pith. machine review for the scientific record. sign in

arxiv: 2605.06266 · v1 · submitted 2026-05-07 · 💻 cs.CV

Recognition: unknown

ZScribbleSeg: A comprehensive segmentation framework with modeling of efficient annotation and maximization of scribble supervision

Ke Zhang , Bomin Wang , Hangqi Zhou , Xiahai Zhuang
Authors on Pith no claims yet

Pith reviewed 2026-05-08 13:30 UTC · model grok-4.3

classification 💻 cs.CV
keywords scribble supervisionweakly supervised segmentationmedical image segmentationEM algorithmspatial regularizationshape constraintsannotation efficiencypseudo-label correction
0
0 comments X

The pith

Scribble annotations alone reach competitive accuracy in medical image segmentation by optimizing their placement and adding spatial shape constraints.

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

The paper sets out to demonstrate that full pixel-by-pixel labels are not required for strong medical segmentation performance. It first identifies scribble patterns that give the most supervision per stroke by maximizing information coverage while simulating realistic variation. Regularization terms then enforce spatial neighborhood consistency and overall shape plausibility, with an EM procedure estimating the proportion of each tissue class to decide labels for the many unlabeled pixels and to fix prediction errors. If successful, this approach lets segmentation models train on far less expert time while still producing usable results on heart MRI, abdominal CT, brain tumors, and prostate scans.

Core claim

We propose ZScribbleSeg that first derives efficient scribble forms through supervision maximization and randomness simulation. We add regularization terms that encode spatial relationships and shape constraints, using the EM algorithm to estimate mixture ratios of label classes. These ratios identify unlabeled pixels for each class and correct erroneous predictions, allowing the integrated framework to deliver competitive segmentation on six medical datasets using only scribble supervision.

What carries the argument

ZScribbleSeg framework that combines maximized scribble supervision with EM-estimated class mixture ratios to apply spatial and shape regularization.

If this is right

  • Segmentation training becomes feasible with only sparse scribble strokes instead of dense pixel labels across cardiac, abdominal, and tumor datasets.
  • Unlabeled regions receive class assignments based on estimated mixture ratios, which reduces errors that arise from incomplete supervision.
  • Shape and spatial regularization produces segmentations that respect anatomical continuity rather than isolated pixel decisions.
  • The same pipeline applies without modification to six distinct segmentation challenges including ACDC, BTCV, and Decathlon tasks.

Where Pith is reading between the lines

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

  • Annotation budgets in radiology departments could shrink dramatically if models need only a few strokes per image.
  • The same mixture-ratio correction idea might improve other weak-supervision settings where partial labels must be propagated.
  • An active-learning loop could be built on top by letting the model suggest the most informative scribble locations for the next human annotation round.

Load-bearing premise

The regularization terms accurately reflect true spatial relationships and shape constraints, and the EM algorithm reliably estimates class mixture ratios without any full ground-truth labels available.

What would settle it

On the ACDC dataset, the Dice scores produced by ZScribbleSeg trained only on scribbles fall more than a few percentage points below the scores of a standard fully supervised model trained on complete labels.

Figures

Figures reproduced from arXiv: 2605.06266 by Bomin Wang, Hangqi Zhou, Ke Zhang, Xiahai Zhuang.

