Sketchpose: Learning to Segment Cells with Partial Annotations

Cl\'ement Cazorla; Nathana\"el Munier; Pierre Weiss; Renaud Morin

REVIEW 3 major objections 3 minor 32 references

Sketchpose: Learning to Segment Cells with Partial Annotations

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Partial annotations can train distance-map cell segmentation networks to the same accuracy as full annotations.

desk verdict Promising practical idea; verdict depends entirely on the experiments, which are not visible in this version. read the letter →

arxiv 2508.17798 v1 pith:VT5DNYKS submitted 2025-08-25 cs.CV

Clément Cazorla , Nathanaël Munier , Renaud Morin , Pierre Weiss This is my paper

classification cs.CV

keywords cellsegmentationdeeplearningdistancemappartialannotationsfrugaltransferNapariinstance

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that cell segmentation networks trained on distance maps do not need fully traced cell outlines. It presents a training procedure that handles partially annotated objects, where the annotator sketches only a portion of each cell and the loss is restricted to the annotated regions. The authors report that this approach yields the same segmentation quality as fully supervised training while saving substantial annotation time, and they demonstrate it in frugal-learning, transfer-learning, and standard settings. The method ships as a Napari plugin, making the reduced annotation burden directly available to biologists.

What carries the argument

The central object is the distance map—a per-pixel map of distance to the nearest cell boundary—which is the output representation used by the most accurate cell segmentation networks. The paper's contribution is a training procedure that makes this distance-map representation learnable from partially annotated objects: the supervised signal is defined so that unannotated regions are treated as missing data rather than as background or as belonging to the cell, allowing the network to train on a mix of complete and incomplete sketches in the same image.

What would settle it

Annotate the same set of images twice—once with full cell outlines and once with partial sketches covering roughly the same total drawing length—then train identical networks and compare detection and boundary accuracy. If the partial-sketch model proves consistently worse on images with densely packed cells or on rare cell morphologies, the claim that partial annotations do not sacrifice segmentation quality would be falsified.

Watch

Extended reading notes

Core claim

The central claim is that a distance-map segmentation network can be trained from partial annotations without sacrificing accuracy. Instead of requiring a complete mask for every cell, the proposed method treats unannotated image regions as missing data and computes the supervised signal only on the annotated parts of each object. The authors show that this partial-annotation training matches fully supervised performance across several learning regimes, meaning that drawing a few strokes per cell rather than tracing entire boundaries is enough to learn the distance-to-boundary representation that drives modern cell segmentation.

Load-bearing premise

The entire savings claim rests on the assumption that a network trained only on partial sketches of cells learns the same distance-to-boundary structure as one trained on complete outlines, and that the skipped regions do not systematically hide the hard cases.

Editorial extensions

If this is right

Biologists can generate training sets by drawing a few strokes per cell instead of tracing entire boundaries, cutting annotation time substantially.
Transfer learning to new microscopy modalities or cell types becomes feasible with only partial sketches, lowering the barrier for domain adaptation.
Segmentation quality remains comparable to fully supervised training, so existing Cellpose-style workflows could adopt the method without performance loss.
The method is embedded in a user-friendly Napari plugin, making it directly usable by non-specialist researchers.

Reading between the lines

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

If partial annotations truly suffice, annotation effort could be cut further by combining the approach with active learning, where the network itself proposes which cells need a sketch next—the paper does not explore this.
The missing-data treatment for distance maps might extend to other instance-segmentation tasks with crowded objects, such as tracking particles or segmenting overlapping fibers, where full outlines are equally expensive.
The claim implies that the bottleneck in training data creation shifts from pixel-level tracing to choosing which cells to annotate; a deliberate sampling of cell diversity could give even better frugal-learning results than random partial annotation.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is a useful practical idea—train distance-map cell segmentation with partial annotations to save labeling time—and the Napari plugin is a real plus. But the version we have is only the abstract and a slice of the introduction, so the central claim ('substantial savings without sacrificing segmentation quality') is untestable from what's in front of us. That doesn't make it wrong; it just means the referee's job is to check the method and evaluation.

