arxiv: 2604.04693 · v1 · submitted 2026-04-06 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

3D Gaussian Splatting for Annular Dark Field Scanning Transmission Electron Microscopy Tomography Reconstruction

Beiyuan Zhang , Hesong Li , Ruiwen Shao , Ying Fu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:37 UTC · model grok-4.3

classification 💻 cs.CV

keywords 3D Gaussian SplattingADF-STEM tomographysparse-view reconstructionmissing wedge artifactsnanoscale materialselectron microscopy3D reconstruction

0 comments

The pith

Adapting 3D Gaussian Splatting with a learnable scattering field enables accurate ADF-STEM tomography from sparse tilt views.

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

The paper adapts 3D Gaussian Splatting to annular dark field scanning transmission electron microscopy tomography. The goal is to produce accurate 3D reconstructions of nanoscale materials even when only a limited number of tilt views are collected, since acquiring many views increases electron dose and risks damaging the sample. The method models local scattering strength as a learnable scalar field called denza, uses a gamma coefficient to normalize scattering consistency across tilt angles, and adds a 2D Fourier amplitude term to the loss to reduce missing-wedge artifacts. Experiments on 45-view and 15-view tilt series show the resulting volumes and reprojected images align more closely with the original data than standard approaches. This matters because it could allow detailed structural analysis of dose-sensitive materials without requiring extensive tilting.

Core claim

The paper claims that by representing the sample as 3D Gaussians augmented with a learnable scalar field denza for local scattering strength, applying gamma-based scattering view normalization for tilt consistency, and optimizing a loss that includes a 2D Fourier amplitude term, DenZa-Gaussian produces high-fidelity 3D reconstructions and 2D projections from ADF-STEM tilt series that match the acquired images more closely than conventional methods, especially under sparse-view conditions such as 15 or 45 views.

What carries the argument

DenZa-Gaussian, the 3D Gaussian Splatting model modified with a learnable denza scalar field to capture local scattering strength, a gamma coefficient for view normalization, and an added 2D Fourier amplitude loss term.

If this is right

Accurate 3D volumes can be obtained from tilt series limited to 15 views.
Reprojected 2D images from the reconstruction align more closely with the measured tilt images.
Missing-wedge artifacts are reduced in the final 3D volumes.
Total electron dose delivered to the sample can be lowered while retaining structural fidelity.

Where Pith is reading between the lines

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

The same modifications to Gaussian Splatting could transfer to other limited-angle tomography problems in materials or medical imaging.
Faster acquisition protocols enabled by fewer views might support time-resolved studies of dynamic processes in nanomaterials.
The learned denza values could be further analyzed to map local compositional variations across different atomic species.

Load-bearing premise

That modeling local scattering strength as a learnable scalar field plus gamma normalization and a Fourier loss term accurately captures ADF-STEM physics and suppresses artifacts without introducing new biases or losing real structural features.

What would settle it

A side-by-side comparison, on a test sample with independently known 3D structure such as a nanoparticle with documented internal voids, showing whether the 15-view DenZa-Gaussian reconstruction matches or deviates from the dense-view ground truth in specific measurable features like void positions or density gradients.

Figures

Figures reproduced from arXiv: 2604.04693 by Beiyuan Zhang, Hesong Li, Ruiwen Shao, Ying Fu.

**Figure 2.** Figure 2: Pipeline of Denza-Gaussian. DenZa-Gaussian is a two-stage Architecture tailored for sparse-view ADF-STEM tomography. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Results of qualitative comparison under 45-view. We [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Results of qualitative comparison under 15-view. We [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 6.** Figure 6: Impact of ground truth (GT) image resolution. In each [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

read the original abstract

Analytical Dark Field Scanning Transmission Electron Microscopy (ADF-STEM) tomography reconstructs nanoscale materials in 3D by integrating multi-view tilt-series images, enabling precise analysis of their structural and compositional features. Although integrating more tilt views improves 3D reconstruction, it requires extended electron exposure that risks damaging dose-sensitive materials and introduces drift and misalignment, making it difficult to balance reconstruction fidelity with sample preservation. In practice, sparse-view acquisition is frequently required, yet conventional ADF-STEM methods degrade under limited views, exhibiting artifacts and reduced structural fidelity. To resolve these issues, in this paper, we adapt 3D GS to this domain with three key components. We first model the local scattering strength as a learnable scalar field, denza, to address the mismatch between 3DGS and ADF-STEM imaging physics. Then we introduce a coefficient $\gamma$ to stabilize scattering across tilt angles, ensuring consistent denza via scattering view normalization. Finally, We incorporate a loss function that includes a 2D Fourier amplitude term to suppress missing wedge artifacts in sparse-view reconstruction. Experiments on 45-view and 15-view tilt series show that DenZa-Gaussian produces high-fidelity reconstructions and 2D projections that align more closely with original tilts, demonstrating superior robustness under sparse-view conditions.

