pith. sign in

arxiv: 2508.04728 · v2 · submitted 2025-08-05 · 📡 eess.IV · cs.CV· physics.ins-det

Neural Field-Based 3D Surface Reconstruction of Microstructures from Multi-Detector Signals in Scanning Electron Microscopy

Shuo Chen , Yijin Li , Xi Zheng , Guofeng Zhang This is my paper

Pith reviewed 2026-05-18 23:59 UTC · model grok-4.3

classification 📡 eess.IV cs.CVphysics.ins-det
keywords 3D surface reconstructionneural fieldsscanning electron microscopyphotometric stereomicrostructuresforward modelingmulti-detector signals
0
0 comments X

The pith

A neural field framework reconstructs precise 3D surfaces of microstructures from multi-detector SEM images.

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

This paper presents a new way to build three-dimensional models of microscopic structures using data from scanning electron microscopes. These microscopes normally produce only two-dimensional intensity maps, but the proposed NFH-SEM method combines views from multiple angles and signals from different detectors. It uses a continuous neural field representation along with a model that learns the physical process of electron scattering to correct for common problems like shadows and missing textures. The approach works across many types of samples and can detect very fine details without requiring special calibration for each one. If successful, it provides a practical tool for researchers studying materials at small scales where 3D shape affects function.

Core claim

NFH-SEM reconstructs high-fidelity 3D surfaces from multi-view, multi-detector SEM images by integrating coarse multi-view geometry with photometric stereo cues from detector signals through a continuous neural field, incorporating a learnable forward model that embeds SEM imaging physics for self-calibrated, shadow-robust reconstruction.

What carries the argument

The continuous neural field that fuses multi-view geometry with photometric stereo cues while embedding a learnable forward model of SEM imaging physics to enable self-calibration.

If this is right

  • Precise recovery of nanoscale features including 478 nm layered structures in two-photon lithography samples.
  • Recovery of 782 nm surface textures on pollen grains and 1.559 μm fracture steps on silicon carbide particles.
  • Robust performance on textureless regions and in the presence of shadowing artifacts.
  • Self-calibrated reconstruction that does not depend on specimen-specific ground-truth data.

Where Pith is reading between the lines

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

  • Adapting the framework to other microscopy modalities could broaden 3D characterization capabilities in materials science.
  • Testing on dynamic samples during in-situ experiments might open applications in real-time observation.
  • Scaling the method with larger multi-detector datasets could improve generalization to new material classes.

Load-bearing premise

A single learnable model of how SEM detectors capture signals works accurately for many different microstructures without any specimen-specific adjustments or 3D ground truth during training.

What would settle it

Direct comparison of the reconstructed 3D surfaces against independent measurements obtained from techniques like atomic force microscopy or focused ion beam sectioning on the same specimens.

read the original abstract

The 3D characterization of microstructures is crucial for understanding and designing functional materials. However, the scanning electron microscope (SEM), widely used in scientific research, captures only 2D electron intensity distributions. Existing SEM 3D reconstruction methods struggle with textureless regions, shadowing artifacts, and calibration dependencies, whereas advanced learning-based approaches fail to generalize to microscopic SEM domains due to the lack of physical priors and domain-specific data. We introduce NFH-SEM, a neural field-based hybrid framework that reconstructs high-fidelity 3D surfaces from multi-view, multi-detector SEM images. NFH-SEM integrates coarse multi-view geometry with photometric stereo cues from detector signals through a continuous neural field, incorporating a learnable forward model that embeds SEM imaging physics for self-calibrated, shadow-robust reconstruction. NFH-SEM achieves precise recovery across diverse specimens, revealing 478 nm layered features in two-photon lithography samples, 782 nm surface textures on pollen grains, and 1.559 $\mu$m fracture steps on silicon carbide particles, demonstrating its accuracy and broad applicability. Our code and real-world dataset are available at https://github.com/zju3dv/NFH-SEM.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces NFH-SEM, a neural field-based hybrid framework for reconstructing high-fidelity 3D surfaces from multi-view, multi-detector SEM images. It integrates coarse multi-view geometry with photometric stereo cues from detector signals through a continuous neural field and incorporates a learnable forward model that embeds SEM imaging physics for self-calibrated, shadow-robust reconstruction. The work demonstrates the method on diverse specimens and reports specific recovered feature sizes including 478 nm layered features in two-photon lithography samples, 782 nm surface textures on pollen grains, and 1.559 μm fracture steps on silicon carbide particles, while releasing code and a real-world dataset.

