Pith. sign in

REVIEW 4 major objections 5 minor 37 references

Imbalanced Medical Image Segmentation with Pixel-dependent Noisy Labels

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a collaborative two-branch framework with curriculum dynamic thresholds and a noise balance loss yields state-of-the-art medical image segmentation under pixel-dependent noisy labels, with more than 3% average Dice…

desk verdict Solid empirical method paper for pixel-dependent noisy labels; the class-adaptive dynamic threshold is the most novel piece, but an anomalous clean upper bound and an alpha/beta reporting contradiction need fixing. read the letter →

arxiv 2501.06678 v1 pith:ZWVK2XCZ submitted 2025-01-12 cs.CV cs.AI

Erjian Guo , Zicheng Wang , Zhen Zhao , Luping Zhou This is my paper · ORCID
classification cs.CVcs.AI
keywords medicalimagesegmentationpixel-dependentlabelnoisecollaborativelearningcurriculumdynamicthresholdbalancelossclassimbalancewithnoisylabels
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

The paper argues that noisy labels in medical image segmentation should be treated as pixel-dependent rather than class-dependent, because annotation errors concentrate on boundaries and small objects, and that existing fixed-threshold denoising worsens class imbalance by dropping minority-class pixels. It proposes CLCS, a two-branch collaborative framework in which the two branches are pushed to disagree via a discrepancy loss, and pixels are voted clean or noisy using both branch predictions and the original label. A curriculum dynamic threshold adapts per class to learning progress, admitting more small-object pixels over time, while a noise balance loss keeps suspected noisy pixels in training instead of discarding them. The authors report state-of-the-art Dice and mIoU on Endovis18 and RIGA under SFDA noise, SFDA plus morphological noise, and real rater noise, with more than 3% average improvement over prior methods.

What carries the argument

The load-bearing object is the Curriculum Dynamic Threshold (CDT), defined class-wise as $T_c(t)= [\hat{\sigma}_c(t)/(2-\hat{\sigma}_c(t))]\tau$, where $\hat{\sigma}_c(t)=\sigma_c(t)/\max_c \sigma_c(t)$ and $\sigma_c(t)$ counts pixels whose predicted confidence exceeds a base threshold $\tau$ and whose predicted class is $c$. This turns the model's own confidence statistics into an adaptive, curriculum-like selection criterion: early in training low thresholds admit more minor-class pixels; as classes are learned, thresholds rise and selection becomes stricter. Around it, the Collaborative Confidence Voting (CCV) requires both branches' high-confidence predictions to equal the original label for a pixel to enter the clean set, and the Noise Balance Loss (NBL) applies confidence-weighted cross-entropy and reverse cross-entropy to the remaining pixels.

What would settle it

Train a model with the proposed selection but reveal the oracle clean masks during evaluation, and compute the clean-label ratio inside the selected clean set per class at several epochs. If the selected set's clean ratio is not consistently higher than the original dataset's, or if the gains over a fixed-threshold variant vanish when per-class noise rates are roughly equal, the curriculum assumption is not what drives the reported improvements.

Watch

Extended reading notes

Core claim

The central discovery is that combining two-branch collaborative voting with a curriculum dynamic threshold and a noise balance loss can separate clean from noisy pixels well enough to improve segmentation under realistic pixel-level noise, even when noise rates differ sharply by class (17.0%, 32.1%, and 44.3% for Shaft, Wrist, and Clasper on Endovis18 SFDA noise). The two branches, kept divergent by minimizing cosine similarity between their feature maps, provide complementary evidence; a pixel is labeled clean only if both branches agree with the original label at high confidence. The per-class threshold is derived from the count of high-confidence predictions for that class, normalized by the maximum class count and shaped by a convex function, so small classes start with lower thresholds and are gradually included. Suspected noisy pixels are not dropped but trained with a confidence-weighted combination of cross-entropy and reverse cross-entropy. On Endovis18 SFDA noise average Dice rises from 64.50% for the best baseline to 67.51%, on RIGA SFDA noise from 85.49% to 89.38%, and on real rater noise from 86.45% to 89.07%.

Load-bearing premise

The central premise is that the number of pixels a class can push above a confidence threshold is a faithful measure of how well that class has been learned, so lowering thresholds for low-count classes lets in clean minority-class pixels rather than just noisy ones.

