arxiv: 2604.10609 · v1 · submitted 2026-04-12 · 💻 cs.CV · q-bio.QM

Recognition: unknown

Self-supervised Pretraining of Cell Segmentation Models

Kaden Stillwagon , Alexandra Dunnum VandeLoo , Benjamin Magondu , Craig R. Forest

Authors on Pith no claims yet

Pith reviewed 2026-05-10 14:59 UTC · model grok-4.3

classification 💻 cs.CV q-bio.QM

keywords self-supervised learninginstance segmentationmicroscopycell imagesdomain adaptationpretrainingzero-shot performancerepresentation learning

0 comments

The pith

Self-supervised pretraining on unlabeled cell images produces representations better aligned with microscopy than those from natural-image models.

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

The paper aims to establish that continuing self-supervised training on unlabeled cell microscopy images adapts initial representations from a natural-image model to yield better instance segmentation of cells. A sympathetic reader would care because high-quality labeled datasets for cells are scarce, which limits progress in analyzing their spatial and temporal properties, while models pretrained on everyday photos carry object and texture assumptions that do not match microscope images. If the claim holds, abundant unlabeled microscopy data can be used to create models that perform well even when tested on new types of microscope images without extra labeled examples. The work reports concrete gains on a standard cell benchmark and effective transfer to previously unseen microscopy datasets.

Core claim

The framework adapts representations from a natural-image model to microscopy data by performing continued self-supervised training on unlabeled cell images before supervised fine-tuning for instance segmentation. This produces higher segmentation performance on the LIVECell benchmark and demonstrates strong zero-shot performance on three out-of-distribution microscopy datasets compared to direct initialization with weights from natural-image segmentation models.

What carries the argument

Continued self-supervised pretraining on unlabeled cell images to adapt natural-image model representations to the microscopy domain.

Load-bearing premise

That the observed improvements in segmentation performance stem from the continued self-supervised pretraining step creating representations more suitable for microscopy rather than from other elements of the model training or data selection process.

What would settle it

A direct comparison where the continued self-supervised pretraining is omitted but all other training steps and data are kept identical, and the resulting model shows no improvement or worse performance on the cell segmentation benchmark and out-of-distribution tests.

Figures

Figures reproduced from arXiv: 2604.10609 by Alexandra Dunnum VandeLoo, Benjamin Magondu, Craig R. Forest, Kaden Stillwagon.

**Figure 1.** Figure 1: DINOCell Model Architecture. Schematic of the DINOCell model architecture containing a DINOv2- pretrained [14] ViT-B [48] encoder that has been domain-adapted on microscopy images, a lightweight upsampling decoder, and convolutional objective prediction heads. The model outputs are combined and decoded using gradienttracking to produce instance segmentation masks. 3.2.3 Model Objective and Post-Processing… view at source ↗

**Figure 2.** Figure 2: Three Stage Training Process of the DINOCell Model.(a) Stage 1: A ViT-B [48] image encoder is pretrained on 1.2 billion images using DINOv2 [14], producing strong object- and part-aware visual embeddings. DINOCell is initialized with these pretrained weights. (b) Stage 2 (Domain Adaptation): The encoder is further trained with DINOv2 on a curated dataset of 130k unlabeled microscopy images to adapt the rep… view at source ↗

**Figure 3.** Figure 3: Evolution of Image Encoder Representations through DINOCell Training. We compute a PCA between image encoder patches embeddings and show their first three components where each component is mapped to a different color channel (red, green, blue). This is done for four images from different datasets (LIVECell [39], Glioma C6 [42], Cellpose cyto [17], and cpg0024 [29]) and at each stage of DINOCell’s three st… view at source ↗

**Figure 4.** Figure 4: Effects of pretraining (PT) and domain adaptation (DA) on model performance and encoder representations. Across epochs, we find that there is a significant effect of self-supervised pretraining on both model performance and encoder representations, while effects of domain adaptation are less significant. (a) Encoder representations of a single image from the cpg0024 [29] subset of the Cellpainting Gallery… view at source ↗