Significance. If the central claims hold, the framework could advance non-destructive 3D microstructure characterization in materials science by handling textureless regions and shadowing without heavy calibration dependencies. The release of code and a real-world dataset is a clear strength that supports reproducibility and community follow-up. However, the absence of quantitative error metrics, independent metrology comparisons, and forward-model ablations limits the strength of the accuracy assertions and reduces immediate applicability for metric-scale claims.

major comments (2)
  1. [Abstract] Abstract: The claims of 'precise recovery' with specific dimensions (478 nm layered features, 782 nm surface textures, 1.559 μm fracture steps) are presented without error bars, standard deviations, or quantitative comparisons to independent baselines such as AFM or FIB tomography. This directly undermines the central claim of sub-micron accuracy across specimens.
  2. [Methods] The joint optimization of the learnable forward model with the neural field surface introduces circularity risk, as model parameters are fitted to the same multi-detector images used for reconstruction. No ablation isolating the contribution of the embedded SEM physics (multi-detector cues and shadow modeling) is described, which is load-bearing for the self-calibration and generalization assertions.
minor comments (1)
  1. [Abstract] The abstract states that the method 'reveals' specific feature sizes but does not clarify whether these are direct measurements from the reconstructed surface or derived quantities, nor how they were extracted.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential of NFH-SEM for non-destructive 3D microstructure characterization. We address each major comment below with clarifications and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claims of 'precise recovery' with specific dimensions (478 nm layered features, 782 nm surface textures, 1.559 μm fracture steps) are presented without error bars, standard deviations, or quantitative comparisons to independent baselines such as AFM or FIB tomography. This directly undermines the central claim of sub-micron accuracy across specimens.

    Authors: We agree that presenting specific feature sizes in the abstract without accompanying error bars or direct quantitative comparisons to independent metrology (e.g., AFM or FIB) weakens the strength of the accuracy claims. These dimensions were extracted by measuring distinct geometric features on the optimized neural field surfaces. In the revised manuscript, we will add uncertainty estimates derived from the optimization residuals and multi-view consistency checks, and we will explicitly qualify that the reported sizes represent observed feature scales rather than validated sub-micron accuracy metrics. We will also expand the discussion of limitations regarding destructive reference methods. revision: yes

  2. Referee: [Methods] The joint optimization of the learnable forward model with the neural field surface introduces circularity risk, as model parameters are fitted to the same multi-detector images used for reconstruction. No ablation isolating the contribution of the embedded SEM physics (multi-detector cues and shadow modeling) is described, which is load-bearing for the self-calibration and generalization assertions.

    Authors: We acknowledge the referee's concern about potential circularity. The forward model parameters are not freely optimized; they are initialized from established SEM physics (detector angles, emission models) and regularized by multi-view geometric consistency losses that are independent of the photometric signals. Nevertheless, we agree that an explicit ablation is needed to isolate the contribution of the embedded physics. In the revision we will add an ablation study that compares the full NFH-SEM pipeline against (i) a version without the learnable forward model and (ii) a version without explicit shadow modeling, reporting quantitative differences in surface fidelity and self-calibration stability. revision: yes

Circularity Check

1 steps flagged

Learnable forward model jointly optimized with surface on same images creates moderate circularity in self-supervised reconstruction claims.

specific steps
  1. fitted input called prediction [Abstract and method overview]
    "NFH-SEM integrates coarse multi-view geometry with photometric stereo cues from detector signals through a continuous neural field, incorporating a learnable forward model that embeds SEM imaging physics for self-calibrated, shadow-robust reconstruction."

    The learnable forward model parameters are fitted to the identical multi-detector SEM images that are also used to optimize the neural field for 3D surface recovery. Consequently, the 'precise recovery' of specific feature dimensions is a direct outcome of this joint fitting process rather than an external prediction or validation.

full rationale

The paper's core method optimizes a neural field surface representation together with a learnable forward model that embeds SEM physics, both fitted directly to the input multi-view multi-detector images for self-calibration. This setup means the reported nanoscale feature sizes (478 nm layers, etc.) are outputs of the same optimization that matches the observed intensities, rather than independent predictions validated against external metrology. However, the multi-view geometry prior and released real dataset provide some independent grounding, preventing full reduction to tautology. No self-citation chains or uniqueness theorems are invoked as load-bearing, and the derivation does not rename known results or smuggle ansatzes via prior work.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that a jointly optimized neural field plus learnable forward model can recover metric 3D geometry from intensity signals without external calibration. No explicit free parameters are named in the abstract, but the learnable forward model necessarily introduces fitted weights. No new physical entities are postulated.