Editorial extensions

If this is right

  • On surgical instrument segmentation with SFDA noise, the reported average Dice is 67.51% versus 64.50% for the best previous method, with similar gains under SFDA plus morphological noise.
  • On retinal cup and disc segmentation with SFDA noise, the reported average Dice is 89.38% versus 85.49% for the best previous method, and on real rater noise it is 89.07% versus 86.45%.
  • The class-wise dynamic threshold increases most for minority classes (a 35% increase in selected Clasper pixels versus 7% for Shaft on Endovis18), directly countering the class-imbalance skew introduced by fixed-threshold denoising.
  • Because suspected noisy pixels are still used through Noise Balance Loss rather than discarded, the method avoids the information loss that comes from hard clean/noisy cuts, which suggests it will degrade more gracefully as the noise selector becomes less accurate.
  • The framework is end-to-end and evaluates with a single branch, so the improvements do not come from a larger model at test time.

Reading between the lines

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

  • A natural extension would be to apply the same per-class count-based threshold in semi-supervised segmentation, where unlabeled pixels could be pseudo-labeled with a curriculum that begins with easy classes; the paper's $\sigma_c(t)$ statistic is a ready-made confidence-history measure for such a setting.
  • Because the threshold depends on the model's own confidence, I would expect the gains to shrink if noise is spatially correlated with object edges in a way that the model learns confidently wrong; testing on structured boundary noise would clarify the boundary of the claim.
  • The clean-ratio figures suggest the voting mainly helps small classes, so a diagnostic extension would be to report per-class Dice gain against per-class clean-ratio gain, checking whether the minority-class improvement is what drives the overall result.
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.

Referee Report

4 major / 5 minor

Summary. The paper proposes CLCS, a two-branch collaborative learning framework for medical image segmentation with pixel-dependent noisy labels and class imbalance. The method combines (i) a boosted collaborative learning module with a cosine discrepancy loss, (ii) a curriculum dynamic threshold (CDT) that adapts per-class confidence thresholds based on the model's own learning-status proxy, (iii) a collaborative confidence voting (CCV) step that partitions pixels into clean and noisy sets, and (iv) a noise balance loss (NBL) that trains on the noisy set with a confidence-weighted mix of CE and RCE. Experiments on Endovis18 and RIGA under SFDA noise, SFDA+ED noise, and real annotation noise report state-of-the-art Dice/mIoU, with more than 3% average improvement over the best baselines in several settings.

Significance. If the empirical claims hold, the paper addresses a real and practical gap: existing noisy-label segmentation methods often assume class-dependent noise or use fixed thresholds that further imbalance minor classes. The paper's strengths are the fairness-oriented experimental setup (multiple baselines re-run, same backbone, ablations over components, p-values reported) and the release of code. The central risk is whether the curriculum selection mechanism, specifically the learning-status proxy in Eqs. (2)-(4), reliably separates clean from noisy pixels; this mechanism is the engine of the reported gains, and its validation is currently indirect.