**Figure 5.** Figure 5: DINOCell per-dataset performance on the test split of the Finetuning Dataset, shown for both (a) aggregate performance, and (b) representative images. (a) Maximum Matching Accuracy (MMA) on the test split of each dataset used for fine-tuning. MMA evaluates how well predicted instances correspond to ground-truth cells (maximum score = 1.0). DINOCell achieves high performance on most datasets, indicating str… view at source ↗

**Figure 6.** Figure 6: DINOCell outperforms Cellpose-SAM , SAMCell-LIVECell, and SAMCell-cyto both visually and quantitatively across all metrics. (a) Qualitative results are shown using segmentation outputs for representative images across the LIVECell [39] and zero-shot datasets. On all data sets, the cells appear visually to be thoroughly and accurately segmented as compared to models from prior publications on the same image… view at source ↗

read the original abstract

Instance segmentation enables the analysis of spatial and temporal properties of cells in microscopy images by identifying the pixels belonging to each cell. However, progress is constrained by the scarcity of high-quality labeled microscopy datasets. Many recent approaches address this challenge by initializing models with segmentation-pretrained weights from large-scale natural-image models such as Segment Anything Model (SAM). However, representations learned from natural images often encode objectness and texture priors that are poorly aligned with microscopy data, leading to degraded performance under domain shift. We propose DINOCell, a self-supervised framework for cell instance segmentation that leverages representations from DINOv2 and adapts them to microscopy through continued self-supervised training on unlabeled cell images prior to supervised fine-tuning. On the LIVECell benchmark, DINOCell achieves a SEG score of 0.784, improving by 10.42% over leading SAM-based models, and demonstrates strong zero-shot performance on three out-of-distribution microscopy datasets. These results highlight the benefits of domain-adapted self-supervised pretraining for robust cell segmentation.

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

DINOCell adapts DINOv2 with extra self-supervised training on cell images before fine-tuning and reports a 10% SEG lift on LIVECell plus zero-shot transfer, but the gains are not isolated from other training choices.

read the letter

The paper's main move is to take DINOv2, run continued self-supervised pretraining on unlabeled cell microscopy images, and then fine-tune for instance segmentation. They get a SEG score of 0.784 on the LIVECell benchmark, which they say is 10.42% above leading SAM-based models, and they show reasonable zero-shot results on three other microscopy datasets. That is the concrete empirical piece they add relative to prior work that just uses SAM weights directly. The approach targets the real issue that natural-image pretraining carries objectness and texture biases that do not match cell data well, and using unlabeled cell images to adapt the backbone is a direct way to address domain shift when labels are scarce. The zero-shot numbers are a useful check on whether the adapted features actually help under distribution shift. The soft spot is exactly the one the stress test flags: the abstract gives no ablation that applies the identical fine-tuning protocol to unmodified DINOv2 weights. Without that control, the reported improvement could come from differences in augmentations, loss, optimizer settings, or even the choice of backbone rather than the domain-adaptive pretraining step itself. Error bars and full training protocol details are also missing from the summary, so the reliability of the 10% figure is hard to judge yet. This is aimed at bioimage analysts and computational biologists who need better starting points for segmentation when labeled cell data is limited. A reader in that group would get a practical idea to try, even if the current evidence is not yet conclusive. It deserves peer review because the problem is well-motivated, the benchmark results are specific, and the missing controls are straightforward to add.

Referee Report

1 major / 3 minor

Summary. The paper introduces DINOCell, a self-supervised framework that initializes from DINOv2 weights and performs continued self-supervised pretraining on unlabeled cell microscopy images before supervised fine-tuning for instance segmentation. It claims a SEG score of 0.784 on the LIVECell benchmark (10.42% improvement over leading SAM-based models) and strong zero-shot performance on three out-of-distribution microscopy datasets, attributing the gains to better alignment of representations with microscopy data via domain-adaptive pretraining.