free parameters (1)
  • learnable forward model weights
    Parameters of the module that predicts detector signals from surface geometry and viewing direction; these are optimized on the input images.
axioms (1)
  • domain assumption Multi-view geometry from coarse reconstruction provides a reliable initialization that can be refined by photometric cues.
    Invoked when the method integrates coarse multi-view geometry with photometric stereo inside the neural field.

pith-pipeline@v0.9.0 · 5756 in / 1469 out tokens · 40882 ms · 2026-05-18T23:59:56.064759+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

67 extracted references · 67 canonical work pages · 1 internal anchor

  1. [1]

    Scanning electron microscopy: physics of im- age formation and microanalysis.Measurement Science and Technology11, 1826–1826 (2000)

    Reimer, L. Scanning electron microscopy: physics of im- age formation and microanalysis.Measurement Science and Technology11, 1826–1826 (2000)

    work page 2000

  2. [2]

    Suga, M.et al.Recent progress in scanning electron mi- croscopy for the characterization of fine structural details of nano materials.Progress in Solid State Chemistry42, 1–21 (2014)

    work page 2014

  3. [3]

    Mitragotri, S.et al.Accelerating the translation of nanoma- terials in biomedicine.ACS Nano9, 6644–6654 (2015)

    work page 2015

  4. [4]

    C.et al.Epitaxial growth of three-dimensionally architectured optoelectronic devices.Nature Materials10, 676–681 (2011)

    Nelson, E. C.et al.Epitaxial growth of three-dimensionally architectured optoelectronic devices.Nature Materials10, 676–681 (2011)

    work page 2011

  5. [5]

    & Barthlott, W

    Koch, K., Bhushan, B. & Barthlott, W. Diversity of structure, morphology and wetting of plant surfaces.Soft Matter4, 1943–1963 (2008)

    work page 1943

  6. [6]

    Nature584, 387–392 (2020)

    Zhao, S.et al.Additive manufacturing of silica aerogels. Nature584, 387–392 (2020)

    work page 2020

  7. [7]

    P., Kirkpatrick, A

    Tafti, A. P., Kirkpatrick, A. B., Alavi, Z., Owen, H. A. & Yu, Z. Recent advances in 3D SEM surface reconstruction. Micron78, 54–66 (2015)

    work page 2015

  8. [8]

    D., Job, P

    Ball, A. D., Job, P. A. & Walker, A. E. L. SEM- microphotogrammetry, a new take on an old method for gen- erating high-resolution 3D models from SEM images.Jour- nal of microscopy267, 214–226 (2017)

    work page 2017

  9. [9]

    & Reiss, G

    Eulitz, M. & Reiss, G. 3D reconstruction of SEM images by use of optical photogrammetry software.Journal of struc- tural biology191, 190–196 (2015)

    work page 2015

  10. [10]

    Zhang, X.et al.Superalloys fracture process inference based on overlap analysis of 3D models.Communications Engi- neering3, 108 (2024)

    work page 2024

  11. [11]

    Extracting three-dimensional information from SEM images by means of photogrammetry.Micron 134, 102873 (2020)

    Kozikowski, P. Extracting three-dimensional information from SEM images by means of photogrammetry.Micron 134, 102873 (2020)

    work page 2020

  12. [12]

    Sch ¨onberger, J. L. & Frahm, J.-M. Structure-from-motion revisited. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 4104–4113 (2016)

    work page 2016

  13. [13]

    & Singer, A

    ¨Ozyes ¸il, O., V oroninski, V ., Basri, R. & Singer, A. A sur- vey of structure from motion.Acta Numerica26, 305–364 (2017)

    work page 2017

  14. [14]

    Iglhaut, J.et al.Structure from motion photogrammetry in forestry: A review.Current Forestry Reports5, 155–168 (2019)

    work page 2019

  15. [15]

    & Krysztof, M

    Sł ´owko, W. & Krysztof, M. Detector system for three- dimensional imaging in the variable pressure/environmental SEM.Acta Physica Polonica A123, 877–879 (2013)

    work page 2013

  16. [16]