Figure 1
Figure 1. Figure 1: Roadmap of the proposed ZScribbleSeg framework. 2.1. Weakly supervised segmentation Recently, a variety of weakly supervised segmentation strate￾gies have been developed to reduce the manual annotation ef￾forts (Lin et al., 2024; Yang et al., 2024; Kweon et al., 2023; Liu et al., 2025b). Among them, the scribbles are of particular interest for the application to medical image annotation, given by their adv… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the training losses for the proposed ZScribbleNet, which consists of modeling of efficient scribbles, computation of priors and shape regularization term. The scribble modeling includes mixup augmentation, regularized with global consistency (Lglobal). The prior comprises class mixture ratios (π) and spatial prior. These components contribute to the spatial prior loss (Lspatial) and the shape r… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of supervision augmentation and global consistency. Supervision maximization is achieved with the mix augmentation to increase the annotated proportion and data variety. Global consistency requires the segmentation result of mixed image and unmixed image to be consistent. Introduce randomness via occlusion: We propose to simu￾late randomly distributed scribbles via occlusion. Specifically, one… view at source ↗
Figure 4
Figure 4. Figure 4: Two under segmentation examples indicated by yellow arrows: (a) foreground classes (left and right ventricles) in ACDC dataset; (b) back￾ground in MyoPS dataset. Considering that the annotated ratio of classes can be imbal￾anced, the scribble-supervised learning also brings challenges to the estimation of class mixture ratios π. Imbalanced classes: Firstly, the number of pixels of different label class (de… view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the spatial prior loss (Lspatial) for correction of prediction, via class mixture ratios (π) and the spatial prior (with spatial energy). 2001; McLachlan and Krishnan, 2007) to iteratively estimate π. In the E-step, we compute the posterior responsibility of class ck for each unlabeled pixel xi using the current estimate π [t] . Let pˆℓ(ck | xi) denote the network prediction (posterior) and… view at source ↗
Figure 6
Figure 6. Figure 6: Performance of segmentation networks trained using partial cross entropy computed from Points scribble form with different number of pixels Npix, with comparisons to fully supervised models (FullSupUNet): (a) and (c) visualize Dice scores with respect to different Npix on ACDC and MyoPS, respectively. The performance of models trained by the Random walk form, with increasing step length l, compared with mo… view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of cardiac segmentation on ACDC dataset. The two slices were from the median and the worst cases by the average Dice scores of all compared methods. change the shape of target structure. Therefore, it can be diffi￾cult for the segmentation model to learn the shape prior, leading to the HD increase in some structures. This problem is allevi￾ated when combined with the global consistency loss, … view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of cardiac segmentation on LGE CMR using MSCMRseg dataset. The two slices were from the median and the worst cases by the average Dice scores of all compared methods view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of irregular segmentation of myocardial pathologies on MyoPS dataset. The two slices were from the median cases by average Dice scores of edema or scar segmentation of all compared methods view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of results on Prostate dataset. The selected subjects were the best, median and worst cases by the average Dice scores of all compared methods. 4.4.3. Irregular pathology segmentation For segmentation of objects with heterogeneous shape fea￾tures, it becomes particularly challenging to learn the accu￾rate shape information for inference. We evaluated ZScribble￾Seg on such challenging task of… view at source ↗
Figure 11
Figure 11. Figure 11: Sensitivity analysis of key hyperparameters on ACDC and MSCMRseg. We vary the spatial prior weight λ2 and the neighborhood radius r, while keeping all other settings fixed. median, and the worst results selected by the average dice score of all compared methods. ZScribbleSeg generated segmenta￾tions with more realistic shape and size than the other scribble based methods, and this advantage was particular… view at source ↗
read the original abstract

Curating fully annotated datasets for medical image segmentation is labour-intensive and expertise-demanding. To alleviate this problem, prior studies have explored scribble annotations for weakly supervised segmentation. Existing solutions mainly compute losses on annotated areas and generate pseudo labels by propagating annotations to adjacent regions. However, these methods often suffer from inaccurate and unrealistic segmentations due to insufficient supervision and incomplete shape information. In contrast, we first investigate the principle of good scribble annotations, which leads to efficient scribble forms via supervision maximization and randomness simulation. We further introduce regularization terms to encode the spatial relationship and the shape constraints, where the EM algorithm is utilized to estimate the mixture ratios of label classes. These ratios are critical in identifying the unlabeled pixels for each class and correcting erroneous predictions, thus the accurate estimation lays the foundation for the incorporation of spatial prior. Finally, we integrate the efficient scribble supervision with the prior into a framework, referred to as ZScribbleSeg, and apply it to multiple scenarios. Leveraging only scribble annotations, ZScribbleSeg achieves competitive performance on six segmentation tasks including ACDC, MSCMRseg, BTCV, MyoPS, Decathlon-BrainTumor and Decathlon-Prostate. Our code will be released via https://github.com/DLwbm123/ZScribbleSeg.