Significance. If the performance gains are shown to arise specifically from the continued self-supervised pretraining step, the work would provide a useful method for adapting natural-image self-supervised models to microscopy, addressing domain shift and reducing dependence on scarce labeled cell data. Evaluation on an external public benchmark plus OOD zero-shot tests offers a reasonable basis for assessing robustness.

major comments (1)

The central claim attributes the 10.42% SEG improvement and OOD gains to continued self-supervised pretraining on unlabeled cell images. However, no ablation is reported that applies the identical supervised fine-tuning protocol, augmentations, and loss to the unmodified original DINOv2 weights. This control experiment is load-bearing for isolating the contribution of the domain-adaptive pretraining from other training choices or backbone effects.

minor comments (3)

The reported SEG scores lack error bars, standard deviations, or statistical significance tests, which weakens assessment of the reliability of the numeric gains.
The abstract and provided description omit key details of the full training protocol, including hyperparameters, pretraining epochs, dataset sizes, and exact augmentations; these must be expanded in the methods for reproducibility.
References to 'leading SAM-based models' in the abstract are not named or cited; specific model names and citations should appear in the results section for precise comparison.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback and for highlighting an important aspect of our experimental design. We address the major comment below and commit to incorporating the requested control in the revised manuscript.

read point-by-point responses

Referee: The central claim attributes the 10.42% SEG improvement and OOD gains to continued self-supervised pretraining on unlabeled cell images. However, no ablation is reported that applies the identical supervised fine-tuning protocol, augmentations, and loss to the unmodified original DINOv2 weights. This control experiment is load-bearing for isolating the contribution of the domain-adaptive pretraining from other training choices or backbone effects.

Authors: We agree that this control experiment is essential for rigorously isolating the contribution of the continued self-supervised pretraining on unlabeled cell images. The original manuscript emphasizes comparisons to SAM-based models to demonstrate practical gains on the LIVECell benchmark and zero-shot transfer. However, to directly address the referee's point and substantiate the attribution to domain-adaptive pretraining, we will add the requested ablation in the revised version. Specifically, we will fine-tune the unmodified original DINOv2 weights using the identical supervised fine-tuning protocol, augmentations, and loss function employed for DINOCell, and report the resulting SEG score on LIVECell together with zero-shot performance on the three out-of-distribution microscopy datasets. These results will be presented in the experiments section (with an updated table) to allow clear comparison. We believe this addition will strengthen the manuscript without altering our core claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external benchmarks

full rationale

The paper's derivation chain consists of (1) taking DINOv2 weights, (2) continued self-supervised pretraining on unlabeled microscopy images, and (3) supervised fine-tuning for instance segmentation. The load-bearing claims are quantitative improvements on the independent LIVECell benchmark (SEG 0.784, +10.42% over SAM baselines) plus zero-shot results on three separate OOD microscopy datasets. These metrics are computed on held-out test data using standard segmentation metrics and are not defined in terms of the method's own fitted quantities. No equations appear in the provided text; no self-definitional loops, fitted-input-as-prediction, or load-bearing self-citations are present. The attribution of gains specifically to the continued pretraining step is an empirical question (addressed by the skeptic via missing ablations), but that is a question of experimental design, not circularity in the derivation itself. The result is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The approach rests on the standard assumption that self-supervised objectives on unlabeled microscopy images yield transferable features for downstream segmentation; no free parameters or invented physical entities are described in the abstract.

axioms (1)

domain assumption Self-supervised pretraining on unlabeled cell images produces representations better aligned with microscopy than natural-image pretraining
Central premise stated in the abstract for why adaptation helps.

invented entities (1)

DINOCell no independent evidence
purpose: Name for the proposed self-supervised adaptation pipeline
Framework introduced by the authors; no independent evidence outside the paper.

pith-pipeline@v0.9.0 · 5486 in / 1284 out tokens · 53867 ms · 2026-05-10T14:59:13.703862+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

55 extracted references · 15 canonical work pages · 7 internal anchors

[1]

Way, Maria Kost-Alimova, Tsukasa Shibue, William F

Gregory P. Way, Maria Kost-Alimova, Tsukasa Shibue, William F. Harrington, Stanley Gill, Federica Piccioni, Tim Becker, Hamdah Shafqat-Abbasi, William C. Hahn, Anne E. Carpenter, Francisca Vazquez, and Shantanu Singh. Predicting cell health phenotypes using image-based morphology profiling.Molecular Biology of the Cell, 32(9):995–1005, April 2021