major comments (4)
  1. [III-B2, Eqs. (2)-(4) and Fig. 10] The learning-status proxy σ_c(t) counts the model's own confident predictions on noisy labels and is used to set the per-class curriculum thresholds T_c(t). No evidence is provided that σ_c(t) tracks true per-class learning status: Fig. 10 shows that the selected clean set has a higher clean ratio than the original labels, but this is an indirect check, and the fixed-threshold ablation in Sec. IV-F is a weak baseline that cannot distinguish a well-calibrated dynamic schedule from any adaptive adjustment. Because the clean/noise partition drives both L_clean and NBL, this is a load-bearing uncertainty. Please provide a comparison against an oracle threshold using true per-class accuracy, or a direct correlation analysis between σ_c(t) and true per-class accuracy over training.
  2. [Table III] On RIGA, CLCS under SFDA-Noise (89.38 Dice) and under Real-Noise (89.07 Dice) surpasses the Clean Upper Bound (88.52 Dice). The paper does not discuss this surprising result. If the clean upper bound is trained on clean labels with the same split and protocol, the fact that noisy-label training beats it requires an explanation (e.g., a regularization effect of the noise or a difference in how the clean labels are obtained). Without this clarification, the magnitude and interpretation of the reported improvements are unclear.
  3. [IV-A vs. IV-H/I, Tables VII-VIII] The implementation details state that the loss weights α and β are set to 1 and 0.01, respectively, but Table VII and Sec. IV-I report the optimal α as 0.01, and Table VIII and Sec. IV-H report the optimal β as 1. This is a direct internal inconsistency that makes the exact configuration used for the main results unreproducible. Please state the final hyperparameter values for each reported experiment and correct the conflicting sentences.
  4. [Algorithm 1 and Sec. IV-A] The algorithm uses a warmup stage ("if iteration < warmup iteration") but no warmup duration is reported in Sec. IV-A. The warmup length is a free parameter that affects how much early noisy-label fitting occurs and therefore influences the final results. Please specify the warmup length used in all experiments.
minor comments (5)
  1. [IV-F] The fixed-threshold baseline is described as setting the threshold for each class to the proportion of that class's pixels, but it is not explained how a class-pixel-proportion becomes a confidence threshold for sample selection; please clarify the conversion.
  2. [Table IV] In the SFDA-Noise block, the SCE row reports HD95 as "10.96(±10.96)", which is likely a typo (the error bar seems implausible); please verify and correct this value.
  3. [Table IX] The checkmark columns are ambiguous; it is not clear which rows use CCV with the convex mapping and which use CCV without it. Please label the rows or columns explicitly so that the comparison in Sec. IV-J can be verified.
  4. [IV-E] P-values are reported only for a subset of comparisons and the statistical test is not named. Report the p-values for all main comparisons and specify the test (e.g., paired t-test) and number of runs.
  5. [III-C, Eq. (7)] The symbol ℓ_rce is introduced in Eq. (7) but is never defined in the text; please define the reverse cross-entropy loss explicitly or cite the exact equation from [25].

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CLCS's curriculum threshold is self-referential feedback, not a derivation that reduces to its inputs; the central claim is an empirical benchmark comparison.

full rationale

The paper's central claim is an empirical performance comparison on Endovis18 and RIGA, not a derivation from first principles. The Curriculum Dynamic Threshold (CDT) uses the model's own confident-prediction count sigma_c(t) to set per-class thresholds, which is a self-referential training heuristic; however, the paper does not define clean labels in terms of sigma_c(t) and then claim to derive them. Instead, it validates the selected clean set against ground-truth clean labels in Sec. IV-G, reporting higher clean ratios in Fig. 10, and ablates the components in Table IX. No equation in the paper reduces a predicted quantity to a fitted parameter by construction: the segmentation outputs are evaluated on held-out test samples, and the hyperparameters (tau, alpha, beta) are reported with ablation studies. Even if some hyperparameters were tuned on the test set, that would be a correctness or generalization risk, not circularity. The cited prior works are standard references, and the paper does not rely on a load-bearing self-citation or an imported uniqueness theorem. The closest self-referential element is the feedback loop between the model's confidence and its own training targets, which is a known limitation of self-training approaches, not a circular argument. Therefore, no significant circularity is present.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a small set of algorithmic assumptions: the memorization effect ordering, the validity of the confidence-count proxy for per-class learning status, the transfer of SCE-style noise robustness to pixel-level CE/RCE, and the reliability of the two-branch voting rule. The free parameters tau, alpha, beta are selected by test-set ablations, and the warmup duration is unspecified. No new physical or conceptual entities are introduced; all components are algorithmic.

free parameters (4)
  • base confidence threshold tau = 0.90
    Chosen by ablation over {0.80, 0.85, 0.90, 0.95} on both datasets (Table V); best test performance. This is tuning on the test set.
  • noise balance loss weight alpha = 0.01
    Chosen by ablation (Table VII); optimal on Endovis18 SFDA test set.
  • discrepancy loss weight beta = 1
    Chosen by ablation (Table VIII); optimal on Endovis18 SFDA test set. Note: implementation section says beta=0.01, contradicting Table VIII.
  • warmup duration = not specified
    The algorithm requires a warmup stage (Alg. 1) but no epoch count or criterion is given; this is an unspecified free hyperparameter affecting results.
assumptions (5)
  • domain assumption Deep networks first memorize clean patterns before overfitting noisy labels (memorization effect).
    Relied on for the warmup stage and curriculum selection; cited from [5], [26]-[28] in Sec. II-A and III-C. If this ordering fails for pixel-level noise, the warmup and selection could be misled.
  • ad hoc to paper The count of confident predictions per class, sigma_c(t), is a valid proxy for per-class learning status.
    Introduced in Eqs. (2)-(4), Sec. III-B2; this is the load-bearing assumption behind the curriculum dynamic threshold.
  • standard math RCE loss has a similar global minimizer under clean and noisy data and is noise-robust.
    Cited from [25] and used to justify the NBL combination in Sec. III-C; assumes the theoretical guarantee transfers to pixel-level CE/RCE combinations.
  • domain assumption Two branches with a discrepancy loss maintain diverse predictions sufficient for reliable voting.
    Sec. III-B1; if branches collapse or diverge too much, the CCV voting (Eq. 6) becomes unreliable.
  • ad hoc to paper Clean labels are those where both branches are confident and agree with the original label.
    CCV rule, Eq. (6); this operational definition of clean is assumed to correlate with true label correctness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Imbalanced Medical Image Segmentation with Pixel-dependent Noisy Labels." pith.science (2026). https://pith.science/paper/ZWVK2XCZ

