Pith. sign in

REVIEW 2 major objections 5 minor 87 references

A binary whole-lesion prior plus inter-class orthogonality turns image-level labels into mutually exclusive multiclass lesion maps in neuroimaging.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 04:09 UTC pith:APRFL2OK

load-bearing objection Solid, well-evidenced extension of their 2025 MICCAI idea: binary structural guidance plus explicit orthogonality really does help on boundaries and rare subtypes. the 2 major comments →

arxiv 2607.03203 v1 pith:APRFL2OK submitted 2026-07-03 eess.IV

Mutually Exclusive Multiclass Lesion Segmentation in Neuroimaging: Binary-Guided Weak Supervision with Inter-Class Orthogonality

Ashutosh Kumar , Vivek Dhamale , Vaanathi Sundaresan This is my paper
classification eess.IV
keywords weakly supervised segmentationneuroimagingmulticlass segmentationlesion detectionclass activation mapsinter-class orthogonalitybinary-guided refinementpseudo-label refinement
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

When lesion subtypes co-occur and share similar intensity, weakly supervised methods produce overlapping activations and miss rare classes because nothing forces pixels to pick one class. BiMEx-MS first finds the whole lesion with a class-frequency-agnostic binary map, then assigns subtypes only inside that region using multi-scale class maps, a class-specific aggregator, and a three-part loss that separates each class from background, makes class features orthogonal, and keeps the multiclass union aligned with the binary prior. Hierarchical morphological cleanup then yields exclusive pseudo-labels for a final segmenter. On brain-tumor MRI and intracranial-hemorrhage CT, against sixteen baselines, the method is the only one with edema HD95 under 40 mm and reaches strong Dice on rare subtypes such as subdural hemorrhage, with gains largest on boundaries. Backbone swaps and cross-dataset tests indicate that the structural prior, not network size, drives the result.

Core claim

Reliable multiclass weakly supervised segmentation of co-occurring neuroimaging lesions is achieved by decomposing the problem into whole-lesion localization and exclusive class assignment: a binary module supplies a frequency-agnostic structural prior that confines multiclass competition, while a multi-exit CAM pipeline with contrastive pretraining, class-specific attention aggregation, and a tri-partite loss (per-class separation, inter-class orthogonality, binary–multiclass consensus) plus hierarchical morphological refinement produces mutually exclusive maps that outperform prior WSSS methods especially on boundary metrics and rare subtypes.

What carries the argument

BiMEx-MS: binary-guided dual streams where F_bin_ag gates multiclass multi-scale CAMs into Ω+, aggregated by a class-specific attention network and trained with the tri-partite loss L_c + L_sep + L_agree, then refined morphologically into exclusive pseudo-labels.

Load-bearing premise

The binary module trained only on the image-level union of all lesion classes must still produce a complete enough lesion mask that covers both common and rare or multifocal subtypes so later exclusivity steps have something to assign.

What would settle it

Train and evaluate with the binary prior ablated or deliberately incomplete on multifocal or rare subtypes; if multiclass Dice and HD95 then match or beat the full model on edema and rare hemorrhage classes, the claim that structural guidance is necessary collapses.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Image-level labels alone can yield pixel-exclusive multiclass lesion masks usable for clinical volumetry when a whole-lesion prior is enforced.
  • Boundary metrics and rare-subtype performance become the natural stress tests for multiclass WSSS, not only aggregate Dice.
  • Cross-scanner and cross-continental transfer improve when localization is anchored to a class-agnostic binary domain rather than subtype-specific cues.
  • Uncertainty maps concentrate on inter-class boundaries inside the lesion, supporting confidence-aware review of subtype assignment.
  • Long-tailed lesion subtype sets can be expanded class-by-class without catastrophic collapse if exclusivity and the binary prior are maintained.

