work page 2020

  8. [8]

    Today Sustainability100799

    Patra K and Pal H 2024 Mater. Today Sustainability100799

    work page 2024

Show all 52 references
  1. [9]

    Gryzi´ nski M 1965Phys. Rev. 138 A336

  2. [10]

    Xu T, Zheng H, Wang J, Banhart F and Sun L 2023 Electron beam irradiation effects and in-situ irradiation of nanomaterials In-Situ Transmission Electron Microscopy(Springer) pp 17–51

    2023

  3. [11]

    187 103–111

    Jiang T T, Wang X D, Wang J J, Zhou Y X, Zhang D L, Lu L, Jia C L, Wuttig M, Mazzarello R and Zhang W 2020 Acta Mater. 187 103–111

    2020

  4. [12]

    9 900–933

    Zhan Z, Liu Y, Wang W, Du G, Cai S and Wang P 2024 Nanoscale Horiz. 9 900–933

    2024

  5. [13]

    2025 ACS Applied Materials & Interfaces17 12852–12861

    Xie Y, Zhang R, Xie Y, Huang X, Bustillo K C, Jorgens D M, Abousleiman Y, Ophus C, Hull K L, Minor A M et al. 2025 ACS Applied Materials & Interfaces17 12852–12861

    2025

  6. [14]

    2023 Science advances 9 eadc9721

    Xie Y, Wang J, Savitzky B H, Chen Z, Wang Y, Betzler S, Bustillo K, Persson K, Cui Y, Wang L W et al. 2023 Science advances 9 eadc9721

    2023

  7. [15]

    8 327–339

    Yuan B and Yu Y 2022 Chem. 8 327–339

    2022

  8. [16]

    Microanal.28 2036–2046

    Radi´ c D, Peterlechner M, Posselt M and Bracht H 2022 Microsc. Microanal.28 2036–2046

    2022

  9. [17]

    Microanal.28 1437– 1443

    Quigley F, McBean P, O’Donovan P, Peters J J and Jones L 2022 Microsc. Microanal.28 1437– 1443

    2022

  10. [18]

    Phys.: Mater.4 042005

    Levin B D 2021 J. Phys.: Mater.4 042005

    2021

  11. [19]

    Bustillo K C, Zeltmann S E, Chen M, Donohue J, Ciston J, Ophus C and Minor A M 2021 Acc. Chem. Res. 54 2543–2551 4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion 17

    2021

  12. [20]

    Panova O, Ophus C, Takacs C J, Bustillo K C, Balhorn L, Salleo A, Balsara N and Minor A M 2019 Nature materials 18 860–865

    2019

  13. [21]

    Crystallogr.25 495–501

    Hoppe W 1969 Fundam. Crystallogr.25 495–501

    1969

  14. [22]

    Crystallogr.25 502–507

    Hoppe W and Strube G 1969 Fundam. Crystallogr.25 502–507

    1969

  15. [23]

    Crystallogr.25 508–514

    Hoppe W 1969 Fundam. Crystallogr.25 508–514

    1969

  16. [24]

    2025 Nature Communications 16 261

    Xu X, Xia L, Zheng C, Liu Y, Yu D, Li J, Zhong S, Li C, Song H, Liu Y et al. 2025 Nature Communications 16 261

    2025

  17. [25]

    Maign´ e A and Wolf M 2018Microscopy 67 i86–i97

  18. [26]

    Song J, Allen C S, Gao S, Huang C, Sawada H, Pan X, Warner J, Wang P and Kirkland A I 2019 Sci. Rep. 9 3919

    2019

  19. [27]

    Crystallogr

    Yamada K, Akaishi N, Yatabe K and Takayama Y 2024 Appl. Crystallogr. 57

    2024

  20. [28]

    2018 arXiv preprint arXiv:1804.03999

    Oktay O, Schlemper J, Folgoc L L, Lee M, Heinrich M, Misawa K, Mori K, McDonagh S, Hammerla N Y, Kainz B et al. 2018 arXiv preprint arXiv:1804.03999

    2018 arXiv

  21. [29]

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

    2016

  22. [30]

    Lu L, Shin Y, Su Y and Karniadakis G E 2019 arXiv preprint arXiv:1903.06733

    2019 arXiv

  23. [31]

    Imambi S, Prakash K B and Kanagachidambaresan G 2021 Programming with TensorFlow: solution for edge computing applications87–104

    2021

  24. [32]

    2017 Photo-realistic single image super-resolution using a generative adversarial network Proceedings of the IEEE conference on computer vision and pattern recognitionpp 4681–4690

    Ledig C, Theis L, Husz´ ar F, Caballero J, Cunningham A, Acosta A, Aitken A, Tejani A, Totz J, Wang Z et al. 2017 Photo-realistic single image super-resolution using a generative adversarial network Proceedings of the IEEE conference on computer vision and pattern recognitionp...