@misc{pith2026250106678,
  author       = {Pith},
  title        = {Pith review of: Imbalanced Medical Image Segmentation with Pixel-dependent Noisy Labels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZWVK2XCZ}},
  note         = {Machine review of arXiv:2501.06678}
}
read the original abstract

Accurate medical image segmentation is often hindered by noisy labels in training data, due to the challenges of annotating medical images. Prior research works addressing noisy labels tend to make class-dependent assumptions, overlooking the pixel-dependent nature of most noisy labels. Furthermore, existing methods typically apply fixed thresholds to filter out noisy labels, risking the removal of minority classes and consequently degrading segmentation performance. To bridge these gaps, our proposed framework, Collaborative Learning with Curriculum Selection (CLCS), addresses pixel-dependent noisy labels with class imbalance. CLCS advances the existing works by i) treating noisy labels as pixel-dependent and addressing them through a collaborative learning framework, and ii) employing a curriculum dynamic thresholding approach adapting to model learning progress to select clean data samples to mitigate the class imbalance issue, and iii) applying a noise balance loss to noisy data samples to improve data utilization instead of discarding them outright. Specifically, our CLCS contains two modules: Curriculum Noisy Label Sample Selection (CNS) and Noise Balance Loss (NBL). In the CNS module, we designed a two-branch network with discrepancy loss for collaborative learning so that different feature representations of the same instance could be extracted from distinct views and used to vote the class probabilities of pixels. Besides, a curriculum dynamic threshold is adopted to select clean-label samples through probability voting. In the NBL module, instead of directly dropping the suspiciously noisy labels, we further adopt a robust loss to leverage such instances to boost the performance.

Figures

Figures reproduced from arXiv: 2501.06678 by the authors.

Figure 1
Figure 1. Overview of Collaborative Learning with Curriculum Selection (CLCS). An input image is processed by each of the two network branches individually to generate predictions from distinct viewpoints, facilitated by a discrepancy loss. Leveraging the predictions from the two branches and the original label, the model groups the pixels into a clean set and a noise set by the Curriculum Noisy Label Sample Selection (CNS) m… view at source ↗
Figure 2
Figure 2. Illustration of Curriculum Dynamic Threshold (CDT). The M represents a convex function. the pixel-wise logits with a shape of H ×W ×C. In addition, we obtain the prediction of each branch as yˆ k i = arg maxc p k i and the confidence score as pˆ k i = maxc Sof tmax(p k i ). Our two-branch network aims to generate predictions from distinct feature perspectives for a given input image. During the initial stage of trai… view at source ↗
Figure 3
Figure 3. Visual comparison of different noisy labels. Column 1: origi￾nal images; Column 2: SFDA-Noise labels; Column 3: Noise annotations from Rater 6 (fig. c) and SFDA+ED-Noise labels (fig. g); Column 4: clean segmentation labels. C. Noise Patterns To comprehensively verify the robustness of each method, we conduct experiments with SFDA noise, SFDA combined with morphological changes, and manual annotation noise. For the S… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visual comparison with SFDA-Noise on Endovis18. [Dice/mIoU] are given. E. Performance Comparison In this section, we evaluate the segmentation performance of different methods on two medical image datasets with different types of noise. We also perform ablation studies…
Figure 5
Figure 5. Figure 5: Visual comparison of the segmentation results from different methods. The segmentation results with Real-Noise on the training dataset. The symbol [. / .] denotes [Dice / mIoU] scores. F. Impact of Curriculum Dynamic Threshold If we use a fixed threshold based on the i…
Figure 6
Figure 6. Figure 6: As shown, the threshold of each class exhibits a gradual increase during the initial 50 epochs of training, and becomes stable afterwards. In the early phases of training, the incorporation of a non-linear mapping function helps mitigate abrupt fluctuations due to the …
Figure 7
Figure 7. Figure 7: Number of pixels with different predictions by the two branches during training. Due to the discrepancy loss, half of the total pixels have divergent predictions after 200 epochs of training. noisy labels). Therefore, rclean(c) indicates the percentage of pixels with c…
Figure 9
Figure 9. Figure 9: Evolution of the numbers of selected wrist pixels and selected clasper pixels during training process [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Class-wise clean label ratios in the selected clean sub￾set and the original dataset. The clean label ratio is calculated as rclean(c) = P 1(yclean P =y|(y=c)) 1(y=c) , where yclean represents the clean labels and y represents the training set labels (possibly contain…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 31 canonical work pages