    & Sł ´owko, W

    Drzazga, W., Paluszynski, J. & Sł ´owko, W. Three- dimensional characterization of microstructures in a SEM. Measurement Science and Technology17, 28 (2005)

    work page 2005

  17. [17]

    & Sł ´owko, W

    Paluszy ´nski, J. & Sł ´owko, W. Surface reconstruction with the photometric method in SEM.Vacuum78, 533–537 (2005)

    work page 2005

  18. [18]

    Giardino, M., Menon, D. M. N. & Janner, D. L. Regular- ization techniques for 3D surface reconstruction from four quadrant backscattered electron detector images.Ultrami- croscopy250, 113746 (2023)

    work page 2023

  19. [19]

    Specific features of the miniature ionisation BSE multi-detector unit for 3D imaging in environmental condi- tions.Micron126, 102752 (2019)

    Sł ´owko, W. Specific features of the miniature ionisation BSE multi-detector unit for 3D imaging in environmental condi- tions.Micron126, 102752 (2019)

    work page 2019

  20. [20]

    Neggers, J.et al.Principal image decomposition for multi- detector backscatter electron topography reconstruction.Ul- tramicroscopy227, 113200 (2021)

    work page 2021

  21. [21]

    & Vanzi, M

    Pintus, R., Podda, S. & Vanzi, M. An automatic alignment procedure for a four-source photometric stereo technique ap- plied to scanning electron microscopy.IEEE Transactions on Instrumentation and Measurement57, 989–996 (2008)

    work page 2008

  22. [22]

    Woodham, R. J. Photometric method for determining surface orientation from multiple images.Optical engineering19, 139–144 (1980)

    work page 1980

  23. [23]

    & Goesele, M

    Ackermann, J. & Goesele, M. A survey of photometric stereo techniques.Found. Trends. Comput. Graph. Vis.9, 149–254 (2015)

    work page 2015

  24. [24]

    Shi, B.et al.A benchmark dataset and evaluation for non- lambertian and uncalibrated photometric stereo.IEEE Trans- actions on Pattern Analysis and Machine Intelligence41, 271–284 (2019)

    work page 2019

  25. [25]

    & Kibbey, T

    Yan, S., Adegbule, A. & Kibbey, T. C. G. A hybrid 3D SEM reconstruction method optimized for complex geologic ma- terial surfaces.Micron99, 26–31 (2017)

    work page 2017

  26. [26]

    & Scherer, S

    Danzl, R. & Scherer, S. Integrating shape from shading and shape from stereo for variable reflectance surface reconstruc- tion from SEM images. InProceedings 26th Workshop of the Austrian Association for Pattern Recognition (AAPR), 281– 288 (2002)

    work page 2002

  27. [27]

    Computer Graphics Forum41, 641–676 (2022)

    Xie, Y .et al.Neural fields in visual computing and beyond. Computer Graphics Forum41, 641–676 (2022)

    work page 2022

  28. [28]

    Wang, P.et al.NeuS: Learning neural implicit surfaces by volume rendering for multi-view reconstruction.Advances in Neural Information Processing Systems34, 27171–27183 (2021)

    work page 2021

  29. [29]

    & Lipman, Y

    Yariv, L., Gu, J., Kasten, Y . & Lipman, Y . V olume rendering of neural implicit surfaces.Advances in Neural Information Processing Systems34, 4805–4815 (2021)

    work page 2021

  30. [30]

    & Geiger, A

    Oechsle, M., Peng, S. & Geiger, A. UNISURF: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction. In2021 IEEE/CVF International Conference on Computer Vision (ICCV), 5569–5579 (2021)

    work page 2021

  31. [31]

    Mildenhall, B.et al.NeRF: Representing scenes as neural radiance fields for view synthesis.Communications of the ACM65, 99–106 (2021)

    work page 2021

  32. [32]

    T., Mildenhall, B., Verbin, D., Srinivasan, P

    Barron, J. T., Mildenhall, B., Verbin, D., Srinivasan, P. P. & Hedman, P. Mip-NeRF 360: Unbounded anti-aliased neural radiance fields. In2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 5460–5469 (2022). 13

    work page 2022

  33. [33]

    & Kanazawa, A

    Yu, A., Ye, V ., Tancik, M. & Kanazawa, A. pixelNeRF: Neural radiance fields from one or few images. In2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 4576–4585 (2021)

    work page 2021

  34. [34]