Where Pith is reading between the lines

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

  • The same binary-then-exclusive split may transfer to other co-occurring medical structures (e.g., infarct core vs penumbra, or multi-compartment cardiac lesions) where intensity contrast is weak.
  • If foundation-model boundary seeds replace the learned binary CAM, the framework could reduce dependence on contrastive pretraining while keeping the exclusivity losses.
  • Volumetric (3D) extension is the clearest next test: inter-slice consistency could either strengthen Ω+ or expose failures of independent 2D binary priors.
  • Prospective radiologist reading of high-entropy inter-class zones would test whether the reported uncertainty is clinically actionable, not only anatomically plausible.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript proposes BiMEx-MS, a weakly supervised multiclass lesion segmentation framework for neuroimaging that decomposes the problem into binary whole-lesion localization and exclusive subtype assignment. A binary multi-exit module produces a class-frequency-agnostic structural prior F_bin_ag that gates multiclass CAMs (Eq. 8); multi-scale class-discriminative maps from a multi-exit classifier with supervised contrastive pretraining are aggregated by a class-specific attention network; exclusivity is enforced by a tri-partite loss (per-class separation L_c, inter-class orthogonality L_sep, binary-multiclass agreement L_agree; Eqs. 13–17); hierarchical morphological refinement then yields pseudo-labels for a final segmentation network. Evaluation spans BraTS 2020/2023 SSA (MRI) and RSNA-ICH o BHSD (CT) against sixteen WSSS baselines, with backbone sweeps, progressive class expansion, three ablations, cross-dataset transfer, and epistemic/aleatoric uncertainty analysis. The central claim is that structural guidance, not model capacity, drives the observed gains, especially on boundary metrics and rare subtypes (Edema HD95 29.56 mm; SDH Dice 0.704).

Significance. If the results hold, the work is a solid and practically useful contribution to multiclass medical WSSS. Explicit mutual exclusivity under image-level labels is a genuine clinical need for co-occurring, adjacent lesion subtypes (tumor core/edema; hemorrhage subtypes), and the paper supplies a clean architectural decomposition, a geometrically motivated loss, and unusually thorough empirical support: sixteen baselines, four datasets, six backbones, progressive expansion, and uncertainty maps. Public code and the disclosed extension of the authors’ 2025 proof-of-concept further strengthen reproducibility. The concentration of gains on HD95/ASSD and rare classes, together with the backbone-capacity ablation, makes a credible case that the binary prior and orthogonality constraints are the operative ingredients rather than encoder size.

major comments (2)
  1. The load-bearing assumption that the binary prior F_bin_ag (Section 3.2.1, Eqs. 1–3, 8) is sufficiently complete and class-frequency-agnostic for rare and multifocal subtypes is only indirectly supported. Ablation I (Fig. 5) shows that removing binary guidance produces the largest drop, and progressive expansion (Table 5) keeps EPH competitive, but the manuscript never reports a direct coverage metric (e.g., recall of rare-class voxels by the thresholded binary prior, or false-negative rate of Ω+ on multifocal cases). A short quantitative check of binary-prior completeness on the rarest classes would close this gap and make the central claim more robust.
  2. Loss weights (λ_c=1, λ_sep=1, λ_agree=5) and thresholds (τ_bin, τ_c, τ_conf) are stated as fixed empirically (Section 4.2) without sensitivity analysis. While standard in the subfield, L_agree is weighted five times higher than the other terms; a brief sensitivity sweep (or at least reporting performance under equal weights) would confirm that the reported gains are not an artifact of this particular hyperparameter choice.
minor comments (5)
  1. Notation for the multi-exit activations F^(l,c) and the refined maps ˜F^(l,c) is dense; a short glossary or consistent subscript convention would help readers navigate Sections 3.2–3.3.
  2. Figure 4 qualitative panels are informative but the color legend for ICH subtypes (EPH/IPH/IVH/SDH) is only mentioned in the caption; adding an in-figure legend would improve readability.
  3. The hierarchical morphological refinement (Section 3.4) prioritizes the class with smallest extent for disputed pixels; a one-sentence justification or reference for this heuristic would be useful.
  4. Table 1 reports slice counts; adding the corresponding patient/volume counts next to the train/val/test rows would make subject-level isolation easier to verify at a glance.
  5. A few minor typos appear (e.g., “HausdorffDistance” missing space; “corresponing” in Section 4.1). A light proofreading pass is recommended.

