arxiv: 2605.08566 · v1 · submitted 2026-05-08 · 💻 cs.CV · cs.LG· q-bio.QM

Recognition: 2 theorem links

· Lean Theorem

MicroDiffuse3D: A Foundation Model for 3D Microscopy Imaging Restoration

Brian Wong, Dan Fu, Erin Dunnington, Hanwen Xu, King Wai Chiu, Sheng Wang, Tangqi Fang, Yongkang Li

Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:15 UTC · model grok-4.3

classification 💻 cs.CV cs.LGq-bio.QM

keywords 3D microscopyimage restorationfoundation modelsuper-resolutiondenoisingchemical imagingvolumetric reconstruction

0 comments

The pith

A pretrained foundation model restores high-quality 3D structure from sparse and degraded microscopy measurements.

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

The paper presents MicroDiffuse3D as a pretrained foundation model that recovers clear volumetric images from low-resolution, noisy, or sparsely sampled inputs in chemical microscopy. Chemical imaging supplies label-free biochemical detail but is constrained by slow 3D acquisition times. The model enables higher-throughput scanning by learning restoration across multiple degradation types, including 16-fold sparse super-resolution, combined resolution and noise loss, and low-SNR denoising. In the sparse super-resolution case the restored volumes show improved depth continuity and fewer artifacts, raising segmentation accuracy by 10.58 percent and line-profile agreement by 15.59 percent.

Core claim

MicroDiffuse3D is a pretrained foundation model for 3D microscopy image restoration that recovers high-quality volumetric structure from degraded low-resolution measurements acquired at substantially higher throughput. Evaluated on three settings—16-fold sparse super-resolution, joint resolution-plus-noise degradation, and low-SNR denoising—the model outperforms strong baselines. In the sparse super-resolution regime it yields clearer continuity across depth with fewer artifacts.

What carries the argument

MicroDiffuse3D, a pretrained foundation model that learns to map degraded 3D inputs onto restored high-quality volumetric outputs.

If this is right

Pretrained 3D restoration becomes a broadly applicable strategy for overcoming throughput and SNR limits in volumetric chemical imaging.
High-resolution analysis becomes feasible at scales and speeds that were previously difficult to achieve.
Downstream tasks such as segmentation gain from the improved depth continuity and reduced artifacts.

Where Pith is reading between the lines

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

The same pretraining strategy could be applied to other 3D modalities that face similar acquisition-speed versus resolution trade-offs.
Lower-resolution hardware combined with this restoration step could reduce equipment costs while maintaining usable output quality.
Faster low-resolution scans followed by model restoration might enable real-time or live-sample 3D workflows.

Load-bearing premise

The pretrained foundation model generalizes effectively to the three specific 3D degradation regimes in chemical microscopy without overfitting to the training distribution or requiring heavy task-specific adaptation.

What would settle it

Testing the model on a fourth unseen degradation type or on data acquired from an independent microscope system and checking whether performance still exceeds baselines would falsify the claim of broad applicability.

Figures

Figures reproduced from arXiv: 2605.08566 by Brian Wong, Dan Fu, Erin Dunnington, Hanwen Xu, King Wai Chiu, Sheng Wang, Tangqi Fang, Yongkang Li.

**Figure 1.** Figure 1: Overview of MicroDiffuse3D. a, Conditional Diffusion Architecture. MicroDiffuse3D takes anisotropic volumetric data as condition and processes them via a deep feature encoder. Conditioning features are then injected into the latent diffusion process for high-fidelity reconstruction. b, Large-scale pretraining strategy. The model is pretrained on 2.38 million unpaired SRS-derived slices using two complement… view at source ↗

**Figure 2.** Figure 2: 3D super-resolution across microscopy modalities. a, b. Quantitative evaluation of superresolution performance. Box plots detail PSNR, SSIM, MS-SSIM and LPIPS distributions across independent volumetric samples from SIM and SRS datasets. Our 3D diffusion model consistently outperforms baselines. c. Qualitative assessment of axial structural fidelity. Representative volumetric cross-sections demonstrate th… view at source ↗