    & Heidrich, W

    Zang, G., Idoughi, R., Li, R., Wonka, P. & Heidrich, W. Intratomo: Self-supervised learning-based tomography via sinogram synthesis and prediction. In2021 IEEE/CVF In- ternational Conference on Computer Vision (ICCV), 1940– 1950 (2021)

    work page 1940

  35. [35]

    In2023 IEEE/CVF International Conference on Computer Vision (ICCV), 1–11 (2023)

    Liu, X.et al.Multi-modal neural radiance field for monoc- ular dense SLAM with a light-weight ToF sensor. In2023 IEEE/CVF International Conference on Computer Vision (ICCV), 1–11 (2023)

    work page 2023

  36. [36]

    & Xing, L

    Shen, L., Pauly, J. & Xing, L. NeRP: Implicit neural repre- sentation learning with prior embedding for sparsely sam- pled image reconstruction.IEEE Transactions on Neural Networks and Learning Systems35, 770–782 (2024)

    work page 2024

  37. [37]

    & Kamilov, U

    Liu, R., Sun, Y ., Zhu, J., Tian, L. & Kamilov, U. S. Re- covery of continuous 3D refractive index maps from discrete intensity-only measurements using neural fields.Nature Ma- chine Intelligence4, 781–791 (2022)

    work page 2022

  38. [38]

    Kang, I., Zhang, Q., Yu, S. X. & Ji, N. Coordinate-based neural representations for computational adaptive optics in widefield microscopy.Nature Machine Intelligence6, 714– 725 (2024)

    work page 2024

  39. [39]

    Chen, S.et al.Multi-view neural 3D reconstruction of micro- and nanostructures with atomic force microscopy.Commu- nications Engineering3, 131 (2024)

    work page 2024

  40. [40]

    10388, http://arxiv.org/abs/2501.10388, arXiv:2501.10388 [cs]

    Chen, S., Li, Y . & Zhang, G. OpticFusion: Multi-modal neu- ral implicit 3D reconstruction of microstructures by fusing white light interferometry and optical microscopy. Preprint athttps://doi.org/10.48550/arXiv.2501. 09259(2025)

    work page doi:10.48550/arxiv.2501 2025

  41. [41]

    & Lipman, Y

    Gropp, A., Yariv, L., Haim, N., Atzmon, M. & Lipman, Y . Implicit geometric regularization for learning shapes. InPro- ceedings of the 37th International Conference on Machine Learning, 3789–3799 (2020)

    work page 2020

  42. [42]

    & Keller, A

    M ¨uller, T., Evans, A., Schied, C. & Keller, A. Instant neu- ral graphics primitives with a multiresolution hash encoding. ACM Transactions on Graphics (ToG)41, 1–15 (2022)

    work page 2022

  43. [43]

    Lorensen, W. E. & Cline, H. E. Marching cubes: A high resolution 3D surface construction algorithm. InSeminal graphics: pioneering efforts that shaped the field, 347–353 (1998)

    work page 1998

  44. [44]

    & O’Leary, P

    Harker, M. & O’Leary, P. Least squares surface reconstruc- tion from measured gradient fields. In2008 IEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR), 1–7 (2008)

    work page 2008

  45. [45]

    & O’Leary, P

    Harker, M. & O’Leary, P. Direct regularized surface recon- struction from gradients for industrial photometric stereo. Computers in industry64, 1221–1228 (2013)

    work page 2013

  46. [46]

    & O’Leary, P

    Harker, M. & O’Leary, P. Regularized reconstruction of a surface from its measured gradient field algorithms for spectral, tikhonov, constrained, and weighted regulariza- tion.Journal of Mathematical Imaging and Vision51, 46–70 (2015)

    work page 2015

  47. [47]

    & Shin, Y

    Harinarayana, V . & Shin, Y . Two-photon lithography for three-dimensional fabrication in micro/nanoscale regime: A comprehensive review.Optics & Laser Technology142, 107180 (2021)

    work page 2021

  48. [48]

    & Takada, K

    Kawata, S., Sun, H., Tanaka, T. & Takada, K. Finer features for functional microdevices - micromachines can be created with higher resolution using two-photon absorption.Nature 412, 697–698 (2001)

    work page 2001

  49. [49]

    Sun, H.et al.Real three-dimensional microstructures fabri- cated by photopolymerization of resins through two-photon absorption.Optics letters25, 1110–1112 (2000)

    work page 2000

  50. [50]