Circularity Check

0 steps flagged

Empirical methods paper with disclosed self-citation of a prior proof-of-concept; no derivation reduces by construction to its inputs.

full rationale

BiMEx-MS is an empirical weakly-supervised segmentation framework evaluated on external public benchmarks (BraTS 2020/2023 SSA, RSNA-ICH→BHSD) against sixteen independent baselines. The tri-partite loss (Lc, Lsep, Lagree; Eqs. 13–17) and binary-guided refinement (Eq. 8) are design choices whose necessity is tested by ablations (Fig. 5, Table 6), not algebraic identities that force the reported Dice/HD95 numbers. The sole self-citation (Dhamale & Sundaresan 2025) is openly framed as a minimal proof-of-concept that the present paper substantially extends with hierarchical morphological refinement, full segmentation training, multi-modal evaluation, and backbone/uncertainty analyses; it is not invoked as a uniqueness theorem that forbids alternatives or as a hidden definition of the result. Loss weights (λc=1, λsep=1, λagree=5) and thresholds are hand-chosen by trial-and-error, which is standard for the subfield and does not constitute fitting a parameter then re-labeling a related quantity as a prediction. Progressive class expansion, cross-dataset transfer, and six-backbone consistency further show that the central claim (structural guidance drives performance) is externally falsifiable rather than self-referential. No self-definitional loop, fitted-input-as-prediction, uniqueness import, or ansatz smuggling is present. Score 1 reflects only the minor, disclosed self-citation that is not load-bearing for the quantitative claims.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 3 invented entities

The central claim rests on standard deep-learning practice plus a small set of domain assumptions about lesion mutual exclusivity and a handful of hand-chosen scalars. No new physical entities are postulated; the invented pieces are architectural modules and loss terms whose only evidence is the empirical tables.

free parameters (4)
  • λ_c, λ_sep, λ_agree = 1, 1, 5
    Scalar weights of the tri-partite loss; fixed empirically by trial-and-error at 1, 1, 5 (Section 4.2). Central exclusivity claim depends on these relative strengths.
  • τ_bin, τ_c, τ_conf
    Global and class-specific thresholds plus confidence gate used to hard-binarize CAMs into pseudo-labels (Eqs. 18–19). Directly control pseudo-label quality.
  • β_l exit weights = 0.25, 0.50, 0.75, 1.0
    Monotonic multi-exit classification weights {0.25, 0.50, 0.75, 1.0} (Eq. 5); chosen by design, not learned.
  • focal-loss α_c, γ and contrastive temperature ρ
    Class re-weighting and focusing parameters plus contrastive temperature; affect rare-class representation before aggregation.
axioms (4)
  • domain assumption Foreground lesion subtypes are mutually exclusive at the pixel level (each pixel belongs to at most one foreground class).
    Stated as the mutual-exclusivity constraint on Φ (Section 3.1). Clinically reasonable for the chosen BraTS and ICH taxonomies but not universally true for all pathologies.
  • domain assumption Image-level multi-hot labels plus the MIL paradigm are sufficient to train useful multi-scale CAMs when combined with the binary prior.
    Standard WSSS assumption (Section 2.1); the paper’s contribution is the structural prior that makes it work for co-occurring subtypes.
  • ad hoc to paper The binary union label y_bin yields a class-frequency-agnostic structural prior that covers rare subtypes equally with dominant ones.
    Core design hypothesis of Section 3.2.1; supported by ablations but not independently proven for arbitrary long-tailed distributions.
  • standard math Standard deep-learning optimization (Adam, focal loss, supervised contrastive pretraining, Wide-ResNet-38) converges to the intended geometric configuration of the tri-partite loss.
    Implicit throughout Sections 3–4; no formal convergence proof is claimed.
invented entities (3)
  • Tri-partite exclusivity loss (L_c + L_sep + L_agree) no independent evidence
    purpose: Enforce per-class foreground–background separation, inter-class feature orthogonality, and binary–multiclass spatial consensus under image-level supervision.
    Architectural loss construct; evidence is empirical ablation (Table 6) and t-SNE (Fig. 8), not independent external measurement.
  • Class-specific attention aggregation network A(·; θ_c) no independent evidence
    purpose: Produce per-class multi-scale importance weights instead of a shared AME-CAM-style map.
    Module introduced in Section 3.2.4; utility shown by ablation replacing it with uniform averaging.
  • Hierarchical morphological pseudo-label refinement no independent evidence
    purpose: Enforce strict pixel-wise mutual exclusivity and clean holes/noise before training the final segmentation network.
    Post-processing strategy new to this extension (Section 3.4); evidence is the component ablation in Fig. 5.