The framing is sensible: name Cellpose/StarDist/HoverNet, emphasize the annotation bottleneck, and argue for handling partially drawn objects. That's a real need in bioimage labs. The choice to evaluate across frugal, transfer, and regular learning regimes is the right way to answer the question. Shipping a Napari plugin also raises the chance the method actually gets used.

The soft spots are the usual ones for a methods paper, and I can't tell which are present because the full text is missing. First, the loss formulation: how do they deal with unannotated pixels—ignore them or treat them as background? The abstract says 'partially annotated objects,' but the distance-map targets inside the unannotated parts of a cell are not true distances to the actual boundary; they're distances to the nearest drawn boundary. If the loss ignores those pixels, the distance head is unconstrained there and errors can leak into annotated regions through convolutions. If it includes them, the network learns false targets. The paper needs to state this explicitly and show it doesn't matter. Second—the stress-test note gets this right—real annotations are not random subsets. Annotators will draw easy cells and skip ambiguous ones, so the claimed savings on benchmarks with dense masks may not transfer. That bias can't be checked from the abstract. Third, the visible text doesn't cite prior partial-annotation or weakly-supervised segmentation methods, so I can't tell whether the novelty claim is overreaching.

Verdict: unverdictable on the current evidence, but not because of any visible flaw. The paper deserves a serious referee because the problem is important and the proposed direction is plausible. My recommendation: send it out, but the reviewers should ask for the loss details, the partial-annotation simulation protocol, and a bias-controlled comparison. If the experiments hold up, this is a modest but genuine contribution to the bioimage field.

Referee Report

3 major / 3 minor

Summary. The paper proposes 'Sketchpose', a method for cell segmentation that claims to train distance-map-based networks (e.g., Cellpose, StarDist, HoVer-Net) using partially annotated objects rather than fully annotated datasets. The abstract asserts that the approach yields substantial savings in time and resources without sacrificing segmentation quality, across frugal, transfer, and regular learning settings, and that it is embedded in a Napari plugin. The visible text, however, consists only of the abstract and the beginning of the introduction. No method description, loss formulation, network architecture, dataset details, experimental protocol, results, or quantitative comparisons are provided. The central claim is therefore unverifiable from the submitted content.

Significance. If the claimed performance were substantiated, the contribution would be practically valuable for bioimage analysis, where annotation cost is a major bottleneck. Reducing annotation effort for distance-map-based segmentation without quality loss would have clear utility for transfer learning and frugal learning scenarios. The inclusion of a user-friendly Napari plugin could also lower the barrier for adoption. However, the manuscript as provided does not permit an assessment of the method's validity or its comparative performance. No machine-checked proofs, reproducible code, or quantitative evidence are visible. The significance remains prospective rather than demonstrated.

major comments (3)

[Abstract] The load-bearing claim—'substantial savings in time and resources without sacrificing segmentation quality'—is not supported by any experimental evidence in the visible text. No datasets, baselines, metrics, error bars, or statistical comparisons are reported. Since the entire contribution rests on this empirical assertion, the manuscript currently provides no basis for accepting it.
[§1 Introduction] The core method is not described. The reader is told only that the approach 'still relies on the distance map and handles partially annotated objects,' but no formulation is given for how partial annotations enter the loss, whether unannotated pixels are ignored or treated as background, how distance maps are computed from incomplete masks, or how the network architecture is adapted. Without this description, the approach cannot be evaluated or reproduced.
[§1 Introduction] The manuscript does not address the risk of biased partial annotations. The abstract's frugal/transfer learning claim implicitly assumes that partial annotations are representative of the data distribution. If annotators tend to fully draw easy, isolated cells and skip touching or ambiguous ones, the learned distance maps may be systematically distorted. The text gives no protocol for simulating partial annotations, no controlled-bias experiments, and no theoretical or empirical bound on the quality loss. This gap directly affects the central claim's generalizability.

minor comments (3)