2021
[2]

Cole and Danielle Hunt

Richard W. Cole and Danielle Hunt. Light Microscopy as a Tool to Detect Apoptosis and Other Cellular Changes and Damage.Journal of biomolecular techniques: JBT, 36(1):3fc1f5fe.5d696e01, April 2025

2025
[3]

Welter, Sofia Benavides, Trevor K

Emma M. Welter, Sofia Benavides, Trevor K. Archer, Oksana Kosyk, and Anthony S. Zannas. Machine learning- based morphological quantification of replicative senescence in human fibroblasts.GeroScience, 46(2):2425–2439, April 2024

2024
[4]

Morphology-based deep learning enables accurate detection of senescence in mesenchymal stem cell cultures.BMC biology, 22(1):1, January 2024

Liangge He, Mingzhu Li, Xinglie Wang, Xiaoyan Wu, Guanghui Yue, Tianfu Wang, Yan Zhou, Baiying Lei, and Guangqian Zhou. Morphology-based deep learning enables accurate detection of senescence in mesenchymal stem cell cultures.BMC biology, 22(1):1, January 2024

2024
[5]

S. Oja, P. Komulainen, A. Penttilä, J. Nystedt, and M. Korhonen. Automated image analysis detects aging in clinical-grade mesenchymal stromal cell cultures.Stem Cell Research & Therapy, 9(1):6, January 2018

2018
[6]

Pratik Kamat, Nico Macaluso, Chanhong Min, Yukang Li, Anshika Agrawal, Aaron Winston, Lauren Pan, Bartholomew Starich, Teasia Stewart, Pei-Hsun Wu, Jean Fan, Jeremy Walston, and Jude M. Phillip. Single-cell morphology encodes functional subtypes of senescence in aging human dermal fibroblasts.bioRxiv: The Preprint Server for Biology, page 2024.05.10.59363...

2024
[7]

Vasilevich, Steven Vermeulen, Marloes Kamphuis, Nadia Roumans, Said Eroumé, Dennie G

Aliaksei S. Vasilevich, Steven Vermeulen, Marloes Kamphuis, Nadia Roumans, Said Eroumé, Dennie G. A. J. Hebels, Jeroen van de Peppel, Rika Reihs, Nick R. M. Beijer, Aurélie Carlier, Anne E. Carpenter, Shantanu Singh, and Jan de Boer. On the correlation between material-induced cell shape and phenotypical response of human mesenchymal stem cells.Scientific...

2020
[8]

Isar Nassiri and Matthew N. McCall. Systematic exploration of cell morphological phenotypes associated with a transcriptomic query.Nucleic Acids Research, 46(19):e116, November 2018

2018
[9]

Gilkes, Jude M

Pei-Hsun Wu, Daniele M. Gilkes, Jude M. Phillip, Akshay Narkar, Thomas Wen-Tao Cheng, Jorge Marchand, Meng-Horng Lee, Rong Li, and Denis Wirtz. Single-cell morphology encodes metastatic potential.Science Advances, 6(4):eaaw6938, January 2020

2020
[10]

Sykora, Milan Mrksich, and Neda Bagheri

Zeynab Mousavikhamene, Daniel J. Sykora, Milan Mrksich, and Neda Bagheri. Morphological features of single cells enable accurate automated classification of cancer from non-cancer cell lines.Scientific Reports, 11(1):24375, December 2021. 14 Self-supervised Pretraining of Cell Segmentation ModelsA PREPRINT

2021
[11]

Schneider, Wayne S

Caroline A. Schneider, Wayne S. Rasband, and Kevin W. Eliceiri. NIH Image to ImageJ: 25 years of image analysis.Nature Methods, 9(7):671–675, July 2012

2012
[12]

Roerdink and Arnold Meijster

Jos B.T.M. Roerdink and Arnold Meijster. The Watershed Transform: Definition, Algorithms and Parallelization Strategies.Fundamenta Informaticae, 41(1-2):187, January 2000