pith-pipeline@v1.1.0-grok45 · 36231 in / 3473 out tokens · 38342 ms · 2026-07-12T04:09:09.424578+00:00 · methodology

0 comments
read the original abstract

Weakly supervised segmentation of co-occurring neuroimaging lesion subclasses remains challenging due to overlapping activations, noisy pseudo-labels, and the absence of explicit inter-class exclusivity constraints. We propose BiMEx-MS (Binary-guided Mutually Exclusive Multiclass Segmentation), a framework that decomposes multiclass segmentation into whole-lesion localization and exclusive class assignment: a binary localization module provides a class-frequency-agnostic structural prior confining multiclass predictions within the detected lesion domain, while a multi-exit classification architecture with supervised contrastive pretraining produces multi-scale class-discriminative activation maps aggregated via a class-specific attention network. Inter-class exclusivity is enforced through a tri-partite loss comprising per-class separation, inter-class orthogonality, and binary-multiclass spatial consensus, followed by hierarchical morphological pseudo-label refinement. Evaluated across brain tumor MRI (BraTS 2020, BraTS 2023 SSA) and intracranial hemorrhage CT (RSNA-ICH to BHSD) against sixteen weakly supervised baselines, BiMEx-MS achieves Edema HD95 of 29.56 mm (the only method below 40 mm) and subdural hemorrhage Dice of 0.704, with gains consistently largest on boundary metrics and rare subtypes. Cross-dataset generalization, backbone ablations across six architectures, and uncertainty quantification confirm that structural guidance rather than model capacity drives performance. Code: https://github.com/ashutoshkr45/BiMEx-MS-Neuro.

Figures

Figures reproduced from arXiv: 2607.03203 by Ashutosh Kumar, Vaanathi Sundaresan, Vivek Dhamale.

Figure 1
Figure 1. Figure 1: The proposed BiMEx-MS framework, (a) The binary lesion localization module produces the class-frequency-agnostic structural prior F bin ag via a frozen binary multi-exit classifier and binary aggregation network, (b) Binary-guided refinement, where the structural prior is injected at each scale via element-wise multiplication, confining multiclass competition to the lesion domain Ω+ , (c) The class-specifi… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of BiMEx-MS loss components for multi￾class mutually exclusivity. The component Lc (Eq. (13)) enforces foreground–background separation, Lsep (Eq. (15)) ensures inter￾class feature orthogonality and Lagree (Eq. (16)) provides binary￾multiclass spatial consensus. providing a significant edge over AME-CAM. 3.3. Loss Functions for Mutually Exclusive CAM Learning The aggregated maps F c ag are opt… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of segmentation performance across six encoder backbone architectures on BraTS 2020 and BHSD datasets. Dice [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sample comparison results of BiMEx-MS with state-of-the-art WSSS methods on the full axial field of view of brain slices. Top panel: within-dataset evaluation on BraTS 2020 (a,b) and cross-continental generalization on BraTS 2020→BraTS 2023 SSA (c) for core (red) and edema (blue) classes. Bottom panel: within-dataset evaluation on BHSD (d,e) and cross-dataset generalization for RSNA→BHSD (f) for EPH (dark … view at source ↗
Figure 5
Figure 5. Figure 5: Ablation results showing the effect of framework components on BraTS 2020 and BHSD datasets. Dice ↑ (a), HD95 ↓ (b) and ASSD ↓ (c) shown for the settings: (orange) without binary guidance from F bin ag , (yellow) replacement of the class-specific aggregation network with uniform averaging and (blue) the complete BiMEx-MS framework. Removal of binary guidance causes the most severe degradation, particularly… view at source ↗
Figure 6
Figure 6. Figure 6: Resolution sensitivity analysis on BraTS 2020 and BHSD [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Sample uncertainty maps for BraTS 2020 and BHSD [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: t-SNE visualization of learned feature representations on [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗

