Pith. sign in

REVIEW 4 major objections 6 minor 49 references

NanoMorph-3D: An End-to-End Physics-Driven Unrolling Framework for Nanomaterial Reconstruction

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read NanoMorph-3D claims that unrolling proximal gradient descent with trajectory-aware attention reconstructs 3D nanomaterials from ±60° tilt series with higher fidelity than six baselines, completing the missing wedge and transferring from sim

desk verdict A well-engineered unrolling framework with strong synthetic results, but the sim-to-real transfer claim is unsupported because the real-data evaluation uses training data and the same loss as the training objective. read the letter →

arxiv 2608.03257 v1 pith:4IGXRCEQ submitted 2026-08-04 cs.CV

Beiyuan Zhang , Hesong Li , Ziqi Wu , Ruiwen Shao , Ying Fu This is my paper
classification cs.CV
keywords electrontomographymissingwedgedeepunrollingproximalgradientdescentHAADF-STEMnanomaterialssinogramconsistencyphysics-drivenreconstruction
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

NanoMorph-3D aims to fix the missing-wedge problem in electron tomography of nanomaterials, where hardware limits tilt angles to roughly ±60° and the unsampled wedge of Fourier space produces elongation and topological blur. The paper proposes an end-to-end network that unrolls proximal gradient descent into ten learnable stages, each alternating a projection-consistency gradient step (back-projecting the sinogram residual) with a deep topology-aware proximal map. A Dual-Domain Sinusoidal Attention injects the geometry of Radon projection trajectories into the attention logits, and a Physics-Normalization layer makes features scale-invariant across magnifications and densities. Trained on a large synthetic dataset built from a Nanomorphological Taxonomy with Beer-Lambert nonlinear attenuation, and adapted to real data by an unsupervised dual-stream reprojection loss, the framework reports the highest Fourier shell correlation (FSC, 0.6125) and lowest LPIPS among SIRT, GENFIRE, RESIRE, AET-Net, LPD, and Denza-GS, while completing missing-wedge sinograms and keeping fidelity down to 11 views.

What carries the argument

The key machinery is the unrolled proximal gradient step with the Projection Consistency Unit (PCU), $Z^{(k)} = \mathrm{ReLU}\left(V^{(k-1)} - \frac{\alpha^{(k)}}{M\cdot N} B_\Theta\left(F_\Theta(V^{(k-1)}) - Y\right)\right)$, followed by a deep Topology-Aware Proximal Mapping $H_\theta$; the Dual-Domain Sinusoidal Attention adds the trajectory-compatibility bias $M_{\text{geo}}$ into the attention logits, and Physics-Normalization (density-aware RMS norm) provides scale invariance. The Beer-Lambert-linearized forward model $Y = -\frac{1}{\mu s}\ln\left(1 - \frac{I}{I_0}\right) = A_\Theta V + \eta$ (Eq. 1) grounds both the simulator and the PCU, so the network's data-consistency path is the

What would settle it

Reconstruct a nanomaterial from a dual-axis or full-tilt HAADF-STEM acquisition to obtain a ground-truth volume, then feed only the ±60° subset to NanoMorph-3D and compare its reconstruction and completed sinogram with the ground truth and the actually measured omitted views; if Fourier shell correlation in the wedge region and the withheld-view reprojection error are not clearly better than those of the best iterative baseline, the claimed resolution of missing-wedge hallucinations is not substantiated on real data.

Watch

Extended reading notes

Core claim

The central discovery is that unrolling proximal gradient descent into a learnable architecture with an explicit geometric bias removes missing-wedge hallucinations. Each of the ten unrolled stages performs a gradient step in the projection domain—back-projecting the residual between the forward projection of the current volume and the measured sinogram—followed by a deep proximal mapping that restores long-range 3D topology via hierarchical attention with Physics-Normalization. The geometric bias is a pairwise trajectory-compatibility prior $M_{\text{geo}}$ computed from the sinusoidal locus $u(\theta) = x\cos\theta + z\sin\theta$ of each voxel under the Radon transform; it is added to the