**Figure 3.** Figure 3: 3D Denoising. a, b. Quantitative evaluation of denoising performance. Box plots detail PSNR, SSIM, MS-SSIM and LPIPS distributions across independent volumetric samples from SRS dataset. Our 3D diffusion model consistently outperforms 2D-based baselines. c. Qualitative assessment of recovery and axial structural fidelity. Representative cross-sections of baselines (represented by 3DRCAN, Row 2) and our mod… view at source ↗

**Figure 4.** Figure 4 [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

read the original abstract

Chemical imaging enables label-free visualization of cells, tissues and living systems while providing direct biochemical information that is difficult to obtain with conventional fluorescence microscopy. Despite its promise in applications ranging from intraoperative diagnosis to drug-response analysis, its broader use remains limited by slow data acquisition, particularly for three-dimensional imaging. Here we present MicroDiffuse3D, a pretrained foundation model for 3D microscopy image restoration that recovers high-quality volumetric structure from degraded low-resolution measurements acquired at substantially higher throughput. We evaluated MicroDiffuse3D across three challenging restoration settings, including 3D super-resolution under 16-fold volumetric sparsity, joint degradation in resolution and noise, and 3D denoising in the low signal-to-noise ratio (SNR) regime, where the model delivered clear gains over strong baselines. Under the sparse 3D super-resolution setting, MicroDiffuse3D produced clearer continuity across depth with fewer artifacts and improved segmentation quality by 10.58% and line-profile concordance by 15.59%. Together, our results establish pretrained 3D restoration as a broadly applicable strategy for overcoming the throughput and SNR limitations in volumetric chemical imaging, enabling high-resolution analysis at scales and speeds that were previously difficult to achieve.

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

MicroDiffuse3D shows usable gains on 3D chemical microscopy restoration but the pretraining story is not yet isolated from other factors.

read the letter

The main thing to know is that this paper puts forward a pretrained model for restoring 3D volumes in label-free chemical imaging and reports concrete improvements under sparse sampling, combined degradations, and low SNR. Those gains matter because slow 3D acquisition is a real limit for the field, and anything that lets people scan faster while keeping usable structure is worth attention. The 16-fold volumetric sparsity case with better depth continuity, plus the 10.58% segmentation lift and 15.59% line-profile improvement, are the numbers that stand out from the abstract. What is actually new is the move to apply foundation-model style pretraining specifically to 3D volumetric restoration in this modality rather than 2D or standard fluorescence work. The paper does a reasonable job framing the throughput problem and showing that the model reduces artifacts relative to the baselines they chose. That part is straightforward and useful for readers who need practical restoration tools. The soft spot is that the evidence tying the gains to the pretraining step itself is thin. No ablation that removes pretraining appears, there is no breakdown of the pretraining corpus size or diversity, and no explicit checks for distribution shift or overfitting to the test regimes. Without those, it is hard to rule out that the architecture or task-specific fine-tuning is doing most of the work. The stress-test note on this point holds up from what is shown. This paper is for people working on computational methods for chemical or label-free microscopy who want a starting point for 3D restoration benchmarks. A reader who cares about throughput in intraoperative or drug-screening contexts would find the quantitative comparisons worth looking at, though they would need the code and data to judge reproducibility. It deserves peer review because the underlying problem is important and the claims are specific enough to be tested with the right experiments. I would send it out, with the clear expectation that reviewers will ask for the missing ablations and training details.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces MicroDiffuse3D, a pretrained foundation model for 3D microscopy image restoration in chemical imaging. It claims to recover high-quality volumetric structure from degraded low-resolution measurements and reports gains over baselines across three settings: 16-fold volumetric sparse super-resolution (with clearer depth continuity, fewer artifacts, +10.58% segmentation quality, and +15.59% line-profile concordance), joint resolution+noise degradation, and low-SNR denoising. The work positions pretrained 3D restoration as a broadly applicable strategy to overcome throughput and SNR limits.