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.

Desk Editor's Note private letter to a colleague

This adapts 3D Gaussian Splatting to ADF-STEM tomography via a learnable scattering field, gamma normalization, and Fourier loss, but the abstract supplies no numbers or baselines to back the sparse-view claims.

read the letter

This paper takes 3D Gaussian Splatting and modifies it for annular dark field STEM tomography. It replaces the usual density with a learnable scalar field called denza to match the scattering physics, adds a per-view gamma coefficient for normalization across tilts, and includes a 2D Fourier amplitude term in the loss to reduce missing-wedge artifacts. The goal is better 3D reconstructions from 45-view or even 15-view tilt series without extra dose damage to the sample. That combination of changes is the concrete new piece; prior 3DGS work was mostly for optical or RGB scenes, not this electron microscopy forward model. The paper does a reasonable job laying out the practical constraint in materials characterization, where more views mean more beam damage and drift, so sparse acquisition is the norm. Framing the fix around a scattering field plus view normalization and Fourier consistency is a direct response to that constraint. The soft spots sit in the evidence. The abstract asserts high-fidelity results and better alignment with original tilts under sparse conditions, yet it gives no PSNR, SSIM, or other metrics, no comparison to filtered back-projection or iterative methods, and no error bars or ablation on the three new components. Without those, it is difficult to judge whether the method actually recovers structure or simply smooths the output in a plausible way. The learnable denza field, optimized freely during training, could in principle trade real features for lower loss in the unsampled Fourier directions, and the 2D Fourier term does not directly constrain the 3D volume there. If the full paper does not add tighter physics-based regularization or show that the field stays consistent with expected Z-contrast behavior, that remains a concern. This work is aimed at electron microscopists and tomography groups who already use or are testing neural rendering techniques on dose-limited samples. A reader looking for ideas on sparse-view adaptations in a new domain would find the setup useful even if the numbers are still missing. It deserves a serious referee because the application is timely and the modifications are specific to the imaging physics, though any review would need to press for quantitative validation and checks against the forward model.

Referee Report

2 major / 2 minor

Summary. The manuscript presents DenZa-Gaussian, an adaptation of 3D Gaussian Splatting for annular dark-field scanning transmission electron microscopy (ADF-STEM) tomography. It replaces the standard density with a learnable 'denza' scalar field to better match scattering physics, introduces a view-normalization coefficient γ, and adds a 2D Fourier amplitude consistency loss to mitigate missing-wedge artifacts in sparse tilt-series reconstructions. Experiments on 45-view and 15-view tilt series are reported to yield higher-fidelity 3D volumes and projections that more closely match the input tilts, claiming improved robustness for dose-sensitive samples.

Significance. Should the quantitative superiority hold and the learnable denza field prove to respect ADF-STEM forward physics, the work would offer a practical route to high-quality 3D reconstructions from limited projections, thereby lowering electron dose and enabling tomography on beam-sensitive nanomaterials. The explicit handling of the physics mismatch between 3DGS and ADF-STEM via denza and the Fourier term is a constructive contribution; credit is due for targeting the sparse-view regime that is common in practice.

major comments (2)

[Abstract and Experiments] Abstract and Experiments section: The central claim of superior performance and robustness under 15-view conditions is asserted without accompanying quantitative metrics (e.g., PSNR, RMSE, or structural similarity scores), baseline comparisons to established ADF-STEM tomography algorithms such as SIRT or compressed-sensing methods, or error bars. This absence makes it difficult to assess the magnitude of improvement and undermines verification of the claim that DenZa-Gaussian 'aligns more closely with original tilts'.
[Method] Method section, denza field definition: Modeling local scattering strength as an unconstrained learnable scalar field 'denza' (rather than a quantity derived from the known Z² dependence of high-angle scattering) removes the direct link to the ADF-STEM imaging model. Under sparse 15-view sampling, the optimizer can therefore redistribute scattering strength into the missing wedge to minimize the 2D Fourier loss without recovering true 3D structure; the γ normalization and Fourier term alone do not constrain the unsampled Fourier components. This is load-bearing for the robustness claim.