Load-bearing premise

The whole pipeline rests on Eq. (1): that after Beer-Lambert linearization, HAADF-STEM projections equal a discrete Radon transform of the density field plus noise, and that the same discretized operator appears in both the synthetic training simulator and the network's Projection Consistency Unit; if real beam-sample interaction deviates from this model—via multiple scattering, detector nonlinearity, or misalignment—the physics constraints and the test set are biased in the

Editorial extensions

If this is right

  • Limited-angle HAADF-STEM tilt series can be reconstructed with a single forward pass of a 10-stage network, offering faster inference than iterative solvers and per-scene optimization used by neural rendering methods.
  • The dynamic-view-dropout curriculum lets the model complete withheld projection angles, enabling dose-friendly sparse-view acquisition down to 11 views while retaining SSIM 0.888 and FSC 0.495, both higher than full-41-view GENFIRE.
  • Because the unsupervised dual-stream loss uses unlabeled real tilt series as pseudo-targets via reprojection consistency, the method can adapt to lab-specific noise and vacuum artifacts without paired ground truth.
  • The taxonomy-driven synthetic dataset with Beer-Lambert nonlinear attenuation provides a reusable physics-grounded benchmark for evaluating generalizable nanomaterial reconstruction.
  • Physics-Normalization makes the learned priors invariant to magnification and mass-density scale, so a single trained model is intended to transfer across microscope settings.

Reading between the lines

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

  • A direct extension the paper leaves implicit is that the same recipe—unrolled proximal steps plus a sinusoidal trajectory bias injected into attention—should transfer to any limited-angle Radon-type inverse problem (sparse-view X-ray CT, cryo-electron tomography, ground-penetrating radar), since the geometry prior is operator-specific but not microscope-specific.
  • The ablation results (FSC drops from 0.6125 to 0.354 when the projection-consistency gradient step is removed) suggest that many black-box 3D denoisers could be upgraded simply by alternating with a differentiable forward/back-projector, a change testable on existing networks without redesigning them.
  • A strong test of the physics-driven claim is to mask a real dual-axis or full-tilt dataset and compare the network's completed sinogram in the omitted wedge with the actually measured projections; if they disagree systematically, the sinogram consistency is an artifact of the shared simulator rather than the real forward model.
  • The reliance on Beer-Lambert implies a prediction: on real data with significant multiple scattering or detector nonlinearities, the method's margin over the best iterative baseline should shrink, and a version trained with a physics-perturbed simulator should show a corresponding drop.
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

4 major / 6 minor

Summary. NanoMorph-3D proposes an end-to-end physics-driven unrolling network for limited-angle HAADF-STEM electron tomography. The forward model (Eq. 1) is a Beer-Lambert-linearized discrete Radon transform with additive noise. The network unrolls proximal gradient descent into ten learnable stages, alternating a Projection Consistency Unit (Eq. 3) with a Topology-Aware Proximal Mapping built from hierarchical volumetric attention, dual-domain sinusoidal attention, and physics normalization. A taxonomy-driven synthetic dataset is generated with nonlinear attenuation and instrumental noise, and an unsupervised dual-stream loss (Eq. 7) is applied to real tilt series. Experiments compare against SIRT, GENFIRE, RESIRE, AET-Net, LPD, and Denza-GS, reporting synthetic metrics in Tables 1, 3, and 4 and real-data reprojection metrics in Table 2.

Significance. If the claims were supported, the paper would make a useful contribution: it combines algorithmic unrolling with explicit projection-geometry attention, and it contributes a large synthetic dataset grounded in a morphological taxonomy. The architecture is a reasonable and timely synthesis of learned priors and physics constraints, and the sparse-view robustness results (Table 3) show a plausible direction for low-dose electron tomography. The main weakness is validation: the real-data evaluation is circular with the training objective, and the synthetic evaluation uses the same forward model embedded in the network, so the simulation-to-reality transfer claim is currently not demonstrated. The paper also contains no code or dataset release despite describing a substantial new dataset.