Significance. If the reported gains are shown to stem from pretraining-enabled generalization rather than task-specific factors, the work could meaningfully advance high-throughput 3D chemical microscopy for intraoperative diagnosis and drug-response studies by enabling high-resolution analysis at previously inaccessible scales and speeds.

major comments (2)

[Abstract] The abstract reports quantitative gains (10.58% segmentation quality, 15.59% line-profile concordance) but supplies no information on training datasets, model architecture, baseline implementations, statistical testing, or cross-validation procedures. This information is load-bearing for the central claim that pretraining confers robustness to the three specific 3D degradation regimes.
[Experiments] No ablation removing the pretraining step, pretraining corpus diversity metrics, or explicit distribution-shift tests are described. Without these, the reported improvements in depth continuity and artifact reduction cannot be confidently attributed to foundation-model generalization rather than architecture or in-distribution fine-tuning.

minor comments (1)

[Abstract] The abstract would be clearer if it briefly noted the number of test volumes or the specific nature of the strong baselines used for comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comments point by point below, agreeing where revisions are warranted to better support our claims about MicroDiffuse3D.

read point-by-point responses

Referee: [Abstract] The abstract reports quantitative gains (10.58% segmentation quality, 15.59% line-profile concordance) but supplies no information on training datasets, model architecture, baseline implementations, statistical testing, or cross-validation procedures. This information is load-bearing for the central claim that pretraining confers robustness to the three specific 3D degradation regimes.

Authors: We agree that the abstract would be strengthened by including concise references to these elements. In the revised manuscript we will expand the abstract to note the pretraining on diverse 3D chemical microscopy volumes, the 3D diffusion-based architecture, the baseline methods used, and the cross-validation with statistical testing employed. Full details remain in the Methods and Experiments sections. revision: yes
Referee: [Experiments] No ablation removing the pretraining step, pretraining corpus diversity metrics, or explicit distribution-shift tests are described. Without these, the reported improvements in depth continuity and artifact reduction cannot be confidently attributed to foundation-model generalization rather than architecture or in-distribution fine-tuning.

Authors: We acknowledge that an explicit ablation removing the pretraining step is absent from the current manuscript. We will add a discussion of this limitation in the revised Experiments section and, where computationally feasible, include a limited ablation on a data subset to help isolate the pretraining contribution. Corpus diversity metrics appear in the supplementary materials, and the three distinct degradation regimes (sparse super-resolution, joint degradation, and low-SNR denoising) function as distribution-shift tests; we will clarify this attribution and add explicit shift analysis in the revision. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical performance claims with no derivations or self-referential reductions

full rationale

The paper reports empirical results from training and evaluating a pretrained 3D restoration model on three degradation regimes, with gains quantified via segmentation quality (+10.58%) and line-profile concordance (+15.59%) on held-out test settings. No equations, first-principles derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. The central claims rest on experimental comparisons rather than any step that reduces by construction to its own inputs, satisfying the criteria for a self-contained non-circular finding.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review provides no explicit list of free parameters or axioms; the approach implicitly relies on standard deep-learning assumptions that large neural networks can learn useful image priors from pretraining data.

free parameters (1)

Neural network weights and training hyperparameters
All model parameters are fitted during pretraining and any fine-tuning on microscopy data; exact count and values unknown from abstract.

axioms (1)

domain assumption Deep neural networks pretrained on large image corpora can serve as effective priors for restoring degraded 3D microscopy volumes.
Central premise of the foundation-model strategy described in the abstract.

pith-pipeline@v0.9.0 · 5538 in / 1373 out tokens · 61877 ms · 2026-05-12T01:15:33.529050+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/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

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

MicroDiffuse3D operates as a transformer-based Latent Diffusion Model... decomposed lateral (XY-plane) and axial (Z-axis) attention mechanisms
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

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

Built on large-scale pretraining over a curated corpus of 2.55 million microscopy images... dynamic physical degradation simulations and 3D block-masking

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

44 extracted references · 44 canonical work pages · 1 internal anchor

[2]

Science376(6594), 5197 (2022) https://doi.org/10.1126/science

Ji-Xin Cheng and X. Sunney Xie. “Vibrational spectroscopic imaging of living systems: An emerging platform for biology and medicine”. In:Science350.6264 (2015), aaa8870.doi:10.1126/science. aaa8870.url:http://science.sciencemag.org/content/sci/350/6264/aaa8870.full.pdf

work page doi:10.1126/science 2015
[3]