[Abstract] Typographical/formatting issues: 'hinges onfully annotated datasets' should read 'hinges on fully annotated datasets'; 'Stardist' should be 'StarDist'; 'HoverNet' should be 'HoVer-Net'. Please correct these.
[§1 Introduction] The introduction cites several works (Otsu, Kass, Ronneberger, Lempitsky, etc.) but no reference list or bibliography is present in the visible text. A complete reference section is required.
[§1 Introduction] The 'simplified history' is very brief and does not include a dedicated discussion of weakly supervised or partially supervised segmentation methods, which are directly relevant to the proposed approach. Adding such positioning would help readers understand the novelty.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is evaluated empirically against external segmentation quality, not against its own training targets.

full rationale

The paper's central claim is an empirical one: a distance-map segmentation network trained with partially annotated objects can save annotation effort without sacrificing quality. The abstract states this as 'Our experiments show that it can lead to substantial savings in time and resources without sacrificing segmentation quality.' Nothing in the visible text defines a model quantity in terms of the quantity to be predicted, fits a parameter and then re-predicts a closely related value, or imports a uniqueness theorem from the authors' prior work to force a choice. The method relies on a standard distance-map head and changes the training data regime (partial annotations); whether that training signal is effective is an empirical, externally checkable question. The skeptical concern that partial annotations may be systematically biased is a correctness and generalization risk, not a circularity: it does not make the experimental outcome equivalent to the input by construction. Because no specific circular step can be quoted from the provided text, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

No free parameters or invented entities can be found in the visible text. The central claim rests on the domain assumptions above.

assumptions (2)

domain assumption Partial annotations provide enough supervision to learn a correct distance map.
This is the core premise of the method: unannotated pixels must neither corrupt nor underdetermine the distance-map training.
domain assumption Evaluation on 'regular databases' is representative of real user datasets.
The abstract's claim of savings 'without sacrificing segmentation quality' is only meaningful if the tested benchmarks reflect practical cell-segmentation workloads.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Sketchpose: Learning to Segment Cells with Partial Annotations." pith.science (2026). https://pith.science/paper/VT5DNYKS

@misc{pith2026250817798,
  author       = {Pith},
  title        = {Pith review of: Sketchpose: Learning to Segment Cells with Partial Annotations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VT5DNYKS}},
  note         = {Machine review of arXiv:2508.17798}
}

read the original abstract

The most popular networks used for cell segmentation (e.g. Cellpose, Stardist, HoverNet,...) rely on a prediction of a distance map. It yields unprecedented accuracy but hinges on fully annotated datasets. This is a serious limitation to generate training sets and perform transfer learning. In this paper, we propose a method that still relies on the distance map and handles partially annotated objects. We evaluate the performance of the proposed approach in the contexts of frugal learning, transfer learning and regular learning on regular databases. Our experiments show that it can lead to substantial savings in time and resources without sacrificing segmentation quality. The proposed algorithm is embedded in a user-friendly Napari plugin.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages

[1]

Labkit: labeling and segmentation toolkit for big image data

Matthias Arzt, Joran Deschamps, Christopher Schmied, Tobias Pietzsch, Deborah Schmidt, Pavel Tomancak, Robert Haase, and Florian Jug. Labkit: labeling and segmentation toolkit for big image data. Frontiers in computer science, 4: 0 10, 2022

work page 2022
[2]

Deep watershed transform for instance segmentation

Min Bai and Raquel Urtasun. Deep watershed transform for instance segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5221--5229, 2017

work page 2017
[3]

Ilastik: interactive machine learning for (bio) image analysis

Stuart Berg, Dominik Kutra, Thorben Kroeger, Christoph N Straehle, Bernhard X Kausler, Carsten Haubold, Martin Schiegg, Janez Ales, Thorsten Beier, Markus Rudy, et al. Ilastik: interactive machine learning for (bio) image analysis. Nature methods, 16 0 (12): 0 1226--1232, 2019

work page 2019
[4]

Short exposure to cold atmospheric plasma induces senescence in human skin fibroblasts and adipose mesenchymal stromal cells