major comments (4)
  1. [Sec. 3.6, Sec. 4.1, Table 2] The real-data evaluation is not a generalization test. Section 4.1 states that the seven real series used in the experiments are the same seven series used as unlabeled data in the unsupervised stream. The Table 2 metric (reprojection PSNR/SSIM/LPIPS over measured tilts) is essentially the L_unsup objective of Eq. (7), which is minimized during training. Thus Table 2 measures fit to the training set, not simulation-to-reality transfer. The Sec. 4.2 claim of 'successful simulation-to-reality transfer' is unsupported. Please evaluate on held-out real series or clearly split the seven series into train/test subsets, and report per-series results. Using withheld tilt views as pseudo-ground truth does not escape this problem when the same views are used in the unsupervised objective.
  2. [Sec. 3.5, Sec. 4.1, Eq. (1), Eq. (3)] The synthetic test set is generated with the same discrete Radon operator A_Θ (after Beer-Lambert linearization) that is embedded in the network's Projection Consistency Unit (Eq. 3) and in the unsupervised loss (Eq. 7). Consequently, Tables 1, 3, and 4 validate inversion under an assumed model, not generalization to the real HAADF-STEM measurement process. The paper itself acknowledges in Sec. 5 a 'reliance on the Beer-Lambert law.' To support the central claim, add an independent validation: generate test data with a different ray-tracer or discretization, include misalignment or calibration errors, or use real volumes with independent ground truth (e.g., FIB-SEM correlated volumes or simulated inclusions embedded in experimental backgrounds).
  3. [Sec. 4.2, Table 2] The real-data comparison reports single aggregate numbers over only seven series, without per-series breakdown, confidence intervals, or statistical significance. Several baselines (LPD, AET-Net) achieve PSNR around 15 dB, which is far below the others and suggests possible implementation or hyperparameter issues for those baselines. Since the real-data claim is load-bearing, please report per-series errors, standard deviations, and the exact hyperparameters used for each baseline (iterations, regularizers, learning rates).
  4. [Sec. 3.3, Eq. (3)] The unrolled PGD formulation is underspecified. Eq. (3) uses operators F_Θ and B_Θ, while Eq. (1) defines A_Θ, and Eq. (7) again uses A_Θ. The paper does not state whether F_Θ = A_Θ, whether B_Θ is the exact adjoint or an approximation, or how the gradient step relates to the PGD step for the objective in Eq. (2). Since the method's name and motivation rest on PGD unrolling, please clarify these operators and provide a short derivation of the update rule.
minor comments (6)
  1. [Sec. 3.4, Eq. (5)] The symbol λ is used both as a learnable scalar in Eq. (5) and as the loss-balancing weight in Eq. (7) and Sec. 4.1. Please use distinct symbols to avoid ambiguity.
  2. [Sec. 3.5, Eq. (1)] The text emphasizes 'non-linear exponential attenuation' and the Beer-Lambert law, but Eq. (1) is already in the linear logarithmic domain. Clarify how the nonlinear attenuation is simulated and how domain randomization of μ is applied in the synthetic pipeline.
  3. [Sec. 4.1, Compared Methods] Baseline configurations are not reported. For SIRT, GENFIRE, RESIRE, LPD, and Denza-GS, specify the number of iterations, regularization parameters, and any preprocessing steps. This is necessary for a fair comparison.
  4. [Fig. 2, Sec. 3.3] The architecture description lacks key details: the number of attention heads, window sizes, feature dimensions, transformer depth, and the definition of the 'gradient-free block' used for initialization. These details are needed for reproducibility.
  5. [References [42]] Denza-GS [42] is a self-cited arXiv preprint. Please state its publication status and clearly explain its relationship to the current method, especially since it is used as a strong baseline.
  6. [Sec. 4.1, Data Availability] The paper states that a large-scale dataset is constructed and provided for the community, but no code or data release link is given. Please add an availability statement.