Bond-selective imaging by optically sensing the mid-infrared photothermal effect

Yeran Bai, Jiaze Yin, and Ji-Xin Cheng. “Bond-selective imaging by optically sensing the mid-infrared photothermal effect”. In:Science Advances7.20 (2021), eabg1559.doi:doi:10.1126/sciadv.abg1559. url:https://www.science.org/doi/abs/10.1126/sciadv.abg1559

work page doi:10.1126/sciadv.abg1559 2021
[4]

A 20-year journey on the invention of vibrational photothermal microscopy

J. X. Cheng. “A 20-year journey on the invention of vibrational photothermal microscopy”. In:Nat Methods22.5 (2025). 1548-7105 Cheng, Ji-Xin Orcid: 0000-0002-5607-6683 R35 GM 136223/U.S. De- partment of Health & Human Services — NIH — National Institute of General Medical Sciences (NIGMS)/ Journal Article United States 2025/05/14 Nat Methods. 2025 May;22(...

work page doi:10.1038/s41592-025-02619-0 2025
[5]

Rapid intraoperative histology of unprocessed surgical specimens via fibre- laser-based stimulated Raman scattering microscopy

Daniel A. Orringer et al. “Rapid intraoperative histology of unprocessed surgical specimens via fibre- laser-based stimulated Raman scattering microscopy”. In:Nature Biomedical Engineering1.2 (2017), p. 0027.issn: 2157-846X.doi:10.1038/s41551-016-0027

work page doi:10.1038/s41551-016-0027 2017
[6]

The TRIPOD-LLM reporting guideline for studies using large language models

Todd C. Hollon et al. “Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks”. In:Nature Medicine26.1 (2020), pp. 52–58.issn: 1546-170X.doi: 10.1038/s41591- 019- 0715- 9.url:https://doi.org/10.1038/s41591- 019- 0715- 9%20https: //escholarship.org/content/qt2bx2h425/qt2bx2h425.pdf?t=qeu5ld

work page doi:10.1038/s41591- 2020
[8]

Optical imaging of metabolic dynamics in animals

Lingyan Shi et al. “Optical imaging of metabolic dynamics in animals”. In:Nature Communications 9.1 (2018), p. 2995.issn: 2041-1723.doi:10.1038/s41467-018-05401-3.url:https://doi.org/ 10.1038/s41467-018-05401-3

work page doi:10.1038/s41467-018-05401-3.url:https://doi.org/ 2018
[9]

Spectral tracing of deuterium for imaging glucose metabolism

Luyuan Zhang et al. “Spectral tracing of deuterium for imaging glucose metabolism”. In:Nature Biomedical Engineering3.5 (2019), pp. 402–413.issn: 2157-846X.doi:10.1038/s41551-019-0393-4. url:https://doi.org/10.1038/s41551-019-0393-4

work page doi:10.1038/s41551-019-0393-4 2019
[10]

Performance and Analysis of the Alchemical Transfer Method for Binding-Free-Energy Predictions of Diverse Ligands

B. S. Wong et al. “Facilitated Transport of EGFR Inhibitors Plays an Important Role in Their Cellular Uptake”. In:Analytical Chemistry96.4 (2024), pp. 1547–1555.issn: 0003-2700.doi:10.1021/acs. analchem.3c04242. 18

work page doi:10.1021/acs 2024
[11]

Assessing drug uptake and response differences in 2D and 3D cellular environments using stimulated Raman scattering microscopy

Fiona Xi Xu et al. “Assessing drug uptake and response differences in 2D and 3D cellular environments using stimulated Raman scattering microscopy”. In:bioRxiv(2024), p. 2024.04.22.590622.doi:10. 1101/2024.04.22.590622.url:https://www.biorxiv.org/content/biorxiv/early/2024/04/26/ 2024.04.22.590622.full.pdf

work page 2024
[12]

Quantitative Stimulated Raman Scattering Microscopy: Promises and Pitfalls

Bryce Manifold and Dan Fu. “Quantitative Stimulated Raman Scattering Microscopy: Promises and Pitfalls”. In:Annual Reviews of Analytical Chemistry15.1 (2022), pp. 269–289.doi:10 . 1146 / annurev - anchem - 061020 - 015110.url:https : / / www . annualreviews . org / doi / abs / 10 . 1146 / annurev-anchem-061020-015110

work page 2022
[13]