Marion Bourdens, Yannick Jeanson, Marion Taurand, No \'e mie Juin, Audrey Carri \`e re, Franck Cl \'e ment, Louis Casteilla, Anne-Laure Bulteau, and Val \'e rie Planat-B \'e nard. Short exposure to cold atmospheric plasma induces senescence in human skin fibroblasts and adipose mesenchymal stromal cells. Scientific reports, 9 0 (1): 0 8671, 2019

work page 2019
[5]

A survey on active learning and human-in-the-loop deep learning for medical image analysis

Samuel Budd, Emma C Robinson, and Bernhard Kainz. A survey on active learning and human-in-the-loop deep learning for medical image analysis. Medical Image Analysis, 71: 0 102062, 2021

work page 2021
[6]

Caicedo, Jonathan Roth, Allen Goodman, Tim Becker, Kyle W

Juan C. Caicedo, Jonathan Roth, Allen Goodman, Tim Becker, Kyle W. Karhohs, Matthieu Broisin, Csaba Molnar, Claire McQuin, Shantanu Singh, Fabian J. Theis, and Anne E. Carpenter. Evaluation of deep learning strategies for nucleus segmentation in fluorescence images. Cytometry Part A, 95 0 (9): 0 952--965, 2019. doi:https://doi.org/10.1002/cyto.a.23863. UR...

work page doi:10.1002/cyto.a.23863 2019
[7]

napari: a python multi-dimensional image viewer platform for the research community

Chi-Li Chiu, Nathan Clack, et al. napari: a python multi-dimensional image viewer platform for the research community. Microscopy and Microanalysis, 28 0 (S1): 0 1576--1577, 2022

work page 2022
[8]

Lkcell: Efficient cell nuclei instance segmentation with large convolution kernels

Ziwei Cui, Jingfeng Yao, Lunbin Zeng, Juan Yang, Wenyu Liu, and Xinggang Wang. Lkcell: Efficient cell nuclei instance segmentation with large convolution kernels. arXiv preprint arXiv:2407.18054, 2024

arXiv 2024

Show all 32 references

[9]

Omnipose: a high-precision morphology-independent solution for bacterial cell segmentation

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 0 (11): 0 1438--1448, 2022

2022
[10]

Pannuke: an open pan-cancer histology dataset for nuclei instance segmentation and classification

Johanna Gamper, Navid Alemi Koohbanani, Katarzyna Benet, Adnan Khuram, and Nasir Rajpoot. Pannuke: an open pan-cancer histology dataset for nuclei instance segmentation and classification. In European Congress on Digital Pathology (ECDP), pages 11--19. Springer, 2019. doi:10.1...

2019 doi
[11]

Hover-net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images

Simon Graham, Quoc Dang Vu, Shan E Ahmed Raza, Ayesha Azam, Yee Wah Tsang, Jin Tae Kwak, and Nasir Rajpoot. Hover-net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images. Medical Image Analysis, 58: 0 101563, 2019

2019
[12]

Mask r-cnn

Kaiming He, Georgia Gkioxari, Piotr Doll \'a r, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961--2969, 2017

2017
[13]

o rst, Moritz Rempe, Lukas Heine, Constantin Seibold, Julius Keyl, Giulia Baldini, Selma Ugurel, Jens Siveke, Barbara Gr \

Fabian H \"o rst, Moritz Rempe, Lukas Heine, Constantin Seibold, Julius Keyl, Giulia Baldini, Selma Ugurel, Jens Siveke, Barbara Gr \"u nwald, Jan Egger, et al. Cellvit: Vision transformers for precise cell segmentation and classification. Medical Image Analysis, 94: 0 103143, 2024

2024
[14]

Snakes: Active contour models

Michael Kass, Andrew Witkin, and Demetri Terzopoulos. Snakes: Active contour models. International journal of computer vision, 1 0 (4): 0 321--331, 1988

1988
[15]

Srivastava, R

Jitendra Kumar, R. Srivastava, R. Srivastava, A. Mehrotra, R. Srivastava, R. Srivastava, and R. Srivastava. Deep learning framework for recognition of diabetic retinopathy using monocular color fundus images. Computers in Biology and Medicine, 109: 0 283--293, 2019. doi:10.101...