    2017

  25. [33]

    Microanal.25 563–582

    Ophus C 2019 Microsc. Microanal.25 563–582

    2019

  26. [34]

    Luo H, Zhang J, Liu X, Zhang L and Liu J 2024 Remote Sens. 16 773

    2024

  27. [35]

    An T, Zhang X, Huo C, Xue B, Wang L and Pan C 2022 IEEE J. Sel. Top. Appl. Earth Obser. Remote Sens. 15 1373–1388

    2022

  28. [36]

    2013 The materials project: a materials genome approach to accelerating materials innovation

    Jain A, Ong S, Hautier G, Chen W, Richards W, Dacek S, Cholia S, Gunter D, Skinner D, Ceder G et al. 2013 The materials project: a materials genome approach to accelerating materials innovation

    2013

  29. [37]

    Berman H M, Westbrook J, Feng Z, Gilliland G, Bhat T N, Weissig H, Shindyalov I N and Bourne P E 2000 Nucleic Acids Res.28 235–242

    2000

  30. [38]

    Madsen J and Susi T 2021 Open Research Europe1 24

    2021

  31. [39]

    2021 Microsc

    Savitzky B H, Zeltmann S E, Hughes L A, Brown H G, Zhao S, Pelz P M, Pekin T C, Barnard E S, Donohue J, DaCosta L R et al. 2021 Microsc. Microanal.27 712–743

    2021

  32. [40]

    Ni X, McGlinchey R P, Jiang J and Lee J C 2019 Journal of molecular biology431 3913–3919

    2019

  33. [41]

    2019 Nature 572 80–85

    Kato H E, Zhang Y, Hu H, Suomivuori C M, Kadji F M N, Aoki J, Krishna Kumar K, Fonseca R, Hilger D, Huang W et al. 2019 Nature 572 80–85

    2019

  34. [42]

    Bai L, Kovach A, You Q, Kenny A and Li H 2019Nature structural & molecular biology26 704–711

  35. [43]

    Zheng H, Sivonxay E, Christensen R, Gallant M, Luo Z, McDermott M, Huck P, Smedskjær M M and Persson K A 2024 npj Computational Materials10 295

    2024

  36. [44]

    Seth A, Kulkarni R P and Sai Gautam G 2025 ACS Materials Au5 458–468

    2025

  37. [45]

    2022 Nat

    Kalinin S V, Ophus C, Voyles P M, Erni R, Kepaptsoglou D, Grillo V, Lupini A R, Oxley M P, Schwenker E, Chan M K et al. 2022 Nat. Rev. Methods Primers2 11

    2022

  38. [46]

    Lee J, Lee M, Park Y, Ophus C and Yang Y 2023 Phys. Rev. Appl.19 054062

    2023

  39. [47]

    2021 Nature 600 279–284

    Wu Y, Han X, Su Y, Glidewell M, Daniels J S, Liu J, Sengupta T, Rey-Suarez I, Fischer R, Patel A et al. 2021 Nature 600 279–284

    2021

  40. [48]

    Clement David-Olawade A, Olawade D B, Vanderbloemen L, Rotifa O B, Fidelis S C, Egbon E, Akpan A O, Adeleke S, Ghose A and Boussios S 2025 Diagnostics 15 689

    2025

  41. [49]

    Microanal.29 395–407 4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion 18

    Friedrich T, Yu C P, Verbeeck J and Van Aert S 2023 Microsc. Microanal.29 395–407 4D-MISR: A unified model for low-dose super-resolution imaging via feature fusion 18

    2023

  42. [50]

    Microanal.22 306–307

    Bosch E G, Lazic I and Lazar S 2016 Microsc. Microanal.22 306–307

    2016

  43. [51]

    Varnavides G, Ribet S M, Zeltmann S E, Yu Y, Savitzky B H, Byrne D O, Allen F I, Dravid V P, Scott M C and Ophus C 2023 arXiv preprint arXiv:2309.05250

    2023 arXiv

  44. [52]

    Rodenburg J M and Faulkner H M 2004 Appl. Phys. Lett.85 4795–4797

    2004

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.