  1. [1]

    Generalizing deep learning for medical image segmentation to unseen domains via deep stacked transformation,

    L. Zhang, X. Wang, D. Yang, T. Sanford, S. Harmon, B. Turkbey, B. J. Wood, H. Roth, A. Myronenko, D. Xu, et al., “Generalizing deep learning for medical image segmentation to unseen domains via deep stacked transformation,” IEEE transactions on medical imaging, vol. 39, no. 7, pp. 2531–2540, 2020

    work page 2020

  2. [2]

    Retinal vessel segmen- tation with skeletal prior and contrastive loss,

    Y . Tan, K.-F. Yang, S.-X. Zhao, and Y .-J. Li, “Retinal vessel segmen- tation with skeletal prior and contrastive loss,” IEEE Transactions on Medical Imaging, vol. 41, no. 9, pp. 2238–2251, 2022

    work page 2022

  3. [3]

    Joint class-affinity loss correction for robust medi- cal image segmentation with noisy labels,

    X. Guo and Y . Yuan, “Joint class-affinity loss correction for robust medi- cal image segmentation with noisy labels,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part IV , pp. 588–598, Springer, 2022

    work page 2022

  4. [4]

    Superpixel-guided iterative learning from noisy labels for medical image segmentation,

    S. Li, Z. Gao, and X. He, “Superpixel-guided iterative learning from noisy labels for medical image segmentation,” in Medical Image Com- puting and Computer Assisted Intervention–MICCAI 2021: 24th Interna- tional Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part I 24 , pp. 525–535, Springer, 2021. 12 IEEE TRANSACTIONS ON MEDI...

    work page 2021

  5. [5]

    A closer look at memorization in deep networks,

    D. Arpit, S. Jastrz˛ ebski, N. Ballas, D. Krueger, E. Bengio, M. S. Kanwal, T. Maharaj, A. Fischer, A. Courville, Y . Bengio,et al., “A closer look at memorization in deep networks,” in International conference on machine learning, pp. 233–242, PMLR, 2017

    work page 2017

  6. [6]

    Does label smoothing mitigate label noise?,

    M. Lukasik, S. Bhojanapalli, A. Menon, and S. Kumar, “Does label smoothing mitigate label noise?,” in International Conference on Ma- chine Learning, pp. 6448–6458, PMLR, 2020

    work page 2020

  7. [7]

    Are anchor points really indispensable in label-noise learning?,

    X. Xia, T. Liu, N. Wang, B. Han, C. Gong, G. Niu, and M. Sugiyama, “Are anchor points really indispensable in label-noise learning?,” Ad- vances in neural information processing systems , vol. 32, 2019

    work page 2019

  8. [8]

    Provably end-to-end label-noise learning without anchor points,

    X. Li, T. Liu, B. Han, G. Niu, and M. Sugiyama, “Provably end-to-end label-noise learning without anchor points,” in International Conference on Machine Learning , pp. 6403–6413, PMLR, 2021

    work page 2021

Show all 37 references
  1. [9]

    Co-teaching: Robust training of deep neural networks with extremely noisy labels,

    B. Han, Q. Yao, X. Yu, G. Niu, M. Xu, W. Hu, I. Tsang, and M. Sugiyama, “Co-teaching: Robust training of deep neural networks with extremely noisy labels,” Advances in neural information processing systems, vol. 31, 2018