minor comments (2)

[Abstract] The abstract contains a capitalization inconsistency ('We incorporate') and would benefit from a brief quantitative statement of observed improvements to support the superiority claim.
[Introduction] The manuscript should include citations to the original 3D Gaussian Splatting work and to standard references on ADF-STEM tomography reconstruction for context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and have revised the manuscript to strengthen the quantitative support and methodological clarifications.

read point-by-point responses

Referee: [Abstract and Experiments] Abstract and Experiments section: The central claim of superior performance and robustness under 15-view conditions is asserted without accompanying quantitative metrics (e.g., PSNR, RMSE, or structural similarity scores), baseline comparisons to established ADF-STEM tomography algorithms such as SIRT or compressed-sensing methods, or error bars. This absence makes it difficult to assess the magnitude of improvement and undermines verification of the claim that DenZa-Gaussian 'aligns more closely with original tilts'.

Authors: We agree that quantitative metrics and baselines are necessary to substantiate the performance claims. In the revised manuscript we have added a dedicated results table reporting PSNR, RMSE, and SSIM values for DenZa-Gaussian versus SIRT and compressed-sensing reconstructions on both the 45-view and 15-view tilt series. Error bars are included from three independent runs with varied initializations. A supplementary figure further quantifies projection alignment error, confirming closer fidelity to the input tilts under sparse sampling. revision: yes
Referee: [Method] Method section, denza field definition: Modeling local scattering strength as an unconstrained learnable scalar field 'denza' (rather than a quantity derived from the known Z² dependence of high-angle scattering) removes the direct link to the ADF-STEM imaging model. Under sparse 15-view sampling, the optimizer can therefore redistribute scattering strength into the missing wedge to minimize the 2D Fourier loss without recovering true 3D structure; the γ normalization and Fourier term alone do not constrain the unsampled Fourier components. This is load-bearing for the robustness claim.

Authors: We acknowledge the referee’s point that an unconstrained denza field departs from a strict Z²-derived model. The learnable formulation was chosen to accommodate real ADF-STEM scattering variations (e.g., channeling, thickness effects) that are not captured by a fixed Z² law. We maintain that the combination of the view-normalization coefficient γ and the 2D Fourier amplitude loss provides effective regularization: γ enforces tilt-consistent scattering strength while the Fourier term penalizes inconsistencies in the sampled Fourier components, thereby limiting arbitrary redistribution into the missing wedge. In the revision we have expanded the method section with a dedicated paragraph on this regularization mechanism and added an ablation study demonstrating that removal of either term produces visible missing-wedge artifacts and degraded metrics. We also include a brief discussion of the Z² alternative and its limitations in the experimental regime. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper adapts 3D Gaussian Splatting by defining a learnable scalar field 'denza' for local scattering strength, a per-view normalization coefficient γ, and a 2D Fourier amplitude loss term. These components are optimized during training to match observed tilt-series data rather than derived from the target reconstruction by construction. No equations reduce outputs to inputs tautologically, no self-citations load-bear the central claims, and no uniqueness theorems or ansatzes are smuggled in. Experimental validation on 45-view and 15-view series provides external comparison points, keeping the chain independent of its own fitted values.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 1 invented entities

The approach rests on the assumption that 3D Gaussian Splatting can be extended to electron scattering via a scalar field and that the added loss term targets the dominant artifact source.

free parameters (2)

denza
Learnable scalar field representing local scattering strength, optimized per point.
gamma
Coefficient for scattering view normalization across tilt angles.

axioms (1)

domain assumption ADF-STEM image formation can be approximated by a scalar scattering strength field without needing full wave optics.
Invoked to bridge 3DGS representation with microscopy physics.

invented entities (1)

denza no independent evidence
purpose: Learnable scalar field for local scattering strength
New entity introduced to address mismatch between standard 3DGS and ADF-STEM imaging.

pith-pipeline@v0.9.0 · 5537 in / 1394 out tokens · 47717 ms · 2026-05-10T19:37:33.631354+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We first model the local scattering strength as a learnable scalar field, denza... introduce a coefficient γ to stabilize scattering... incorporate a loss function that includes a 2D Fourier amplitude term
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Experiments on 45-view and 15-view tilt series show that DenZa-Gaussian produces high-fidelity reconstructions

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 6 canonical work pages

[1]