Circularity Check

2 steps flagged · score 6.0 of 10

Real-data 'sim-to-real transfer' claim is a training-fit report: Table 2 measures reprojection consistency on the same seven tilt series used to minimize Eq. (7), and synthetic benchmarks use the same forward operator embedded in the network.

  1. fitted input called prediction [Sec. 3.6 (Eq. 7), Sec. 4.1, Sec. 4.2, Table 2]
    "To bridge the domain gap, we assembled seven real-world HAADF-STEM tilt series: three from a public nanomaterial database [18] and four acquired in-house. The real series are used as unlabeled experimental data in the unsupervised domain-adaptation stream during training. ... We split the measured angles into two disjoint subsets Θin and Θout, and reconstruct Vreal=Net(YΘin) from the input subset alone. We then enforce a held-out reprojection loss on the withheld angles: Lunsup = ∥AΘout[Vreal]−YΘout∥1 (7) ... Table 2 evaluates generalization on experimental HAADF-STEM tilt series via 2D reproj"

    The seven real series evaluated in Table 2 are exactly the seven series on which the unsupervised stream was trained (Sec. 4.1), and Table 2's metric—2D reprojection consistency over measured tilts—is essentially the same objective as Eq. (7), which the network minimized on those same series. Thus the 'successful simulation-to-reality transfer' claim in Sec. 4.2 reports how well the model fits its own training data, not transfer to unseen experimental data. No held-out real series or train/test split is provided.

  2. other [Sec. 3.3 Eq. (3), Sec. 3.5, Tables 1/3/4]
    "To adhere to the forward physical model defined in Sec. 3.2, we simulate the HAADF-STEM acquisition process over the generated volumes. Rather than relying on linear projections, we map the voxel grids to a physical scale of 100 nm and enforce a non-linear exponential attenuation governed by the Beer-Lambert law. ... where FΘ represents the physical forward projection operator governing beam attenuation."

    The synthetic test volumes in Sec. 3.5 are projected with the same Beer-Lambert/Radon forward operator that the network's Projection Consistency Unit (Eq. 3) enforces as a data-consistency constraint. Tables 1, 3, and 4 therefore evaluate the network on test data generated by the very operator embedded in the network. This does not externally validate the physical model or the missing-wedge completion; it tests only whether the unrolled network can invert the assumed operator, so part of the claimed 'physics-driven' superiority is ensured by construction.

full rationale

The methodological core—unrolling proximal gradient descent into a learnable architecture, the dual-domain attention, and the physics normalization—is a self-contained engineering contribution and is not circular in its derivation. The circularity is concentrated in the evaluation that supports the headline claims. First, the real-data 'simulation-to-reality transfer' claim (Sec. 4.2) is supported by Table 2, which reports reprojection PSNR/SSIM/LPIPS on the same seven HAADF-STEM tilt series used as unlabeled training data in Sec. 3.6, with a metric essentially identical to the Lunsup of Eq. (7). That is training-set fit, not generalization to new experimental data. Second, the synthetic benchmark (Tables 1, 3, 4) uses volumes generated by the same forward operator FΘ that the Projection Consistency Unit (Eq. 3) enforces, so it cannot independently validate the forward-physics assumption; it only tests inversion of the assumed operator. The paper's own conclusion acknowledges reliance on the Beer-Lambert law. Self-citations (e.g., Denza-GS [42]) appear only as comparison baselines and are not load-bearing for the derivation. Because the central transfer claim reduces to a training objective on the evaluation set, the paper earns a partial circularity score of 6.

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

The central claim rests on five assumptions and five free parameters. The most consequential is the identity between the physical forward model used for simulation and the one baked into the network layers, which makes the synthetic accuracy numbers partially self-consistent by construction. The remaining assumptions concern the realism of the synthetic taxonomy and the transferability of the unsupervised loss. No new physical entities are proposed; the taxonomy, sinusoidal attention, and normalization are algorithmic constructs.