2000
[13]

Segment Anything

Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollár, and Ross Girshick. Segment Anything, April 2023. arXiv:2304.02643 [cs]

work page internal anchor Pith review arXiv 2023
[14]

DINOv2: Learning Robust Visual Features without Supervision

Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick La...

work page internal anchor Pith review Pith/arXiv arXiv 2024
[15]

U-Net: Convolutional Networks for Biomedical Image Segmentation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-Net: Convolutional Networks for Biomedical Image Segmentation, May 2015. arXiv:1505.04597 [cs]

work page internal anchor Pith review Pith/arXiv arXiv 2015
[16]

U-Net: deep learning for cell counting, detection, and morphometry.Nature Methods, 16(1):67–70, January 2019

Thorsten Falk, Dominic Mai, Robert Bensch, Özgün Çiçek, Ahmed Abdulkadir, Yassine Marrakchi, Anton Böhm, Jan Deubner, Zoe Jäckel, Katharina Seiwald, Alexander Dovzhenko, Olaf Tietz, Cristina Dal Bosco, Sean Walsh, Deniz Saltukoglu, Tuan Leng Tay, Marco Prinz, Klaus Palme, Matias Simons, Ilka Diester, Thomas Brox, and Olaf Ronneberger. U-Net: deep learning...

2019
[17]

Cellpose: a generalist algorithm for cellular segmentation.Nature Methods, 18(1):100–106, January 2021

Carsen Stringer, Tim Wang, Michalis Michaelos, and Marius Pachitariu. Cellpose: a generalist algorithm for cellular segmentation.Nature Methods, 18(1):100–106, January 2021

2021
[18]

Cutler, Carsen Stringer, Teresa W

Kevin J. Cutler, Carsen Stringer, Teresa W. Lo, Luca Rappez, Nicholas Stroustrup, S. Brook Peterson, Paul A. Wiggins, and Joseph D. Mougous. Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation.Nature Methods, 19(11):1438–1448, November 2022

2022
[19]

Segment anything in medical images.Nature Communications, 15(1):654, January 2024

Jun Ma, Yuting He, Feifei Li, Lin Han, Chenyu You, and Bo Wang. Segment anything in medical images.Nature Communications, 15(1):654, January 2024

2024
[20]

Malta, Saahil Sanganeriya, Emilio Aponte, Caitlin van Zyl, Danfei Xu, and Craig Forest

Alexandra Dunnum VandeLoo, Nathan J. Malta, Saahil Sanganeriya, Emilio Aponte, Caitlin van Zyl, Danfei Xu, and Craig Forest. SAMCell: Generalized label-free biological cell segmentation with segment anything.PLOS ONE, 20(9):e0319532, September 2025

2025
[21]

CellSAM: a foundation model for cell segmentation

Markus Marks, Uriah Israel, Rohit Dilip, Qilin Li, Changhua Yu, Emily Laubscher, Ahamed Iqbal, Elora Pradhan, Ada Ates, Martin Abt, Caitlin Brown, Edward Pao, Shenyi Li, Alexander Pearson-Goulart, Pietro Perona, Georgia Gkioxari, Ross Barnowski, Yisong Yue, and David Van Valen. CellSAM: a foundation model for cell segmentation. Nature Methods, 22(12):2585...

2025
[22]

Cellpose-SAM: superhuman generalization for cellular segmentation, May 2025

Marius Pachitariu, Michael Rariden, and Carsen Stringer. Cellpose-SAM: superhuman generalization for cellular segmentation, May 2025. Pages: 2025.04.28.651001 Section: New Results

2025
[23]

org/abs/2111.06377

Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked Autoencoders Are Scalable Vision Learners, December 2021. arXiv:2111.06377 [cs]

work page arXiv 2021
[24]

A Simple Framework for Contrastive Learning of Visual Representations

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A Simple Framework for Contrastive Learning of Visual Representations, July 2020. arXiv:2002.05709 [cs]

work page internal anchor Pith review arXiv 2020
[25]