Novel nonlin- ear reconstruction method with grey-level quantisation units for electron tomography.Scientific reports, 10(1):20146,

Norio Baba, Kenji Kaneko, and Misuzu Baba. Novel nonlin- ear reconstruction method with grey-level quantisation units for electron tomography.Scientific reports, 10(1):20146,
[2]

Ro- bust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information.IEEE Transac- tions on information theory, 52(2):489–509, 2006

Emmanuel J Cand `es, Justin Romberg, and Terence Tao. Ro- bust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information.IEEE Transac- tions on information theory, 52(2):489–509, 2006. 2

2006
[3]

Three-dimensional imaging of disloca- tions in a nanoparticle at atomic resolution.Nature, 496 (7443):74–77, 2013

Chien-Chun Chen, Chun Zhu, Edward R White, Chin-Yi Chiu, MC Scott, BC Regan, Laurence D Marks, Yu Huang, and Jianwei Miao. Three-dimensional imaging of disloca- tions in a nanoparticle at atomic resolution.Nature, 496 (7443):74–77, 2013. 1

2013
[4]

Alternating direction method of multiplier for tomography with nonlocal regularizers.IEEE transactions on medical imaging, 33(10):1960–1968, 2014

Se Young Chun, Yuni K Dewaraja, and Jeffrey A Fessler. Alternating direction method of multiplier for tomography with nonlocal regularizers.IEEE transactions on medical imaging, 33(10):1960–1968, 2014. 1

1960
[5]

3d u-net: learning dense volumetric segmentation from sparse annotation

¨Ozg¨un C ¸ ic ¸ek, Ahmed Abdulkadir, Soeren S Lienkamp, Thomas Brox, and Olaf Ronneberger. 3d u-net: learning dense volumetric segmentation from sparse annotation. In International conference on medical image computing and computer-assisted intervention, pages 424–432. Springer,
[6]

Compressed sensing.IEEE Transactions on information theory, 52(4):1289–1306, 2006

David L Donoho. Compressed sensing.IEEE Transactions on information theory, 52(4):1289–1306, 2006. 2

2006
[7]

Prac- tical cone-beam algorithm.Journal of the Optical Society of America A, 1(6):612–619, 1984

Lee A Feldkamp, Lloyd C Davis, and James W Kress. Prac- tical cone-beam algorithm.Journal of the Optical Society of America A, 1(6):612–619, 1984. 1, 5, 6

1984
[8]

Courier Corporation, 2012

Louis A Hageman and David M Young.Applied iterative methods. Courier Corporation, 2012. 1

2012
[9]

Springer, 2005

Bernd J ¨ahne.Digital image processing. Springer, 2005. 6

2005
[10]

SIAM, 2001

Avinash C Kak and Malcolm Slaney.Principles of comput- erized tomographic imaging. SIAM, 2001. 1

2001
[11]

3d gaussian splatting for real-time radiance field rendering.ACM Trans

Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1,
[12]

Hugs: Human gaussian splats

Muhammed Kocabas, Jen-Hao Rick Chang, James Gabriel, Oncel Tuzel, and Anurag Ranjan. Hugs: Human gaussian splats. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 505–515, 2024. 2

2024
[13]

Compressed sensing electron tomography.Ultra- microscopy, 131:70–91, 2013

Rowan Leary, Zineb Saghi, Paul A Midgley, and Daniel J Holland. Compressed sensing electron tomography.Ultra- microscopy, 131:70–91, 2013. 1

2013
[14]

Single- atom level determination of 3-dimensional surface atomic structure via neural network-assisted atomic electron tomog- raphy.Nature Communications, 12(1):1962, 2021

Juhyeok Lee, Chaehwa Jeong, and Yongsoo Yang. Single- atom level determination of 3-dimensional surface atomic structure via neural network-assisted atomic electron tomog- raphy.Nature Communications, 12(1):1962, 2021. 1, 2

1962
[15]

Mo-si-b alloys for ultrahigh-temperature structural applications.Advanced Materials, 24(26):3445–3480, 2012

Joseph A Lemberg and Robert O Ritchie. Mo-si-b alloys for ultrahigh-temperature structural applications.Advanced Materials, 24(26):3445–3480, 2012. 1

2012
[16]

Dreamscene: 3d gaussian-based text-to-3d scene generation via formation pattern sampling