Review of bio-optical imaging systems with a high space-bandwidth product

Jongchan Park et al. “Review of bio-optical imaging systems with a high space-bandwidth product”. In:Advanced Photonics3.4 (2021), p. 044001.doi:10 . 1117 / 1 . AP . 3 . 4 . 044001.url:https : //doi.org/10.1117/1.AP.3.4.044001

work page doi:10.1117/1.ap.3.4.044001 2021
[14]

AI Feynman: A physics-inspired method for symbolic regression.Science Advances, 6(16):eaay2631, 2020

Wei Min and Xin Gao. “Absolute signal of stimulated Raman scattering microscopy: A quantum electrodynamics treatment”. In:Science Advances10.50 (2024), eadm8424.doi:10.1126/sciadv. adm8424. eprint:https : / / www . science . org / doi / pdf / 10 . 1126 / sciadv . adm8424.url:https : //www.science.org/doi/abs/10.1126/sciadv.adm8424

work page doi:10.1126/sciadv 2024
[15]

Content-aware image restoration: pushing the limits of fluorescence microscopy

Martin Weigert, Uwe Schmidt, Tobias Boothe, et al. “Content-aware image restoration: pushing the limits of fluorescence microscopy”. In:Nature Methods15 (2018), pp. 1090–1097.doi:10 . 1038 / s41592-018-0216-7

work page 2018
[16]

Deep learning enables cross-modality super-resolution in fluorescence microscopy

Hongda Wang et al. “Deep learning enables cross-modality super-resolution in fluorescence microscopy”. In:Nature Methods16 (2019), pp. 103–110.doi:10.1038/s41592-018-0239-0

work page doi:10.1038/s41592-018-0239-0 2019
[17]

Evaluation and development of deep neural networks for image super-resolution in optical microscopy

Chang Qiao et al. “Evaluation and development of deep neural networks for image super-resolution in optical microscopy”. In:Nature Methods18.2 (2021), pp. 194–202.doi:10.1038/s41592-020-01048- 5

work page doi:10.1038/s41592-020-01048- 2021
[18]

Zero-shot learning enables instant denoising and super- resolution in optical fluorescence microscopy

Chang Qiao, Yunmin Zeng, Qian Meng, et al. “Zero-shot learning enables instant denoising and super- resolution in optical fluorescence microscopy”. In:Nature Communications15 (2024), p. 4180.doi: 10.1038/s41467-024-48575-9

work page doi:10.1038/s41467-024-48575-9 2024
[19]

Incorporating the image formation process into deep learning improves network performance

Yijun Li, Yijun Su, Min Guo, et al. “Incorporating the image formation process into deep learning improves network performance”. In:Nature Methods19 (2022), pp. 1427–1437.doi:10.1038/s41592- 022-01652-7

work page doi:10.1038/s41592- 2022
[20]

Three-dimensional residual channel attention networks denoise and sharpen fluores- cence microscopy image volumes

Jiji Chen et al. “Three-dimensional residual channel attention networks denoise and sharpen fluores- cence microscopy image volumes”. In:Nature Methods18.6 (2021), pp. 678–687

work page 2021
[21]

Reference-free Axial Super-resolution of 3D Microscopy Images using Implicit Neural Representation with a 2D Diffusion Prior

Kyungryun Lee and Won-Ki Jeong. “Reference-free Axial Super-resolution of 3D Microscopy Images using Implicit Neural Representation with a 2D Diffusion Prior”. In:Medical Image Computing and Computer Assisted Intervention – MICCAI 2024. arXiv:2408.08616. 2024

work page arXiv 2024
[22]

Addressing fairness in artificial intelligence for medical imaging.Nature Communications, 13:4581, 2022

Hyoungjun Park et al. “Deep learning enables reference-free isotropic super-resolution for volumetric fluorescence microscopy”. In:Nature Communications13 (2022), p. 3297.doi:10.1038/s41467-022- 30949-6

work page doi:10.1038/s41467-022- 2022
[23]