free parameters (5)
  • per-stage step size α^(k) (k=1..10) = not reported
    Eq. 3: each unrolled stage has a learnable step size; trained values are not disclosed, so the physics gradient step magnitude is a fitted parameter.
  • loss balancing weight λ = 0.1
    Sec. 4.1: 'the loss balancing weight λ set to 0.1'; no sensitivity analysis provided.
  • number of unrolled stages K = 10
    Sec. 4.1: PGD unrolled for K=10 stages; ablation over K not reported.
  • material attenuation coefficient μ = randomized, range not given
    Sec. 3.5: 'domain randomization to the material attenuation coefficient μ'; the chosen range controls the simulator and the forward model, and it is not reported.
  • density fluctuation amplitude in synthetic volumes = ±20%
    Sec. 3.5: internal density variations constrained to ±20% amplitude; a data-generation choice that shapes the synthetic topology distribution.
assumptions (5)
  • domain assumption HAADF-STEM intensity at mesoscale follows Beer-Lambert exponential attenuation and, after logarithmic linearization, the projection is Y = A_Θ V + η (Eq. 1).
    Sec. 3.2, Eq. (1). The entire physics-driven unrolling and the synthetic data generation depend on this. The authors list reliance on Beer-Lambert as a limitation in Sec. 5.
  • domain assumption The discrete Radon transform A_Θ and its adjoint B_Θ are accurate, aligned discretizations of the true projection geometry, and the same discretization is used in the simulator and in the network's PCU.
    Sec. 3.3 Eq. (3) and Sec. 3.5. If the network's forward operator differs from the real acquisition geometry, the data consistency constraints are mismatched; if it matches the simulator but not reality, synthetic results are optimistic.
  • domain assumption The procedurally generated taxonomy volumes (dense, layered, hollow, 1D, porous, with grain boundaries and defects) are representative of real nanomaterial morphology.
    Sec. 3.5. The claim of generalizable topological recovery depends on this representativeness.
  • domain assumption The unsupervised held-out reprojection loss (Eq. 7) transfers to real data and does not merely overfit the training views.
    Sec. 3.6. The sim-to-real claim rests on this premise; it is not validated on truly held-out experimental volumes with 3D ground truth.
  • ad hoc to paper Unrolling PGD into a learnable network with a learned proximal operator retains the convergence and data-consistency properties of PGD.
    Sec. 3.3. No convergence analysis is given for the learned proximal operator; the paper relies on the standard PGD framework implicitly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NanoMorph-3D: An End-to-End Physics-Driven Unrolling Framework for Nanomaterial Reconstruction." pith.science (2026). https://pith.science/paper/4IGXRCEQ

@misc{pith2026260803257,
  author       = {Pith},
  title        = {Pith review of: NanoMorph-3D: An End-to-End Physics-Driven Unrolling Framework for Nanomaterial Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4IGXRCEQ}},
  note         = {Machine review of arXiv:2608.03257}
}
read the original abstract

Precise 3D characterization of nanomaterials is essential for unlocking structure-property relationships. However, standard electron tomography is fundamentally limited by the missing wedge problem. Consequently, conventional algorithms suffer from severe geometric distortions, a challenge further complicated by pervasive noise interference. Current learning-based methods either rely on physics-blind post-processing or employ end-to-end architectures constrained by local receptive fields, failing to capture complex 3D topologies. We propose NanoMorph-3D, a unified end-to-end framework grounded in a comprehensive Nanomorphological Taxonomy. Powered by a large-scale synthetic dataset explicitly modeling non-linear electron attenuation, we design a Physics-Driven Unrolled Network mapping proximal gradient descent into a learnable architecture. To capture complex internal topologies, we formulate a hierarchical attention mechanism with Physics-Normalization for long-range 3D dependencies and scale invariance. Crucially, our Dual-Domain strategy leverages Sinusoidal Attention to explicitly model physical projection trajectories, enforcing strict sinogram consistency to mitigate missing wedge artifacts. Finally, an unsupervised dual-stream mechanism bridges the simulation-to-reality gap. Experiments demonstrate NanoMorph-3D reconstructs diverse topologies with superior fidelity and speed.

Figures

Figures reproduced from arXiv: 2608.03257 by the authors.