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

87 extracted references · 5 linked inside Pith

  1. [1]

    Expert Systems with Applications , volume=

    Efficient active contour model for medical image segmentation and correction based on edge and region information , author=. Expert Systems with Applications , volume=. 2022 , publisher=

    2022

  2. [2]

    IEEE Journal of Biomedical and Health Informatics , volume=

    Symmetry-aware deep learning for cerebral ventricle segmentation with intra-ventricular hemorrhage , author=. IEEE Journal of Biomedical and Health Informatics , volume=. 2022 , publisher=

    2022

  3. [3]

    IEEE transactions on medical imaging , volume=

    Unetpp: Redesigning skip connections to exploit multiscale features in image segmentation , author=. IEEE transactions on medical imaging , volume=. 2019 , publisher=

    2019

  4. [4]

    International conference on medical image computing and computer-assisted intervention , pages=

    Bounding box tightness prior for weakly supervised image segmentation , author=. International conference on medical image computing and computer-assisted intervention , pages=. 2021 , organization=

    2021

  5. [5]

    International Conference on Medical Imaging with Deep Learning , pages=

    Weakly supervised deep nuclei segmentation using points annotation in histopathology images , author=. International Conference on Medical Imaging with Deep Learning , pages=. 2019 , organization=

    2019

  6. [6]

    Medical image analysis , volume=

    Constrained-CNN losses for weakly supervised segmentation , author=. Medical image analysis , volume=. 2019 , publisher=

    2019

  7. [7]

    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

  8. [8]

    Medical Image Analysis , volume=

    Weakly supervised segmentation on neural compressed histopathology with self-equivariant regularization , author=. Medical Image Analysis , volume=. 2022 , publisher=

    2022

  9. [9]

    Computers in Biology and Medicine , volume=

    Application of deep reinforcement learning for spike sorting under multi-class imbalance , author=. Computers in Biology and Medicine , volume=. 2023 , publisher=

    2023

  10. [10]

    ACM Transactions on Intelligent Systems and Technology (TIST) , volume=

    Self-weighted robust LDA for multiclass classification with edge classes , author=. ACM Transactions on Intelligent Systems and Technology (TIST) , volume=. 2020 , publisher=

    2020

  11. [11]

    Medical image analysis , volume=

    Weakly supervised segmentation with cross-modality equivariant constraints , author=. Medical image analysis , volume=. 2022 , publisher=

    2022

  12. [12]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Learning deep features for discriminative localization , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  13. [13]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Dilated residual networks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  14. [14]

    IEEE transactions on image processing , volume=

    Layercam: Exploring hierarchical class activation maps for localization , author=. IEEE transactions on image processing , volume=. 2021 , publisher=

    2021

  15. [15]

    Proceedings of the IEEE international conference on computer vision , pages=

    Grad-cam: Visual explanations from deep networks via gradient-based localization , author=. Proceedings of the IEEE international conference on computer vision , pages=

  16. [16]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    A weakly supervised consistency-based learning method for covid-19 segmentation in ct images , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  17. [17]

    Computers in biology and medicine , volume=

    TSSK-Net: Weakly supervised biomarker localization and segmentation with image-level annotation in retinal OCT images , author=. Computers in biology and medicine , volume=. 2023 , publisher=

    2023

  18. [18]