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.

Referee Report

2 major / 1 minor

Summary. The paper introduces ZScribbleSeg, a weakly supervised segmentation framework for medical images that relies solely on scribble annotations. It first derives principles for efficient scribble forms via supervision maximization and randomness simulation, then adds regularization terms encoding spatial relationships and shape constraints. An EM algorithm estimates class mixture ratios to identify and correct unlabeled pixels, which are incorporated as spatial priors. The framework is evaluated on six tasks (ACDC, MSCMRseg, BTCV, MyoPS, Decathlon-BrainTumor, Decathlon-Prostate), claiming competitive performance.

Significance. If the results hold, the work addresses a practical bottleneck in medical imaging by minimizing annotation effort while maintaining segmentation quality through principled regularization and EM-based pseudo-labeling. The explicit modeling of scribble efficiency and the promised code release would support reproducibility and adoption in data-scarce clinical settings.

major comments (2)
  1. [Abstract] Abstract: The central claim that ZScribbleSeg 'achieves competitive performance' on the six listed datasets is unsupported by any quantitative metrics, baseline comparisons, error bars, or ablation results. This absence prevents verification of whether the regularization and EM steps deliver the asserted gains over prior scribble methods.
  2. [Method (regularization and EM steps)] The EM-based estimation of mixture ratios (in the regularization and EM steps): Because training uses only scribbles, the EM step has no access to ground-truth class proportions. No direct validation of estimation accuracy (e.g., comparison to oracle ratios, sensitivity analysis, or ablation removing the EM correction) is provided, yet this estimation is load-bearing for the pseudo-label correction and spatial-prior incorporation that underpin the competitive-performance claim.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by including at least one key quantitative result (e.g., Dice score on ACDC) to allow readers to gauge the claimed competitiveness immediately.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful for the referee's insightful comments, which have helped us identify areas for improvement in our manuscript. Below, we provide point-by-point responses to the major comments and outline the revisions we plan to make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that ZScribbleSeg 'achieves competitive performance' on the six listed datasets is unsupported by any quantitative metrics, baseline comparisons, error bars, or ablation results. This absence prevents verification of whether the regularization and EM steps deliver the asserted gains over prior scribble methods.

    Authors: We agree that the abstract would be strengthened by including quantitative support for the performance claims. In the revised manuscript, we will update the abstract to report key metrics such as mean Dice scores on the six datasets (ACDC, MSCMRseg, BTCV, MyoPS, Decathlon-BrainTumor, Decathlon-Prostate), along with brief comparisons to prior scribble-supervised baselines. This will allow readers to directly assess the contributions of the regularization and EM components. revision: yes

  2. Referee: [Method (regularization and EM steps)] The EM-based estimation of mixture ratios (in the regularization and EM steps): Because training uses only scribbles, the EM step has no access to ground-truth class proportions. No direct validation of estimation accuracy (e.g., comparison to oracle ratios, sensitivity analysis, or ablation removing the EM correction) is provided, yet this estimation is load-bearing for the pseudo-label correction and spatial-prior incorporation that underpin the competitive-performance claim.