Figure 1
Figure 1. Overview of NanoMorph-3D. (a) Missing Wedge [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Architecture of NanoMorph-3D. Synthetic and real HAADF-STEM tilt series are processed by a dual-stream pipeline [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Dual-Domain Sinusoidal Attention. We bridge spa [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative evaluation on the synthetic dataset. Reconstructions of four morphological archetypes are shown with [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative evaluation on real experimental data. Reconstructed volumes are shown on top and their sinograms [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 38 canonical work pages

  1. [1]

    Jonas Adler and Ozan Öktem. 2018. Learned primal-dual reconstruction.IEEE transactions on medical imaging37, 6 (2018), 1322–1332

    work page 2018

  2. [2]

    Anders H Andersen and Avinash C Kak. 1984. Simultaneous algebraic recon- struction technique (SART): a superior implementation of the ART algorithm. Ultrasonic imaging6, 1 (1984), 81–94

    work page 1984

  3. [3]

    Zhile Chen, Yuhui Quan, and Hui Ji. 2024. Unsupervised deep unrolling networks for phase unwrapping. InProceedings of the ieee/cvf conference on computer vision and pattern recognition. 25182–25192

    work page 2024

  4. [4]

    Özgün Çiçek, Ahmed Abdulkadir, Soeren S Lienkamp, Thomas Brox, and Olaf Ronneberger. 2016. 3D U-Net: learning dense volumetric segmentation from sparse annotation. InInternational conference on medical image computing and computer-assisted intervention. Springer, 424–432

    work page 2016

  5. [5]

    Moustafa A Darwish, Walaa Abd-Elaziem, Ammar Elsheikh, and Abdelhameed A Zayed. 2024. Advancements in nanomaterials for nanosensors: a comprehensive review.Nanoscale Advances6, 16 (2024), 4015–4046

    work page 2024

  6. [6]

    Peter Ercius, Osama Alaidi, Matthew J Rames, and Gang Ren. 2015. Electron to- mography: a three-dimensional analytic tool for hard and soft materials research. Advanced materials27, 38 (2015), 5638–5663

    work page 2015

  7. [7]

    Kyle Gao, Yina Gao, Hongjie He, Dening Lu, Linlin Xu, and Jonathan Li. 2022. Nerf: Neural radiance field in 3d vision, a comprehensive review.arXiv preprint arXiv:2210.00379(2022)

    arXiv 2022

  8. [8]

    Mauro Gemmi, Enrico Mugnaioli, Tatiana E Gorelik, Ute Kolb, Lukas Palatinus, Philippe Boullay, Sven Hovmöller, and Jan Pieter Abrahams. 2019. 3D electron diffraction: the nanocrystallography revolution.ACS central science5, 8 (2019), 1315–1329

    work page 2019

Show all 49 references
  1. [9]

    Masoud Ghasemi, Michael Geitner, Agatha O’Connell, and Enrique D Gomez

  2. [10]

    Lee W Goldman. 2007. Principles of CT and CT technology.Journal of nuclear medicine technology35, 3 (2007), 115–128

    2007

  3. [11]

    Kuang Gong, Dufan Wu, Kyungsang Kim, Jaewon Yang, Tao Sun, Georges El Fakhri, Youngho Seo, and Quanzheng Li. 2019. MAPEM-Net: an unrolled neural network for Fully 3D PET image reconstruction. In15th International meeting on fully three-dimensional image reconstruction in radio...

    2019

  4. [12]

    Joost Batenburg, H

    Bart Goris, Wouter Van den Broek, K. Joost Batenburg, H. H. Mezerji, and Sara Bals. 2011. Electron tomography based on a total variation minimization recon- struction technique.Ultramicroscopy113 (2011), 120–130. doi:10.1016/j.ultramic. 2011.11.004

    2011 doi

  5. [13]

    Zhanli Hu and Hairong Zheng. 2014. Improved total variation minimization method for few-view computed tomography image reconstruction.BioMedical Engineering OnLine13, 1 (2014), 70

    2014

  6. [14]

    Marzieh Jahanian, Seyede Hosseini, Zahra Dehkordi, Khadije Sadeghi, Sharareh Kalhori, Seyed Ayyoubzadeh, and Mahnaz Ahmadi. 2024. Machine learning approaches for recognition and classification of nanomaterial morphology.Ma- terials Today Communications41 (2024), 110818–110818....