2019 doi
[16]

Tgf 1--induced recruitment of human bone mesenchymal stem cells is mediated by the primary cilium in a smad3-dependent manner

Marie-No \"e lle Labour, Mathieu Riffault, S ren T Christensen, and David A Hoey. Tgf 1--induced recruitment of human bone mesenchymal stem cells is mediated by the primary cilium in a smad3-dependent manner. Scientific reports, 6 0 (1): 0 35542, 2016

2016
[17]

Morpholibj: integrated library and plugins for mathematical morphology with imagej

David Legland, Ignacio Arganda-Carreras, and Philippe Andrey. Morpholibj: integrated library and plugins for mathematical morphology with imagej. Bioinformatics, 32 0 (22): 0 3532--3534, 2016

2016
[18]

Deep image prior

Victor Lempitsky, Andrea Vedaldi, and Dmitry Ulyanov. Deep image prior. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9446--9454. IEEE, 2018

2018
[19]

Segmentation of nuclei in histopathology images by deep regression of the distance map

Peter Naylor, Marick La \'e , Fabien Reyal, and Thomas Walter. Segmentation of nuclei in histopathology images by deep regression of the distance map. IEEE transactions on medical imaging, 38 0 (2): 0 448--459, 2018

2018
[20]

A threshold selection method from gray-level histograms

Nobuyuki Otsu. A threshold selection method from gray-level histograms. IEEE transactions on systems, man, and cybernetics, 9 0 (1): 0 62--66, 1979

1979
[21]

Cellpose 2.0: how to train your own model

Marius Pachitariu and Carsen Stringer. Cellpose 2.0: how to train your own model. Nature methods, 19 0 (12): 0 1634--1641, 2022

2022
[22]

Torchvf: Vector fields for instance segmentation

Ryan Peters. Torchvf: Vector fields for instance segmentation. 2022

2022
[23]

U-net: Convolutional networks for biomedical image segmentation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention--MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part...

2015
[24]

The facial adipose system: its role in facial aging and approaches to volume restoration

Neil S Sadick, Andrew S Dorizas, Nils Krueger, and Amer H Nassar. The facial adipose system: its role in facial aging and approaches to volume restoration. Dermatologic Surgery, 41: 0 S333--S339, 2015

2015
[25]

microbeseg models, 2022

Tim Scherr, Johannes Seiffarth, Bastian Wollenhaupt, Oliver Neumann, Dietrich Kohlheyer, Hanno Scharr, Katharina N\" o h, and Ralf Mikut. microbeseg models, 2022. URL https://zenodo.org/record/7221151

2022
[26]

Cell detection with star-convex polygons

Uwe Schmidt, Martin Weigert, Coleman Broaddus, and Gene Myers. Cell detection with star-convex polygons. In Medical Image Computing and Computer Assisted Intervention--MICCAI 2018: 21st International Conference, Granada, Spain, September 16-20, 2018, Proceedings, Part II 11, p...

2018
[27]

Mathematical morphology and its applications to image processing, volume 2

Jean Serra and Pierre Soille. Mathematical morphology and its applications to image processing, volume 2. Springer Science & Business Media, 2012

2012
[28]

Cellpose3: one-click image restoration for improved cellular segmentation

Carsen Stringer and Marius Pachitariu. Cellpose3: one-click image restoration for improved cellular segmentation. Nature methods, 22 0 (3): 0 592--599, 2025

2025
[29]

Cellpose: a generalist algorithm for cellular segmentation

Carsen Stringer, Tim Wang, Michalis Michaelos, and Marius Pachitariu. Cellpose: a generalist algorithm for cellular segmentation. Nature methods, 18 0 (1): 0 100--106, 2021

2021
[30]

Training deep learning models for cell image segmentation with sparse annotations

Ko Sugawara. Training deep learning models for cell image segmentation with sparse annotations. BioRxiv, pages 2023--06, 2023