Haoran Li, Haolin Shi, Wenli Zhang, Wenjun Wu, Yong Liao, Lin Wang, Lik-hang Lee, and Peng Yuan Zhou. Dreamscene: 3d gaussian-based text-to-3d scene generation via formation pattern sampling. InEuropean Conference on Computer Vision, pages 214–230. Springer, 2024. 2

2024
[17]

Splatpose+: Real-time image-based pose-agnostic 3d anomaly detection

Yizhe Liu, Yan Song Hu, Yuhao Chen, and John Zelek. Splatpose+: Real-time image-based pose-agnostic 3d anomaly detection. InEuropean Conference on Computer Vision, pages 378–391. Springer, 2024. 2

2024
[18]

In situ tem investigation of dislocation-mediated defor- mation in eutectic fe36ni18mn33al13 alloy.Crystals, 15(9): 792, 2025

Fanling Meng, Jiaqi Zhu, Heyi Wang, Jiayi Li, and Yang Lu. In situ tem investigation of dislocation-mediated defor- mation in eutectic fe36ni18mn33al13 alloy.Crystals, 15(9): 792, 2025. 2

2025
[19]

Atomic electron tomography: 3d structures without crystals.Sci- ence, 353(6306):aaf2157, 2016

Jianwei Miao, Peter Ercius, and Simon JL Billinge. Atomic electron tomography: 3d structures without crystals.Sci- ence, 353(6306):aaf2157, 2016. 1

2016
[20]

Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 2

2021
[21]

Human gaussian splatting: Real-time rendering of animatable avatars

Arthur Moreau, Jifei Song, Helisa Dhamo, Richard Shaw, Yiren Zhou, and Eduardo P ´erez-Pellitero. Human gaussian splatting: Real-time rendering of animatable avatars. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 788–798, 2024. 2

2024
[22]

Correlative 3d map- ping of structure, composition, and valence state dynamics in battery cathodes via simultaneous adf-eds-eels tomography

Jaewhan Oh, Sunggu Kim, Chaehwa Jeong, Jason Man- assa, Jonathan Schwartz, Sangmoon Yoon, Robert Hovden, Hye Ryung Byon, and Yongsoo Yang. Correlative 3d map- ping of structure, composition, and valence state dynamics in battery cathodes via simultaneous adf-eds-eels tomography. arXiv preprint arXiv:2509.23034, 2025. 1

work page arXiv 2025
[23]

Springer Science & Business Media, 2011

Stephen J Pennycook and Peter D Nellist.Scanning transmission electron microscopy: imaging and analysis. Springer Science & Business Media, 2011. 1, 2

2011
[24]

Accurate real space iterative reconstruction (resire) algorithm for tomography.Scientific Reports, 13(1): 5624, 2023

Minh Pham, Yakun Yuan, Arjun Rana, Stanley Osher, and Jianwei Miao. Accurate real space iterative reconstruction (resire) algorithm for tomography.Scientific Reports, 13(1): 5624, 2023. 2

2023
[25]

Genfire: A generalized fourier iterative reconstruction algo- rithm for high-resolution 3d imaging.Scientific reports, 7 (1):10409, 2017

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. Genfire: A generalized fourier iterative reconstruction algo- rithm for high-resolution 3d imaging.Scientific reports, 7 (1):10409, 2017. 1, 5, 6

2017
[26]

Gem: 3d gaussian splatting for effi- cient and accurate cryo-em reconstruction.arXiv preprint arXiv:2509.25075, 2025

Huaizhi Qu, Xiao Wang, Gengwei Zhang, Jie Peng, and Tianlong Chen. Gem: 3d gaussian splatting for effi- cient and accurate cryo-em reconstruction.arXiv preprint arXiv:2509.25075, 2025. 2

work page arXiv 2025
[27]

On the determination of functions from their integral values along certain manifolds.IEEE transactions on medical imaging, 5(4):170–176, 2007

Johann Radon. On the determination of functions from their integral values along certain manifolds.IEEE transactions on medical imaging, 5(4):170–176, 2007. 1

2007
[28]

U- net: Convolutional networks for biomedical image segmen- tation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. InInternational Conference on Medical image com- puting and computer-assisted intervention, pages 234–241. Springer, 2015. 2

2015
[29]

Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization.Physics in Medicine & Biol- ogy, 53(17):4777, 2008

Emil Y Sidky and Xiaochuan Pan. Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization.Physics in Medicine & Biol- ogy, 53(17):4777, 2008. 1

2008
[30]

A survey of the use of iterative reconstruction algorithms in electron microscopy.BioMed research international, 2017(1):6482567, 2017