    2024

  7. [15]

    2005.Digital image processing

    Bernd Jähne. 2005.Digital image processing. Springer

    2005

  8. [16]

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George Drettakis, et al

  9. [17]

    Juhyeok Lee, Chaehwa Jeong, and Yongsoo Yang. 2021. Single-atom level deter- mination of 3-dimensional surface atomic structure via neural network-assisted atomic electron tomography.Nature communications12, 1 (2021), 1962

    2021

  10. [18]

    Barnaby DA Levin, Elliot Padgett, Chien-Chun Chen, MC Scott, Rui Xu, Wolf- gang Theis, Yi Jiang, Yongsoo Yang, Colin Ophus, Haitao Zhang, et al . 2016. Nanomaterial datasets to advance tomography in scanning transmission electron microscopy.Scientific data3, 1 (2016), 160041

    2016

  11. [19]

    Ruijing Li, Houjin Chen, Yahui Peng, and Jupeng Li. 2019. Comparison of Four Forward Models for Breast Imaging in Ultrasound Computed Tomography.Chi- nese Journal of Electronics28, 4 (2019), 805–816. doi:10.1049/cje.2019.05.008

    2019 doi

  12. [20]

    Risheng Liu, Shichao Cheng, Long Ma, Xin Fan, and Zhongxuan Luo. 2019. Deep proximal unrolling: Algorithmic framework, convergence analysis and applications.IEEE Transactions on Image Processing28, 10 (2019), 5013–5026

    2019

  13. [21]

    Yun-Tao Liu, Heng Zhang, Hui Wang, Chang-Lu Tao, Guo-Qiang Bi, and Z Hong Zhou. 2022. Isotropic reconstruction for electron tomography with deep learning. Nature communications13, 1 (2022), 6482

    2022

  14. [22]

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. 2021. Nerf: Representing scenes as neural radiance fields for view synthesis.Commun. ACM65, 1 (2021), 99–106

    2021

  15. [23]

    David Nicolaides. 2001. Mesoscale Modelling.Molecular Simulation26, 1 (2001), 51–72. doi:10.1080/08927020108024200

    2001 doi

  16. [24]

    Minh Pham, Yakun Yuan, Arjun Rana, Stanley Osher, and Jianwei Miao. 2023. Accurate real space iterative reconstruction (RESIRE) algorithm for tomography. Scientific Reports13, 1 (2023), 5624

    2023

  17. [25]

    Alan Pryor Jr, Yongsoo Yang, Arjun Rana, Marcus Gallagher-Jones, Jihan Zhou, Yuan Hung Lo, Georgian Melinte, Wah Chiu, Jose A Rodriguez, and Jianwei Miao

  18. [26]

    Huaizhi Qu, Xiao Wang, Gengwei Zhang, Jie Peng, and Tianlong Chen. 2025. GEM: 3D Gaussian Splatting for Efficient and Accurate Cryo-EM Reconstruction. arXiv preprint arXiv:2509.25075(2025)

    2025

  19. [27]

    Michael Radermacher. 2006. Weighted back-projection methods. InElectron tomography: methods for three-dimensional visualization of structures in the cell. Springer, 245–273

    2006

  20. [28]

    Darius Rückert, Yuanhao Wang, Rui Li, Ramzi Idoughi, and Wolfgang Heidrich

  21. [29]

    Diana Sannino. 2021. Types and Classification of Nanomaterials.Nanotechnology (2021), 15–38. doi:10.1007/978-981-15-9437-3_2

    2021 doi

  22. [30]

    Jonathan Schwartz, Chris Harris, Jacob Pietryga, Huihuo Kumar, Anastasia Visheratina, Nicholas Kotor, Brianna Major, Patrick Avery, Peter Ercius, Utmarsch Ayachit, Berk Geveci, David Muller, Alessandro Genova, Yi Jiang, Marcus Han- well, and Robert Hovden. 2022. Real-time 3D a...