Emerging properties in self-supervised vision transformers.arXiv preprint arXiv:2104.14294,

Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging Properties in Self-Supervised Vision Transformers, May 2021. arXiv:2104.14294 [cs]

work page arXiv 2021
[26]

UINO-FSS: Unifying Representation Learning and Few-shot Segmentation via Hierarchical Distillation and Mamba-HyperCorrelation, November 2025

Wei Zhuo, Zhiyue Tang, Wufeng Xue, Hao Ding, Junkai Ji, and Linlin Shen. UINO-FSS: Unifying Representation Learning and Few-shot Segmentation via Hierarchical Distillation and Mamba-HyperCorrelation, November 2025. arXiv:2504.15669 [cs]

work page arXiv 2025
[27]

Towards general purpose vision foundation models for medical image analysis: An experimental study of dinov2 on radiology benchmarks.arXiv preprint arXiv:2312.02366, 2023

Mohammed Baharoon, Waseem Qureshi, Jiahong Ouyang, Yanwu Xu, Abdulrhman Aljouie, and Wei Peng. Evaluating General Purpose Vision Foundation Models for Medical Image Analysis: An Experimental Study of DINOv2 on Radiology Benchmarks, September 2024. arXiv:2312.02366 [cs]

work page arXiv 2024
[28]

Beilei Cui, Mobarakol Islam, Long Bai, and Hongliang Ren. Surgical-DINO: adapter learning of foundation models for depth estimation in endoscopic surgery.International Journal of Computer Assisted Radiology and Surgery, 19(6):1013–1020, 2024

2024
[29]

Carpenter, Beth A

Erin Weisbart, Ankur Kumar, John Arevalo, Anne E. Carpenter, Beth A. Cimini, and Shantanu Singh. Cell Painting Gallery: an open resource for image-based profiling.Nature Methods, 21(10):1775–1777, October 2024. 15 Self-supervised Pretraining of Cell Segmentation ModelsA PREPRINT

2024
[30]

Sokolnicki, and Anne E

Vebjorn Ljosa, Katherine L. Sokolnicki, and Anne E. Carpenter. Annotated high-throughput microscopy image sets for validation.Nature Methods, 9(7):637–637, July 2012

2012
[31]

Li, Gabriella Rustici, Aleksandra Tarkowska, Anatole Chessel, Simone Leo, Bálint Antal, Richard K

Eleanor Williams, Josh Moore, Simon W. Li, Gabriella Rustici, Aleksandra Tarkowska, Anatole Chessel, Simone Leo, Bálint Antal, Richard K. Ferguson, Ugis Sarkans, Alvis Brazma, Rafael E. Carazo Salas, and Jason R. Swedlow. Image Data Resource: a bioimage data integration and publication platform.Nature Methods, 14(8):775–781, August 2017

2017
[32]

DIC image reconstruction using an energy minimization framework to visualize optical path length distribution.Scientific Reports, 6:30420, July 2016

Krisztian Koos, József Molnár, Lóránd Kelemen, Gábor Tamás, and Peter Horvath. DIC image reconstruction using an energy minimization framework to visualize optical path length distribution.Scientific Reports, 6:30420, July 2016

2016
[33]

Blood Cell Segmentation Dataset, 2023

Deponker Depto, Shazidur Rahman, Mekayel Hosen, Shapna Akter, Tamanna Reme, Aimon Rahman, Hasib Zunai, Sohel Rahman, and Jeet Lahiri. Blood Cell Segmentation Dataset, 2023

2023
[34]

DeepBacs – Escherichia coli bright field segmentation dataset, October 2021

Christoph Spahn and Mike Heilemann. DeepBacs – Escherichia coli bright field segmentation dataset, October 2021

2021
[35]

DeepBacs – Escherichia coli release from stationary phase - Bright field segmentation dataset and StarDist model, March 2022

Christoph Spahn and Mike Heilemann. DeepBacs – Escherichia coli release from stationary phase - Bright field segmentation dataset and StarDist model, March 2022

2022
[36]