    2018

  2. [10]

    Review–a survey of learning from noisy labels,

    X. Liang, X. Liu, and L. Yao, “Review–a survey of learning from noisy labels,” ECS Sensors Plus , vol. 1, no. 2, p. 021401, 2022

    2022

  3. [11]

    Augmentation strategies for learning with noisy labels,

    K. Nishi, Y . Ding, A. Rich, and T. Hollerer, “Augmentation strategies for learning with noisy labels,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 8022–8031, 2021

    2021

  4. [12]

    Iterative learning with open-set noisy labels,

    Y . Wang, W. Liu, X. Ma, J. Bailey, H. Zha, L. Song, and S.-T. Xia, “Iterative learning with open-set noisy labels,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 8688– 8696, 2018

    2018

  5. [13]

    Combating noisy labels by agree- ment: A joint training method with co-regularization,

    H. Wei, L. Feng, X. Chen, and B. An, “Combating noisy labels by agree- ment: A joint training method with co-regularization,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 13726–13735, 2020

    2020

  6. [14]

    Using trusted data to train deep networks on labels corrupted by severe noise,

    D. Hendrycks, M. Mazeika, D. Wilson, and K. Gimpel, “Using trusted data to train deep networks on labels corrupted by severe noise,” Advances in neural information processing systems , vol. 31, 2018

    2018

  7. [15]

    Active bias: Training more accurate neural networks by emphasizing high variance samples,

    H.-S. Chang, E. Learned-Miller, and A. McCallum, “Active bias: Training more accurate neural networks by emphasizing high variance samples,” Advances in Neural Information Processing Systems , vol. 30, 2017

    2017

  8. [16]

    Generalized cross entropy loss for training deep neural networks with noisy labels,

    Z. Zhang and M. Sabuncu, “Generalized cross entropy loss for training deep neural networks with noisy labels,” Advances in neural information processing systems, vol. 31, 2018

    2018

  9. [17]

    A semi-supervised two-stage approach to learning from noisy labels,

    Y . Ding, L. Wang, D. Fan, and B. Gong, “A semi-supervised two-stage approach to learning from noisy labels,” in 2018 IEEE Winter conference on applications of computer vision (WACV), pp. 1215–1224, IEEE, 2018

    2018

  10. [18]

    Semi-supervised semantic segmentation with cross pseudo supervision,

    X. Chen, Y . Yuan, G. Zeng, and J. Wang, “Semi-supervised semantic segmentation with cross pseudo supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 2613–2622, 2021

    2021

  11. [19]

    Semi-supervised semantic segmen- tation with cross-consistency training,

    Y . Ouali, C. Hudelot, and M. Tami, “Semi-supervised semantic segmen- tation with cross-consistency training,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 12674– 12684, 2020

    2020

  12. [20]

    Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis,

    D. Karimi, H. Dou, S. K. Warfield, and A. Gholipour, “Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis,” Medical image analysis , vol. 65, p. 101759, 2020

    2020

  13. [21]

    When source-free domain adaptation meets learning with noisy labels,

    L. Yi, G. Xu, P. Xu, J. Li, R. Pu, C. Ling, A. I. McLeod, and B. Wang, “When source-free domain adaptation meets learning with noisy labels,” arXiv preprint arXiv:2301.13381 , 2023

    2023 arXiv

  14. [22]

    Characterizing label errors: confident learning for noisy-labeled image segmentation,

    M. Zhang, J. Gao, Z. Lyu, W. Zhao, Q. Wang, W. Ding, S. Wang, Z. Li, and S. Cui, “Characterizing label errors: confident learning for noisy-labeled image segmentation,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Li...

    2020

  15. [23]

    How does disagreement help generalization against label corruption?,

    X. Yu, B. Han, J. Yao, G. Niu, I. Tsang, and M. Sugiyama, “How does disagreement help generalization against label corruption?,” in International Conference on Machine Learning, pp. 7164–7173, PMLR, 2019

    2019

  16. [24]

    Flexmatch: Boosting semi-supervised learning with curriculum pseudo labeling,

    B. Zhang, Y . Wang, W. Hou, H. Wu, J. Wang, M. Okumura, and T. Shi- nozaki, “Flexmatch: Boosting semi-supervised learning with curriculum pseudo labeling,” Advances in Neural Information Processing Systems , vol. 34, pp. 18408–18419, 2021