Deep self-learning enables fast, high-fidelity isotropic resolution restoration for volumetric fluorescence microscopy

Kai Ning, Bin Lu, Xin Wang, et al. “Deep self-learning enables fast, high-fidelity isotropic resolution restoration for volumetric fluorescence microscopy”. In:Light: Science & Applications12 (2023), p. 204. doi:10.1038/s41377-023-01230-2

work page doi:10.1038/s41377-023-01230-2 2023
[24]

InterpolAI: deep learning-based optical flow inter- polation and restoration of biomedical images for improved 3D tissue mapping

Sushama Joshi, Amanda Forjaz, K. S. Han, et al. “InterpolAI: deep learning-based optical flow inter- polation and restoration of biomedical images for improved 3D tissue mapping”. In:Nature Methods 22 (2025), pp. 1556–1567.doi:10.1038/s41592-025-02712-4

work page doi:10.1038/s41592-025-02712-4 2025
[25]

Image Quality Assessment: From Error Visibility to Structural Similarity

Zhou Wang et al. “Image Quality Assessment: From Error Visibility to Structural Similarity”. In: IEEE Transactions on Image Processing13.4 (2004), pp. 600–612

work page 2004
[26]

Multi-Scale Structural Similarity for Image Quality Assessment

Zhou Wang, Eero P. Simoncelli, and Alan C. Bovik. “Multi-Scale Structural Similarity for Image Quality Assessment”. In:The Thirty-Seventh Asilomar Conference on Signals, Systems & Computers. Vol. 2. 2003, pp. 1398–1402. 19

work page 2003
[27]

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric

Richard Zhang et al. “The Unreasonable Effectiveness of Deep Features as a Perceptual Metric”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018, pp. 586– 595

work page 2018
[28]

Scalable Diffusion Models with Transformers

William Peebles and Saining Xie. “Scalable Diffusion Models with Transformers”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 4195–4205

work page 2023
[29]

Sadler and Jiaman Wu and Wei

William Peebles and Saining Xie. “Scalable Diffusion Models with Transformers”. In:Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 2023, pp. 4195–4205.doi: 10.1109/ICCV51070.2023.00387

work page doi:10.1109/iccv51070.2023.00387 2023
[30]

Predicting transcriptional outcomes of novel multigene perturba- tions with GEARS

Chang Qiao et al. “A neural network for long-term super-resolution imaging of live cells with reliable confidence quantification”. In:Nature Biotechnology44.1 (2026), pp. 110–119.doi:10.1038/s41587- 025-02553-8

work page doi:10.1038/s41587- 2026
[31]

Cellpose: A Generalist Algorithm for Cellular Segmentation

Carsen Stringer et al. “Cellpose: A Generalist Algorithm for Cellular Segmentation”. In:Nature Meth- ods18.1 (2021), pp. 100–106

work page 2021
[32]

Cellpose-SAM: superhuman generalization for cellular segmentation

Marius Pachitariu, Michael Rariden, and Carsen Stringer. “Cellpose-SAM: superhuman generalization for cellular segmentation”. In:BioRxiv(2025), pp. 2025–04

work page 2025
[33]

Panoptic Segmentation

Alexander Kirillov et al. “Panoptic Segmentation”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019, pp. 9404–9413

work page 2019
[34]

V-Net: Fully Convolutional Neural Net- works for Volumetric Medical Image Segmentation

Fausto Milletari, Nassir Navab, and Seyed-Ahmad Ahmadi. “V-Net: Fully Convolutional Neural Net- works for Volumetric Medical Image Segmentation”. In:2016 Fourth International Conference on 3D Vision (3DV). IEEE. 2016, pp. 79–87

work page 2016
[35]

A Concordance Correlation Coefficient to Evaluate Reproducibility

Lawrence I-Kuei Lin. “A Concordance Correlation Coefficient to Evaluate Reproducibility”. In:Bio- metrics45.1 (1989), pp. 255–268

work page 1989
[36]

Spectroscopic stimulated Raman scattering imaging of highly dynamic specimens through matrix completion