    Medical image analysis , volume=

    Deep weakly-supervised breast tumor segmentation in ultrasound images with explicit anatomical constraints , author=. Medical image analysis , volume=. 2022 , publisher=

    2022

  19. [19]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    C-cam: Causal cam for weakly supervised semantic segmentation on medical image , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  20. [20]

    Proceedings of the IEEE international conference on computer vision , pages=

    Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation , author=. Proceedings of the IEEE international conference on computer vision , pages=

  21. [21]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Bbam: Bounding box attribution map for weakly supervised semantic and instance segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  22. [22]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Box-driven class-wise region masking and filling rate guided loss for weakly supervised semantic segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  23. [23]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Sparsely annotated semantic segmentation with adaptive gaussian mixtures , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  24. [24]

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

    Ame-cam: Attentive multiple-exit cam for weakly supervised segmentation on mri brain tumor , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2023 , organization=

    2023

  25. [25]

    IEEE Journal of Biomedical and Health Informatics , volume=

    WS-MTST: Weakly supervised multi-label brain tumor segmentation with transformers , author=. IEEE Journal of Biomedical and Health Informatics , volume=. 2023 , publisher=

    2023

  26. [26]

    Computers in Biology and Medicine , volume=

    Weakly supervised learning for multi-class medical image segmentation via feature decomposition , author=. Computers in Biology and Medicine , volume=. 2024 , publisher=

    2024

  27. [27]

    European conference on computer vision , pages=

    What’s the point: Semantic segmentation with point supervision , author=. European conference on computer vision , pages=. 2016 , organization=

    2016

  28. [28]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Seminar learning for click-level weakly supervised semantic segmentation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  29. [29]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Anti-adversarially manipulated attributions for weakly and semi-supervised semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  30. [30]

    European Conference on Computer Vision , pages=

    Semi-supervised semantic segmentation via strong-weak dual-branch network , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  31. [31]

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

    Inter-class Separability Loss for Weakly Supervised Mutually Exclusive Multiclass Segmentation of Brain Tumor Lesions , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2025 , organization=

    2025

  32. [32]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Mctformer+: Multi-class token transformer for weakly supervised semantic segmentation , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2024 , publisher=

    2024

  33. [33]

    2024 , journal=

    Weakly-supervised semantic segmentation with image-level labels: from traditional models to foundation models (2024) , author=. 2024 , journal=

    2024

  34. [34]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    C2am: Contrastive learning of class-agnostic activation map for weakly supervised object localization and semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  35. [35]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Extracting class activation maps from non-discriminative features as well , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  36. [36]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Multi-class token transformer for weakly supervised semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  37. [37]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Class re-activation maps for weakly-supervised semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  38. [38]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Learning pixel-level semantic affinity with image-level supervision for weakly supervised semantic segmentation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  39. [39]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Weakly supervised learning of instance segmentation with inter-pixel relations , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  40. [40]

    Advances in neural information processing systems , volume=

    Supervised contrastive learning , author=. Advances in neural information processing systems , volume=

  41. [41]

    arXiv preprint arXiv:1811.02629 , year=

    Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the BRATS challenge , author=. arXiv preprint arXiv:1811.02629 , year=

    Pith/arXiv arXiv

  42. [42]

    Scientific data , volume=

    Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features , author=. Scientific data , volume=. 2017 , publisher=

    2017

  43. [43]

    IEEE transactions on medical imaging , volume=

    The multimodal brain tumor image segmentation benchmark (BRATS) , author=. IEEE transactions on medical imaging , volume=. 2014 , publisher=

    2014

  44. [44]

    2019 , howpublished =

    Anouk Stein, MD and Carol Wu and Chris Carr and George Shih and Jayashree Kalpathy-Cramer and Julia Elliott and kalpathy and Luciano Prevedello and Marc Kohli, MD and Matt Lungren and Phil Culliton and Robyn Ball and Safwan Halabi MD , title =. 2019 , howpublished =

    2019

  45. [45]