    2021

  17. [25]

    Symmetric cross entropy for robust learning with noisy labels,

    Y . Wang, X. Ma, Z. Chen, Y . Luo, J. Yi, and J. Bailey, “Symmetric cross entropy for robust learning with noisy labels,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 322– 330, 2019

    2019

  18. [26]

    Understanding and improving early stopping for learning with noisy labels,

    Y . Bai, E. Yang, B. Han, Y . Yang, J. Li, Y . Mao, G. Niu, and T. Liu, “Understanding and improving early stopping for learning with noisy labels,” Advances in Neural Information Processing Systems , vol. 34, pp. 24392–24403, 2021

    2021

  19. [27]

    Coresets for robust training of deep neural networks against noisy labels,

    B. Mirzasoleiman, K. Cao, and J. Leskovec, “Coresets for robust training of deep neural networks against noisy labels,” Advances in Neural Information Processing Systems , vol. 33, pp. 11465–11477, 2020

    2020

  20. [28]

    Robustness of accuracy metric and its inspirations in learning with noisy labels,

    P. Chen, J. Ye, G. Chen, J. Zhao, and P.-A. Heng, “Robustness of accuracy metric and its inspirations in learning with noisy labels,” inPro- ceedings of the AAAI Conference on Artificial Intelligence , pp. 11451– 11461, 2021

    2021

  21. [29]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,” IEEE transactions on pattern analysis and machine intelligence , vol. 40, no. 4, pp. 834–848, 2017

    2017

  22. [30]

    2018 robotic scene segmentation challenge,

    M. Allan, S. Kondo, S. Bodenstedt, S. Leger, R. Kadkhodamoham- madi, I. Luengo, F. Fuentes, E. Flouty, A. Mohammed, M. Pedersen, et al. , “2018 robotic scene segmentation challenge,” arXiv preprint arXiv:2001.11190, 2020

    2018 arXiv

  23. [31]

    Isinet: an instance- based approach for surgical instrument segmentation,

    C. González, L. Bravo-Sánchez, and P. Arbelaez, “Isinet: an instance- based approach for surgical instrument segmentation,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Par...

    2020

  24. [32]

    Agreement among ophthalmologists in marking the optic disc and optic cup in fundus images,

    A. Almazroa, S. Alodhayb, E. Osman, E. Ramadan, M. Hummadi, M. Dlaim, M. Alkatee, K. Raahemifar, and V . Lakshminarayanan, “Agreement among ophthalmologists in marking the optic disc and optic cup in fundus images,” International ophthalmology, vol. 37, pp. 701– 717, 2017

    2017

  25. [33]

    2017 robotic instrument segmentation challenge,

    M. Allan, A. Shvets, T. Kurmann, Z. Zhang, R. Duggal, Y .-H. Su, N. Rieke, I. Laina, N. Kalavakonda, S. Bodenstedt, et al., “2017 robotic instrument segmentation challenge,” arXiv preprint arXiv:1902.06426 , 2019

    2017 arXiv

  26. [34]

    Learning calibrated medical image segmentation via multi- rater agreement modeling,

    W. Ji, S. Yu, J. Wu, K. Ma, C. Bian, Q. Bi, J. Li, H. Liu, L. Cheng, and Y . Zheng, “Learning calibrated medical image segmentation via multi- rater agreement modeling,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 12341–12351, 2021

    2021

  27. [35]

    Learning from noisy labels via discrepant collaborative training,

    Y . Han, S. ROY , L. Petersson, and M. Harandi, “Learning from noisy labels via discrepant collaborative training,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pp. 3169–3178, 2020

    2020

  28. [36]

    Adaptive early-learning correction for segmentation from noisy annotations,

    S. Liu, K. Liu, W. Zhu, Y . Shen, and C. Fernandez-Granda, “Adaptive early-learning correction for segmentation from noisy annotations,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2606–2616, 2022

    2022

  29. [37]

    Anti-interference from noisy labels: Mean-teacher-assisted confident learning for medical image segmentation,

    Z. Xu, D. Lu, J. Luo, Y . Wang, J. Yan, K. Ma, Y . Zheng, and R. K.- Y . Tong, “Anti-interference from noisy labels: Mean-teacher-assisted confident learning for medical image segmentation,” IEEE Transactions on Medical Imaging , vol. 41, no. 11, pp. 3062–3073, 2022

    2022

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.