DeepBacs – Staphylococcus aureus widefield segmentation dataset, October 2021

Pedro Matos Pereira and Mariana Pinho. DeepBacs – Staphylococcus aureus widefield segmentation dataset, October 2021

2021
[37]

Greenwald, David Van Valen, Erin Weisbart, Beth A Cimini, Trevor Cheung, Oscar Brück, Gary D

Jun Ma, Ronald Xie, Shamini Ayyadhury, Cheng Ge, Anubha Gupta, Ritu Gupta, Song Gu, Yao Zhang, Gihun Lee, Joonkee Kim, Wei Lou, Haofeng Li, Eric Upschulte, Timo Dickscheid, José Guilherme de Almeida, Yixin Wang, Lin Han, Xin Yang, Marco Labagnara, V ojislav Gligorovski, Maxime Scheder, Sahand Jamal Rahi, Carly Kempster, Alice Pollitt, Leon Espinosa, Tam M...

2022
[38]

A convolutional neural network segments yeast microscopy images with high accuracy

Nicola Dietler, Matthias Minder, V ojislav Gligorovski, Augoustina Maria Economou, Denis Alain Henri Lucien Joly, Ahmad Sadeghi, Chun Hei Michael Chan, Mateusz Kozi´nski, Martin Weigert, Anne-Florence Bitbol, and Sahand Jamal Rahi. A convolutional neural network segments yeast microscopy images with high accuracy. Nature Communications, 11(1):5723, November 2020

2020
[39]

Jackson, Nabeel Khalid, Nicola Bevan, Timothy Dale, Andreas Dengel, Sheraz Ahmed, Johan Trygg, and Rickard Sjögren

Christoffer Edlund, Timothy R. Jackson, Nabeel Khalid, Nicola Bevan, Timothy Dale, Andreas Dengel, Sheraz Ahmed, Johan Trygg, and Rickard Sjögren. LIVECell—A large-scale dataset for label-free live cell segmentation. Nature Methods, 18(9):1038–1045, September 2021

2021
[40]

McCourty, Berardo M

Tianhao Zhang, Heather J. McCourty, Berardo M. Sanchez-Tafolla, Anton Nikolaev, and Lyudmila S. Mihaylova. MorphoSeg: An uncertainty-aware deep learning method for biomedical segmentation of complex cellular morphologies.Neurocomputing, 647:130511, September 2025

2025
[41]

Smith, Fred A

Athul Vijayan, Tejasvinee Atul Mody, Qin Yu, Adrian Wolny, Lorenzo Cerrone, Soeren Strauss, Miltos Tsiantis, Richard S. Smith, Fred A. Hamprecht, Anna Kreshuk, and Kay Schneitz. A deep learning-based toolkit for 3D nuclei segmentation and quantitative analysis in cellular and tissue context.Development, 151(14):dev202800, July 2024

2024
[42]

Glioma C6: A Novel Dataset for Training and Benchmarking Cell Segmentation, November

Roman Malashin, Svetlana Pashkevich, Daniil Ilyukhin, Arseniy V olkov, Valeria Yachnaya, Andrey Denisov, and Maria Mikhalkova. Glioma C6: A Novel Dataset for Training and Benchmarking Cell Segmentation, November
[43]

arXiv:2511.07286 [cs]

work page arXiv
[44]

Reith, Jannik Franzen, Dinesh R

Fabian H. Reith, Jannik Franzen, Dinesh R. Palli, J. Lorenz Rumberger, and Dagmar Kainmueller. SelfAdapt: Unsupervised Domain Adaptation of Cell Segmentation Models, August 2025. arXiv:2508.11411 [cs]

work page arXiv 2025
[45]

A systematic evaluation of computational methods for cell segmentation.Briefings in Bioinformatics, 25(5):bbae407, September 2024

Yuxing Wang, Junhan Zhao, Hongye Xu, Cheng Han, Zhiqiang Tao, Dawei Zhou, Tong Geng, Dongfang Liu, and Zhicheng Ji. A systematic evaluation of computational methods for cell segmentation.Briefings in Bioinformatics, 25(5):bbae407, September 2024

2024
[46]

Jackson, Johan Trygg, Rickard Sjögren, Andreas Dengel, and Sheraz Ahmed

Nabeel Khalid, Fabian Schmeisser, Mohammadmahdi Koochali, Mohsin Munir, Christoffer Edlund, Timothy R. Jackson, Johan Trygg, Rickard Sjögren, Andreas Dengel, and Sheraz Ahmed. Point2Mask: A Weakly Supervised Approach for Cell Segmentation Using Point Annotation. In Guang Yang, Angelica Aviles-Rivero, Michael Roberts, and Carola-Bibiane Schönlieb, editors,...