    Authors: The EM procedure iteratively estimates class mixture ratios by treating the model's soft predictions on unlabeled pixels as observations, conditioned on the scribble annotations and spatial constraints; it does not require ground-truth proportions. While the paper demonstrates the value of this step through overall results and indirect ablations, we acknowledge the absence of direct validation such as oracle comparisons or explicit sensitivity analysis. In the revision, we will add a dedicated ablation removing the EM correction and a sensitivity study on the estimated ratios to substantiate its role. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a framework that first studies principles of good scribble annotations to derive efficient forms via supervision maximization, then adds regularization terms encoding spatial relationships and shape constraints with an EM algorithm to estimate class mixture ratios for pseudo-label correction. These steps are presented as applications of standard EM and regularization principles to scribble data rather than self-definitional loops. No equations equate a prediction directly to a fitted input by construction, no load-bearing self-citations are invoked to justify uniqueness, and no known results are merely renamed. Performance claims rest on empirical results across six external datasets, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on domain assumptions about scribble quality and the ability of regularization plus EM to recover accurate labels from partial annotations; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (3)
  • domain assumption Efficient scribble forms can be derived from principles of supervision maximization and randomness simulation
    Stated as the first investigation leading to the annotation strategy.
  • domain assumption Regularization terms can encode spatial relationships and shape constraints sufficiently to correct predictions on unlabeled pixels
    Used to incorporate prior knowledge without full labels.
  • domain assumption The EM algorithm can accurately estimate mixture ratios of label classes from scribble data
    Critical for identifying unlabeled pixels and fixing erroneous predictions.

pith-pipeline@v0.9.0 · 5546 in / 1343 out tokens · 52516 ms · 2026-05-08T13:30:09.332139+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

136 extracted references · 13 canonical work pages · 3 internal anchors

  1. [1]

    2020 , journal=

    ACCL: Adversarial constrained-CNN loss for weakly supervised medical image segmentation , author=. 2020 , journal=

    2020

  2. [2]

    Advances in neural information processing systems , volume=

    A probabilistic u-net for segmentation of ambiguous images , author=. Advances in neural information processing systems , volume=

  3. [3]

    2019 , pages=

    Fong, Ruth and Patrick, Mandela and Vedaldi, Andrea , booktitle= ICCV, title =. 2019 , pages=

    2019

  4. [4]

    2017 , pages=

    Fong Ruth, Patrick M, Vedaldi A , booktitle= ICCV, title =. 2017 , pages=

    2017

  5. [5]

    Neural computation , volume=

    Training with noise is equivalent to Tikhonov regularization , author=. Neural computation , volume=. 1995 , publisher=

    1995

  6. [6]

    2006 , publisher=

    Pattern Recognition and machine learning , author=. 2006 , publisher=

    2006

  7. [7]

    International Conference on Machine learning , pages=

    Semi-supervised classification based on classification from positive and unlabeled data , author=. International Conference on Machine learning , pages=. 2017 , publisher=

    2017

  8. [8]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Minimizing estimated risks on unlabeled data: A new formulation for semi-supervised medical image segmentation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2022 , publisher=

    2022

  9. [9]

    International Conference on Machine learning , volume=

    Adjusting the outputs of a classifier to new a priori probabilities may significantly improve classification accuracy: evidence from a multi-class problem in remote sensing , author=. International Conference on Machine learning , volume=

  10. [10]

    2007 , publisher=

    The EM algorithm and extensions , author=. 2007 , publisher=

    2007

  11. [11]

    Computing , volume=

    A shortest augmenting path algorithm for dense and sparse linear assignment problems , author=. Computing , volume=. 1987 , publisher=

    1987

  12. [12]

    Journal of the society for industrial and applied mathematics , volume=

    Algorithms for the assignment and transportation problems , author=. Journal of the society for industrial and applied mathematics , volume=. 1957 , publisher=

    1957

  13. [13]

    IEEE Transactions on pattern analysis and machine intelligence , volume=

    Fast approximate energy minimization via graph cuts , author=. IEEE Transactions on pattern analysis and machine intelligence , volume=. 2001 , publisher=

    2001

  14. [14]

    nature , volume=

    Deep learning , author=. nature , volume=. 2015 , publisher=

    2015

  15. [15]

    arXiv preprint arXiv:1904.08128 , year=

    Automated design of deep learning methods for biomedical image segmentation , author=. arXiv preprint arXiv:1904.08128 , year=

    work page arXiv 1904

  16. [16]

    Nature Methods , volume=

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation , author=. Nature Methods , volume=. 2021 , publisher=

    2021

  17. [17]

    Medical Image Analysis , volume=

    Multi-scale patch and multi-modality atlases for whole heart segmentation of MRI , author=. Medical Image Analysis , volume=. 2016 , publisher=