Haonan Lin et al. “Spectroscopic stimulated Raman scattering imaging of highly dynamic specimens through matrix completion”. In:Light: Science & Applications7.5 (2018), pp. 17179–17179.issn: 2047-7538.doi:10.1038/lsa.2017.179.url:https://doi.org/10.1038/lsa.2017.179

work page doi:10.1038/lsa.2017.179.url:https://doi.org/10.1038/lsa.2017.179 2018
[37]

Super-resolution SRS microscopy with A-PoD

Hongje Jang et al. “Super-resolution SRS microscopy with A-PoD”. In:Nature Methods20.3 (2023), pp. 448–458.issn: 1548-7105.doi:10.1038/s41592-023-01779-1.url:https://doi.org/10.1038/ s41592-023-01779-1

work page doi:10.1038/s41592-023-01779-1.url:https://doi.org/10.1038/ 2023
[38]

Denoising of stimulated Raman scattering microscopy images via deep learning

Bryce Manifold et al. “Denoising of stimulated Raman scattering microscopy images via deep learning”. In:Biomedical Optics Express10.8 (2019), pp. 3860–3874.doi:10.1364/BOE.10.003860.url:http: //www.osapublishing.org/boe/abstract.cfm?URI=boe-10-8-3860

work page doi:10.1364/boe.10.003860.url:http: 2019
[39]

Deep learning-driven super-resolution in Raman hyperspectral imaging: Efficient high-resolution reconstruction from low-resolution data

Md Inzamam Ul Haque et al. “Deep learning-driven super-resolution in Raman hyperspectral imaging: Efficient high-resolution reconstruction from low-resolution data”. In:Applied Physics Letters125.20 (2024), p. 204104.issn: 0003-6951.doi:10.1063/5.0228645.url:https://doi.org/10.1063/5. 0228645

work page doi:10.1063/5.0228645.url:https://doi.org/10.1063/5 2024
[40]

Super-resolution vibrational imaging based on photoswitchable Raman probe

Jingwen Shou et al. “Super-resolution vibrational imaging based on photoswitchable Raman probe”. In:Science Advances9.24 (), eade9118.doi:10.1126/sciadv.ade9118.url:https://doi.org/10. 1126/sciadv.ade9118

work page doi:10.1126/sciadv.ade9118.url:https://doi.org/10
[41]

Label-free super-resolution stimulated Raman scattering imaging of biomedical specimens

Julien Guilbert et al. “Label-free super-resolution stimulated Raman scattering imaging of biomedical specimens”. In:Advanced Imaging1.11 (Jan. 2024), p. 011004.url:https://www.researching.cn/ articles/OJfff2f087cb065461

work page 2024
[42]

4Pi stimulated Raman scattering for label-free super-resolution chemical imaging

Jonathan I. Kim et al. “4Pi stimulated Raman scattering for label-free super-resolution chemical imaging”. In:Science Advances12.1 (), eaec0523.doi:10 . 1126 / sciadv . aec0523.url:https : //doi.org/10.1126/sciadv.aec0523

work page doi:10.1126/sciadv.aec0523
[43]

Mean Flows for One-step Generative Modeling

Zhengyang Geng et al. “Mean Flows for One-step Generative Modeling”. In:The Thirty-ninth Annual Conference on Neural Information Processing Systems. 2025.url:https://openreview.net/forum? id=uWj4s7rMnR

work page 2025
[44]

Generative Modeling via Drifting

Mingyang Deng et al. “Generative Modeling via Drifting”. In:arXiv preprint arXiv:2602.04770(2026). 20

work page internal anchor Pith review arXiv 2026
[45]

Learning a Deep Convolutional Network for Image Super-Resolution

Chao Dong et al. “Learning a Deep Convolutional Network for Image Super-Resolution”. In:European Conference on Computer Vision (ECCV). 2014, pp. 184–199

work page 2014
[46]

SwinIR: Image Restoration Using Swin Transformer

Jingyun Liang et al. “SwinIR: Image Restoration Using Swin Transformer”. In:IEEE/CVF Interna- tional Conference on Computer Vision Workshops (ICCVW). 2021, pp. 1833–1844. 21 Supplementary Figure 1:Point spread function illustration. A, Differences in the point spread function between high-resolution (HR) and low-resolution (LR) objectives.The LR objective...

work page 2021