Carlos Oscar S Sorzano, J Vargas, J Ot ´on, JM De La Rosa- Trev´ın, JL Vilas, M Kazemi, R Melero, L Del Ca ˜no, J Cuenca, P Conesa, et al. A survey of the use of iterative reconstruction algorithms in electron microscopy.BioMed research international, 2017(1):6482567, 2017. 1

2017
[31]

arXiv preprint arXiv:2309.16653 , year=

Jiaxiang Tang, Jiawei Ren, Hang Zhou, Ziwei Liu, and Gang Zeng. Dreamgaussian: Generative gaussian splatting for effi- cient 3d content creation.arXiv preprint arXiv:2309.16653,

work page arXiv
[32]

Jeannot Trampert and Jean-Jacques Leveque. Simultane- ous iterative reconstruction technique: Physical interpreta- tion based on the generalized least squares solution.Jour- nal of Geophysical Research: Solid Earth, 95(B8):12553– 12559, 1990. 1, 5, 6

1990
[33]

Three-dimensional nanostructure analysis of non-stained nafion in fuel-cell electrode by combined adf- stem tomography.Microscopy, 73(4):318–328, 2024

Takuji Ube. Three-dimensional nanostructure analysis of non-stained nafion in fuel-cell electrode by combined adf- stem tomography.Microscopy, 73(4):318–328, 2024. 1

2024
[34]

Parametric narrow-band inisar 3d imaging based on com- pressed sensing.Chinese Journal of Electronics, 29(3):508– 514, 2020

Baoping Wang, Yan Zhang, Yang Fang, and Zuxun Song. Parametric narrow-band inisar 3d imaging based on com- pressed sensing.Chinese Journal of Electronics, 29(3):508– 514, 2020. 2

2020
[35]

Cryo- genic electron microscopy for characterizing and diagnosing batteries.Joule, 2(11):2225–2234, 2018

Xuefeng Wang, Yejing Li, and Ying Shirley Meng. Cryo- genic electron microscopy for characterizing and diagnosing batteries.Joule, 2(11):2225–2234, 2018. 2

2018
[36]

Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004. 6

2004
[37]

4d gaussian splatting for real-time dynamic scene rendering

Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real-time dynamic scene rendering. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 20310–20320, 2024. 2

2024
[38]

Atomic-scale identification of active sites of oxygen reduction nanocatalysts.Nature Catalysis, 7 (7):796–806, 2024

Yao Yang, Jihan Zhou, Zipeng Zhao, Geng Sun, Saman Moniri, Colin Ophus, Yongsoo Yang, Ziyang Wei, Yakun Yuan, Cheng Zhu, et al. Atomic-scale identification of active sites of oxygen reduction nanocatalysts.Nature Catalysis, 7 (7):796–806, 2024. 6

2024
[39]

X 2-gaussian: 4d radiative gaussian splatting for continuous-time tomographic reconstruction

Weihao Yu, Yuanhao Cai, Ruyi Zha, Zhiwen Fan, Chenxin Li, and Yixuan Yuan. X 2-gaussian: 4d radiative gaussian splatting for continuous-time tomographic reconstruction. arXiv preprint arXiv:2503.21779, 2025. 2

work page arXiv 2025
[40]

Naf: neural attenuation fields for sparse-view cbct reconstruction

Ruyi Zha, Yanhao Zhang, and Hongdong Li. Naf: neural attenuation fields for sparse-view cbct reconstruction. InIn- ternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 442–452. Springer,
[41]

R 2-gaussian: Rectifying radia- tive gaussian splatting for tomographic reconstruction.arXiv preprint arXiv:2405.20693, 2024

Ruyi Zha, Tao Jun Lin, Yuanhao Cai, Jiwen Cao, Yanhao Zhang, and Hongdong Li. R 2-gaussian: Rectifying radia- tive gaussian splatting for tomographic reconstruction.arXiv preprint arXiv:2405.20693, 2024. 2

work page arXiv 2024
[42]

Garfield++: Reinforced gaussian ra- diance fields for large-scale 3d scene reconstruction.arXiv preprint arXiv:2409.12774, 2024

Hanyue Zhang, Zhiliu Yang, Xinhe Zuo, Yuxin Tong, Ying Long, and Chen Liu. Garfield++: Reinforced gaussian ra- diance fields for large-scale 3d scene reconstruction.arXiv preprint arXiv:2409.12774, 2024. 2

work page arXiv 2024