    2016

  18. [18]

    Medical Image Analysis , volume=

    HELPNet: Hierarchical perturbations consistency and entropy-guided ensemble for scribble supervised medical image segmentation , author=. Medical Image Analysis , volume=. 2025 , publisher=

    2025

  19. [19]

    Machine Vision and Applications , volume=

    Volumetric medical image segmentation via scribble annotations and shape priors , author=. Machine Vision and Applications , volume=

  20. [20]

    International conference on machine learning , pages=

    Puzzle mix: Exploiting saliency and local statistics for optimal mixup , author=. International conference on machine learning , pages=

  21. [21]

    International Conference on Learning Representations , year=

    Co-Mixup: Saliency Guided Joint Mixup with Supermodular Diversity , author=. International Conference on Learning Representations , year=

  22. [22]

    Improved Regularization of Convolutional Neural Networks with Cutout

    Improved Regularization of Convolutional Neural Networks with Cutout , author=. arXiv preprint arXiv:1708.04552 , year=

    work page internal anchor Pith review arXiv

  23. [23]

    International Conference on Computer Vision , pages=

    CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features , author=. International Conference on Computer Vision , pages=

  24. [24]

    International Conference on Learning Representations , year=

    mixup: Beyond Empirical Risk Minimization , author=. International Conference on Learning Representations , year=

  25. [25]

    Exploring simple siamese representation learning

    Xinlei Chen and Kaiming He , title =. arXiv preprint arXiv:2011.10566 , year =

    work page arXiv 2011

  26. [26]

    Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning

    Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , author =. arXiv preprint arXiv:2006.07733 , year =

    work page arXiv 2006

  27. [27]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Cardiac segmentation from LGE MRI using deep neural network incorporating shape and spatial priors , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2019 , publisher=

    2019

  28. [28]

    Medical Image Analysis , volume=

    Segmentation only uses sparse annotations: Unified weakly and semi-supervised learning in medical images , author=. Medical Image Analysis , volume=. 2022 , publisher=

    2022

  29. [29]

    IEEE Transactions on Medical Imaging , volume=

    Learning to segment from scribbles using multi-scale adversarial attention gates , author=. IEEE Transactions on Medical Imaging , volume=. 2021 , publisher=

    2021

  30. [30]

    Nature communications , volume=

    The medical segmentation decathlon , author=. Nature communications , volume=. 2022 , publisher=

    2022

  31. [31]

    and Full, Peter M

    Bernard, Olivier and Lalande, Alain and Zotti, Clement and Cervenansky, Frederick and Yang, Xin and Heng, Pheng-Ann and Cetin, Irem and Lekadir, Karim and Camara, Oscar and Gonzalez Ballester, Miguel Angel and Sanroma, Gerard and Napel, Sandy and Petersen, Steffen and Tziritas, Georgios and Grinias, Elias and Khened, Mahendra and Kollerathu, Varghese Alex...

  32. [32]

    Computers in biology and medicine , volume=

    Data augmentation for medical imaging: A systematic literature review , author=. Computers in biology and medicine , volume=. 2023 , publisher=

    2023

  33. [33]

    European Conference on Computer Vision , pages=

    Sumix: Mixup with semantic and uncertain information , author=. European Conference on Computer Vision , pages=

  34. [34]

    Medical Image Analysis , volume=

    Nuclei segmentation with point annotations from pathology images via self-supervised learning and co-training , author=. Medical Image Analysis , volume=. 2023 , publisher=

    2023

  35. [35]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Multivariate mixture model for myocardial segmentation combining multi-source images , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2019 , publisher=

    2019

  36. [36]

    International Conference on Machine Learning , pages=

    The benefits of mixup for feature learning , author=. International Conference on Machine Learning , pages=

  37. [37]

    IEEE Transactions on Medical Imaging , volume=

    Scribformer: Transformer makes cnn work better for scribble-based medical image segmentation , author=. IEEE Transactions on Medical Imaging , volume=. 2024 , publisher=