    Wang, H.et al.Two-photon polymerization lithography for optics and photonics: fundamentals, materials, technolo- gies, and applications.Advanced Functional Materials33, 2214211 (2023)

    work page 2023

  51. [51]

    & Park, S

    Lee, H., Kim, D.-i., Kwon, S.-h. & Park, S. Magneti- cally actuated drug delivery helical microrobot with mag- netic nanoparticle retrieval ability.ACS applied materials & interfaces13, 19633–19647 (2021)

    work page 2021

  52. [52]

    Marino, A.et al.Two-photon lithography of 3D nanocom- posite piezoelectric scaffolds for cell stimulation.ACS ap- plied materials & interfaces7, 25574–25579 (2015)

    work page 2015

  53. [53]

    & Punyasena, S

    Sivaguru, M., Mander, L., Fried, G. & Punyasena, S. W. Cap- turing the surface texture and shape of pollen: a comparison of microscopy techniques.Plos one7, e39129 (2012)

    work page 2012

  54. [54]

    & Preuss, D

    Edlund, A., Swanson, R. & Preuss, D. Pollen and stigma structure and function: The role of diversity in pollination. Plant Cell16, S84–S97 (2004)

    work page 2004

  55. [55]

    Fan, T.-F.et al.Transformation of hard pollen into soft mat- ter.Nature Communications11, 1449 (2020)

    work page 2020

  56. [56]

    Zhang, H.et al.Managed bumblebees outperform honey- bees in increasing peach fruit set in China: different limiting processes with different pollinators.Plos one10, e0121143 (2015)

    work page 2015

  57. [57]

    & Meredith, J

    Lin, H., Gomez, I. & Meredith, J. C. Pollenkitt wetting mechanism enables species-specific tunable pollen adhesion. Langmuir29, 3012–3023 (2013)

    work page 2013

  58. [58]

    & Misirli, A

    Evrenoso ˘glu, Y . & Misirli, A. Investigations on the pollen morphology of some fruit species.Turkish Journal of Agri- culture and Forestry33, 181–190 (2009)

    work page 2009

  59. [59]

    Materials Today Communications28, 102533 (2021)

    Xu, M.et al.Recent advances and challenges in silicon car- bide (SiC) ceramic nanoarchitectures and their applications. Materials Today Communications28, 102533 (2021)

    work page 2021

  60. [60]

    & Zhang, X

    Wang, Y ., Dong, S., Li, X., Hong, C. & Zhang, X. Synthe- sis, properties, and multifarious applications of SiC nanopar- ticles: A review.Ceramics International48, 8882–8913 (2022)

    work page 2022

  61. [61]

    Pilchak, A. L. Fatigue crack growth rates in alpha titanium: Faceted vs. striation growth.Scripta Materialia68, 277–280 (2013). 14

    work page 2013

  62. [62]

    Murashkin, M.et al.Fatigue behavior of an ultrafine-grained Al-Mg-Si alloy processed by high-pressure torsion.Metals 5, 578–590 (2015)

    work page 2015

  63. [63]

    Birenis, D.et al.Interpretation of hydrogen-assisted fatigue crack propagation in BCC iron based on dislocation structure evolution around the crack wake.Acta Materialia156, 245– 253 (2018)

    work page 2018

  64. [64]

    Paszke, A.et al.PyTorch: An imperative style, high- performance deep learning library.Advances in neural in- formation processing systems32, 8024–8035 (2019)

    work page 2019

  65. [65]

    Instant neural surface reconstruction.Github

    Guo, Y .-C. Instant neural surface reconstruction.Github. https://github.com/bennyguo/instant-nsr- pl(2022)

    work page 2022

  66. [66]

    & Bonnet, S

    Abdallah, Z., Fay, A. & Bonnet, S. 3D SEM metrology of microstructures for high volume manufacturing. In38th European Mask and Lithography Conference (EMLC 2023), vol. 12802, 128020N (SPIE, 2023)

    work page 2023

  67. [67]

    Kingma, D. P. & Ba, J. Adam: A method for stochastic op- timization. Preprint athttps://doi.org/10.48550/ arXiv.1412.6980(2017). Acknowledgements We sincerely thank A. Ren and L. Ma for their assistance with the TPL experiments. We gratefully acknowledge the Edward J. Team for their support of the Wukong 3D model. We also thank N. Rong, J. Niu, and D. Qi ...

    work page internal anchor Pith review Pith/arXiv arXiv 2017