    2022 doi arXiv

  23. [31]

    MC Scott, Chien-Chun Chen, Matthew Mecklenburg, Chun Zhu, Rui Xu, Peter Ercius, Ulrich Dahmen, BC Regan, and Jianwei Miao. 2012. Electron tomography at 2.4-ångström resolution.Nature483, 7390 (2012), 444–447

    2012

  24. [32]

    Donald F Swinehart. 1962. The beer-lambert law.Journal of chemical education 39, 7 (1962), 333

    1962

  25. [33]

    2020.Nanotechnology and photocatalysis for environmental applications

    Muhammad Bilal Tahir, Muhammad Rafique, and Muhammad Shahid Rafique. 2020.Nanotechnology and photocatalysis for environmental applications. Elsevier

    2020

  26. [34]

    Ye Tian, Ying Fu, and Jun Zhang. 2023. Transformer-Based Under-sampled Single-Pixel Imaging.Chinese Journal of Electronics32, 5 (2023), 1151–1159. doi:10.23919/cje.2022.00.284

    2023 doi

  27. [35]

    Jeannot Trampert and Jean-Jacques Leveque. 1990. Simultaneous iterative re- construction technique: Physical interpretation based on the generalized least squares solution.Journal of Geophysical Research: Solid Earth95, B8 (1990), 12553–12559

    1990

  28. [36]

    Marin Van Heel and Michael Schatz. 2005. Fourier shell correlation threshold criteria.Journal of structural biology151, 3 (2005), 250–262

    2005

  29. [37]

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. 2004. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing13, 4 (2004), 600–612

    2004

  30. [38]

    Yanze Wei, Decai Zhao, and Dan Wang. 2023. Mesoscience in Hollow Multi- Shelled Structures.Advanced science (Weinheim, Baden-Wurttemberg, Germany) 11, 8 (2023), e2305408–e2305408. doi:10.1002/advs.202305408

    2023 doi

  31. [39]

    Matthew Weyland and Paul A Midgley. 2007. Electron tomography. (2007)

    2007

  32. [40]

    Zhaochun Yang. 2023. Nano Materials.Material Modeling in Finite Element Analysis(2023), 203–207. doi:10.1201/9781003436317-27

    2023 doi

  33. [41]

    Weihao Yu, Yuanhao Cai, Ruyi Zha, Zhiwen Fan, Chenxin Li, and Yixuan Yuan

  34. [42]

    Beiyuan Zhang, Hesong Li, Ruiwen Shao, and Ying Fu. 2026. 3D Gaussian Splatting for Annular Dark Field Scanning Transmission Electron Microscopy Tomography Reconstruction.arXiv preprint arXiv:2604.04693(2026)

    2026 arXiv

  35. [43]

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang

  36. [2017]

    GENFIRE: A generalized Fourier iterative reconstruction algorithm for high-resolution 3D imaging.Scientific reports7, 1 (2017), 10409

    2017

  37. [2018]

    InProceedings of the IEEE conference on computer vision and pattern recognition

    The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recognition. 586–595

  38. [2022]

    Neat: Neural adaptive tomography.ACM Transactions on Graphics (TOG) 41, 4 (2022), 1–13

    2022

  39. [2023]

    Graph.42, 4 (2023), 139–1

    3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph.42, 4 (2023), 139–1

    2023

  40. [2024]

    Three-dimensional morphology of polymeric membranes from electron tomography.Annual Review of Chemical and Biomolecular Engineering15 (2024)

    2024

  41. [2025]

    InProceedings of the IEEE/CVF International Conference on Computer Vision

    X2-gaussian: 4d radiative gaussian splatting for continuous-time tomo- graphic reconstruction. InProceedings of the IEEE/CVF International Conference on Computer Vision. 24728–24738

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.