    2024

  38. [38]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Weakly supervised medical image segmentation via superpixel-guided scribble walking and class-wise contrastive regularization , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

  39. [39]

    Miccai multi-atlas labeling beyond the cranial vault--workshop and challenge , author=. Proc. MICCAI Multi-Atlas Labeling Beyond Cranial Vault—Workshop Challenge , volume=

  40. [40]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    EFFDNet: A Scribble-Supervised Medical Image Segmentation Method with Enhanced Foreground Feature Discrimination , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

  41. [41]

    Multivariate Mixture Model for Myocardial Segmentation Combining Multi-Source Images , year=

    Zhuang, Xiahai , journal=. Multivariate Mixture Model for Myocardial Segmentation Combining Multi-Source Images , year=

  42. [42]

    Medical Image Analysis , volume=

    Weakly supervised segmentation of retinal layers on OCT images with AMD using uncertainty prototype and boundary regression , author=. Medical Image Analysis , volume=. 2025 , publisher=

    2025

  43. [43]

    Medical Image Analysis , volume=

    The Medical Segmentation Decathlon , author=. Medical Image Analysis , volume=. 2025 , publisher=

    2025

  44. [44]

    Medical image analysis , volume=

    Anomaly-guided weakly supervised lesion segmentation on retinal OCT images , author=. Medical image analysis , volume=. 2024 , publisher=

    2024

  45. [45]

    IEEE Transactions on Medical Imaging , year=

    FedLPPA: Learning personalized prompt and aggregation for federated weakly-supervised medical image segmentation , author=. IEEE Transactions on Medical Imaging , year=

  46. [46]

    Medical Image Computing and Computer Assisted Intervention , pages=

    Multivariate Mixture Model for Cardiac Segmentation from Multi-Sequence MRI , author=. Medical Image Computing and Computer Assisted Intervention , pages=

  47. [47]

    Dice , journal =

    Lee R. Dice , journal =. Measures of the Amount of Ecologic Association Between Species , volume =

  48. [48]

    International Workshop on Statistical Atlases and Computational Models of the Heart , pages=

    An exploration of 2D and 3D deep learning techniques for cardiac MR image segmentation , author=. International Workshop on Statistical Atlases and Computational Models of the Heart , pages=. 2017 , publisher=

    2017

  49. [49]

    European Conference on Computer Vision , pages=

    On regularized losses for weakly-supervised cnn segmentation , author=. European Conference on Computer Vision , pages=

  50. [50]

    Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support Workshop , year=

    Learning to Segment Medical Images with Scribble-Supervision Alone , author=. Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support Workshop , year=

  51. [51]

    IEEE Transactions on Medical Imaging , year=

    Post-DAE: Anatomically Plausible Segmentation via Post-Processing With Denoising Autoencoders , author=. IEEE Transactions on Medical Imaging , year=

  52. [52]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2017 , publisher=

    2017

  53. [53]

    IEEE International Conference on Computer Vision , pages=

    Conditional random fields as recurrent neural networks , author=. IEEE International Conference on Computer Vision , pages=

  54. [54]

    International Conference on Machine learning , pages=

    Manifold mixup: Better representations by interpolating hidden states , author=. International Conference on Machine learning , pages=. 2019 , publisher=

    2019

  55. [55]

    International Conference on Information Processing in Medical Imaging , pages=

    Semi-supervised and task-driven data augmentation , author=. International Conference on Information Processing in Medical Imaging , pages=. 2019 , publisher=

    2019

  56. [56]

    Distilling the Knowledge in a Neural Network

    Distilling the knowledge in a neural network , author=. arXiv preprint arXiv:1503.02531 , year=

    work page internal anchor Pith review arXiv

  57. [57]

    IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Focalmix: Semi-supervised learning for 3d medical image detection , author=. IEEE Conference on Computer Vision and Pattern Recognition , pages=

  58. [58]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Recurrent neural networks for aortic image sequence segmentation with sparse annotations , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2018 , publisher=