2022
[47]

High-Throughput Low-Cost Segmentation of Brightfield Microscopy Live Cell Images, August 2025

Surajit Das, Gourav Roy, and Pavel Zun. High-Throughput Low-Cost Segmentation of Brightfield Microscopy Live Cell Images, August 2025. arXiv:2508.14106 [q-bio] version: 2. 16 Self-supervised Pretraining of Cell Segmentation ModelsA PREPRINT

work page arXiv 2025
[48]

Contrast limited adaptive histogram equalization

Karel Zuiderveld. Contrast limited adaptive histogram equalization. InGraphics gems IV, pages 474–485. Academic Press Professional, Inc., USA, August 1994

1994
[49]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, June 2021. arXiv:2010.11929 [cs]

work page internal anchor Pith review Pith/arXiv arXiv 2021
[50]

CNN-watershed: A watershed transform with predicted markers for corneal endothelium image segmentation.Biomedical Signal Processing and Control, 68:102805, July 2021

Adrian Kucharski and Anna Fabija ´nska. CNN-watershed: A watershed transform with predicted markers for corneal endothelium image segmentation.Biomedical Signal Processing and Control, 68:102805, July 2021

2021
[51]

iBOT: Image BERT Pre-Training with Online Tokenizer

Jinghao Zhou, Chen Wei, Huiyu Wang, Wei Shen, Cihang Xie, Alan Yuille, and Tao Kong. iBOT: Image BERT Pre-Training with Online Tokenizer, January 2022. arXiv:2111.07832 [cs]

work page internal anchor Pith review arXiv 2022
[52]

Decoupled Weight Decay Regularization

Ilya Loshchilov and Frank Hutter. Decoupled Weight Decay Regularization, January 2019. arXiv:1711.05101 [cs]

work page internal anchor Pith review Pith/arXiv arXiv 2019
[53]

Balak, Pavel Karas, Tereza Bolcková, Markéta Štreitová, Craig Carthel, Stefano Coraluppi, Nathalie Harder, Karl Rohr, Klas E

Martin Maška, Vladimír Ulman, David Svoboda, Pavel Matula, Petr Matula, Cristina Ederra, Ainhoa Urbiola, Tomás España, Subramanian Venkatesan, Deepak M.W. Balak, Pavel Karas, Tereza Bolcková, Markéta Štreitová, Craig Carthel, Stefano Coraluppi, Nathalie Harder, Karl Rohr, Klas E. G. Magnusson, Joakim Jaldén, Helen M. Blau, Oleh Dzyubachyk, Pavel Kˇrížek, ...

2014
[54]

Sorokin, Petr Matula, Carlos Ortiz-de Solórzano, and Michal Kozubek

Pavel Matula, Martin Maška, Dmitry V . Sorokin, Petr Matula, Carlos Ortiz-de Solórzano, and Michal Kozubek. Cell Tracking Accuracy Measurement Based on Comparison of Acyclic Oriented Graphs.PLOS ONE, 10(12):e0144959, December 2015

2015
[55]

Model Comparison

I.T. Jolliffe. Principal Component Analysis | Springer Nature Link, 2002. 6 Supplementary Materials 6.1 Fine-tuning Dataset Descriptions BBBC030 [ 32]: The BBBC030 dataset contains 60 images of Chinese Hamster Ovary (CHO) cells. The images were taken on a DIC microscope with a 20x magnification. The dataset is available here: https://bbbc.broadinstitute.o...

work page arXiv 2002