2023
[31]

Watersheds in digital spaces: an efficient algorithm based on immersion simulations

Luc Vincent and Pierre Soille. Watersheds in digital spaces: an efficient algorithm based on immersion simulations. IEEE Transactions on Pattern Analysis & Machine Intelligence, 13 0 (06): 0 583--598, 1991

1991
[32]

Effects of fixation on bacterial cellular dimensions and integrity

Lillian Zhu, Manohary Rajendram, and Kerwyn Casey Huang. Effects of fixation on bacterial cellular dimensions and integrity. iScience, 24 0 (4): 0 102348, 2021. ISSN 2589-0042. doi:https://doi.org/10.1016/j.isci.2021.102348. URL https://www.sciencedirect.com/science/article/pi...

2021

Pith tools

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

[1] [1]

Labkit: labeling and segmentation toolkit for big image data

Matthias Arzt, Joran Deschamps, Christopher Schmied, Tobias Pietzsch, Deborah Schmidt, Pavel Tomancak, Robert Haase, and Florian Jug. Labkit: labeling and segmentation toolkit for big image data. Frontiers in computer science, 4: 0 10, 2022

work page 2022

[2] [2]

Deep watershed transform for instance segmentation

Min Bai and Raquel Urtasun. Deep watershed transform for instance segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5221--5229, 2017

work page 2017

[3] [3]

Ilastik: interactive machine learning for (bio) image analysis

Stuart Berg, Dominik Kutra, Thorben Kroeger, Christoph N Straehle, Bernhard X Kausler, Carsten Haubold, Martin Schiegg, Janez Ales, Thorsten Beier, Markus Rudy, et al. Ilastik: interactive machine learning for (bio) image analysis. Nature methods, 16 0 (12): 0 1226--1232, 2019

work page 2019

[4] [4]

Short exposure to cold atmospheric plasma induces senescence in human skin fibroblasts and adipose mesenchymal stromal cells

Marion Bourdens, Yannick Jeanson, Marion Taurand, No \'e mie Juin, Audrey Carri \`e re, Franck Cl \'e ment, Louis Casteilla, Anne-Laure Bulteau, and Val \'e rie Planat-B \'e nard. Short exposure to cold atmospheric plasma induces senescence in human skin fibroblasts and adipose mesenchymal stromal cells. Scientific reports, 9 0 (1): 0 8671, 2019

work page 2019

[5] [5]

A survey on active learning and human-in-the-loop deep learning for medical image analysis

Samuel Budd, Emma C Robinson, and Bernhard Kainz. A survey on active learning and human-in-the-loop deep learning for medical image analysis. Medical Image Analysis, 71: 0 102062, 2021

work page 2021

[6] [6]

Caicedo, Jonathan Roth, Allen Goodman, Tim Becker, Kyle W

Juan C. Caicedo, Jonathan Roth, Allen Goodman, Tim Becker, Kyle W. Karhohs, Matthieu Broisin, Csaba Molnar, Claire McQuin, Shantanu Singh, Fabian J. Theis, and Anne E. Carpenter. Evaluation of deep learning strategies for nucleus segmentation in fluorescence images. Cytometry Part A, 95 0 (9): 0 952--965, 2019. doi:https://doi.org/10.1002/cyto.a.23863. UR...

work page doi:10.1002/cyto.a.23863 2019

[7] [7]

napari: a python multi-dimensional image viewer platform for the research community

Chi-Li Chiu, Nathan Clack, et al. napari: a python multi-dimensional image viewer platform for the research community. Microscopy and Microanalysis, 28 0 (S1): 0 1576--1577, 2022

work page 2022

[8] [8]

Lkcell: Efficient cell nuclei instance segmentation with large convolution kernels

Ziwei Cui, Jingfeng Yao, Lunbin Zeng, Juan Yang, Wenyu Liu, and Xinggang Wang. Lkcell: Efficient cell nuclei instance segmentation with large convolution kernels. arXiv preprint arXiv:2407.18054, 2024

arXiv 2024