    2018

  59. [59]

    Pattern Anal

    The PASCAL visual object classes challenge 2012 (VOC2012) development kit , author=. Pattern Anal. Stat. Model. Comput. Learn., Tech. Rep , volume=

    2012

  60. [60]

    IEEE International Conference on Computer Vision , pages=

    Pedestrian parsing via deep decompositional network , author=. IEEE International Conference on Computer Vision , pages=

  61. [61]

    IEEE transactions on medical imaging , volume=

    Weakly supervised deep nuclei segmentation using partial points annotation in histopathology images , author=. IEEE transactions on medical imaging , volume=. 2020 , publisher=

    2020

  62. [62]

    IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Scribblesup: Scribble-supervised convolutional networks for semantic segmentation , author=. IEEE Conference on Computer Vision and Pattern Recognition , pages=

  63. [63]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Scribble-based hierarchical weakly supervised learning for brain tumor segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2019 , publisher=

    2019

  64. [64]

    IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Semi-supervised semantic segmentation with cross-consistency training , author=. IEEE Conference on Computer Vision and Pattern Recognition , pages=

  65. [65]

    arXiv preprint arXiv:1610.02242 , year=

    Temporal ensembling for semi-supervised learning , author=. arXiv preprint arXiv:1610.02242 , year=

    work page arXiv

  66. [66]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results , author=. arXiv preprint arXiv:1703.01780 , year=

    work page Pith review arXiv

  67. [67]

    IEEE International Conference on Computer Vision , pages=

    Unpaired image-to-image translation using cycle-consistent adversarial networks , author=. IEEE International Conference on Computer Vision , pages=

  68. [68]

    IEEE International Conference on Computer Vision , pages=

    Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation , author=. IEEE International Conference on Computer Vision , pages=

  69. [69]

    Medical Image Analysis , volume=

    Embracing imperfect datasets: A review of deep learning solutions for medical image segmentation , author=. Medical Image Analysis , volume=. 2020 , publisher=

    2020

  70. [70]

    Neuroimage , volume=

    User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability , author=. Neuroimage , volume=. 2006 , publisher=

    2006

  71. [71]

    Pattern Recognition , volume=

    Weakly supervised segmentation of COVID19 infection with scribble annotation on CT images , author=. Pattern Recognition , volume=. 2022 , publisher=

    2022

  72. [72]

    IEEE International Conference on Computer Vision , pages=

    Constrained convolutional neural networks for weakly supervised segmentation , author=. IEEE International Conference on Computer Vision , pages=

  73. [73]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results , volume =

    Tarvainen, Antti and Valpola, Harri , journal =. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results , volume =

  74. [74]

    IEEE International Conference on Computer Vision , pages=

    Semi supervised semantic segmentation using generative adversarial network , author=. IEEE International Conference on Computer Vision , pages=

  75. [75]

    IEEE transactions on pattern analysis and machine intelligence , year=

    Semi-supervised semantic segmentation with high-and low-level consistency , author=. IEEE transactions on pattern analysis and machine intelligence , year=

  76. [76]

    IEEE transactions on pattern analysis and machine intelligence , year=

    Stc: A simple to complex framework for weakly-supervised semantic segmentation , author=. IEEE transactions on pattern analysis and machine intelligence , year=

  77. [77]

    IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Image-to-image translation with conditional adversarial networks , author=. IEEE Conference on Computer Vision and Pattern Recognition , pages=

  78. [78]

    IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Normalized cut loss for weakly-supervised cnn segmentation , author=. IEEE Conference on Computer Vision and Pattern Recognition , pages=

  79. [79]

    Advances in Neural Information Processing Systems , volume=

    Mixture Proportion Estimation and PU Learning: A Modern Approach , author=. Advances in Neural Information Processing Systems , volume=

  80. [80]

    Advances in Neural Information Processing Systems , volume=

    Positive-Unlabeled Learning with Non-Negative Risk Estimator , author=. Advances in Neural Information Processing Systems , volume=

Showing first 80 references.