    International Workshop on Machine Learning in Medical Imaging , pages=

    Bhsd: A 3d multi-class brain hemorrhage segmentation dataset , author=. International Workshop on Machine Learning in Medical Imaging , pages=. 2023 , organization=

    2023

  46. [46]

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

    Edge-Semantic Synergy Fusion and Adaptive Noise-Aware for Weakly Supervised Pathological Tissue Segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2025 , organization=

    2025

  47. [47]

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

    Pathmamba: Weakly supervised state space model for multi-class segmentation of pathology images , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  48. [48]

    European Conference on Computer Vision , pages=

    Weakly supervised co-training with swapping assignments for semantic segmentation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  49. [49]

    IEEE Transactions on Medical Imaging , volume=

    HAMIL: High-resolution activation maps and interleaved learning for weakly supervised segmentation of histopathological images , author=. IEEE Transactions on Medical Imaging , volume=. 2023 , publisher=

    2023

  50. [50]

    IEEE Transactions on Image Processing , year=

    Weaktr: Exploring plain vision transformer for weakly-supervised semantic segmentation , author=. IEEE Transactions on Image Processing , year=

  51. [51]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Weakly-supervised semantic segmentation for histopathology images based on dataset synthesis and feature consistency constraint , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  52. [52]

    2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=

    Semantic prompt learning for weakly-supervised semantic segmentation , author=. 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=. 2025 , organization=

    2025

  53. [53]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Sparse multimodal vision transformer for weakly supervised semantic segmentation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  54. [54]

    Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    Foundation model assisted weakly supervised semantic segmentation , author=. Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

  55. [55]

    arXiv preprint arXiv:2210.14618 , year=

    Semformer: semantic guided activation transformer for weakly supervised semantic segmentation , author=. arXiv preprint arXiv:2210.14618 , year=

    Pith/arXiv arXiv

  56. [56]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Perturbed and strict mean teachers for semi-supervised semantic segmentation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  57. [57]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Semi-supervised semantic segmentation with cross pseudo supervision , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  58. [58]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Unimatch v2: Pushing the limit of semi-supervised semantic segmentation , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2025 , publisher=

    2025

  59. [59]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , pages=

    Score-CAM: Score-weighted visual explanations for convolutional neural networks , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , pages=

  60. [60]

    2020 international joint conference on neural networks (IJCNN) , pages=

    Eigen-cam: Class activation map using principal components , author=. 2020 international joint conference on neural networks (IJCNN) , pages=. 2020 , organization=

    2020

  61. [61]

    2016 , eprint=

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning , author=. 2016 , eprint=

    2016

  62. [62]

    2017 , eprint=

    On Calibration of Modern Neural Networks , author=. 2017 , eprint=

    2017

  63. [63]

    Neurocomputing , volume=

    Aleatoric uncertainty estimation with test-time augmentation for medical image segmentation with convolutional neural networks , author=. Neurocomputing , volume=. 2019 , publisher=

    2019

  64. [64]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    Pith/arXiv arXiv

  65. [65]

    IEEE transactions on medical imaging , volume=

    Comparison and evaluation of methods for liver segmentation from CT datasets , author=. IEEE transactions on medical imaging , volume=. 2009 , publisher=

    2009

  66. [66]

    Pattern recognition , volume=

    Wider or deeper: Revisiting the resnet model for visual recognition , author=. Pattern recognition , volume=. 2019 , publisher=

    2019

  67. [67]

    ArXiv , pages=

    The brain tumor segmentation (brats) challenge 2023: Glioma segmentation in sub-saharan africa patient population (brats-africa) , author=. ArXiv , pages=

    2023

  68. [68]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  69. [69]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Densely connected convolutional networks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  70. [70]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Mobilenetv2: Inverted residuals and linear bottlenecks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  71. [71]

    International conference on machine learning , pages=

    Efficientnet: Rethinking model scaling for convolutional neural networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  72. [72]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Swin transformer: Hierarchical vision transformer using shifted windows , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  73. [73]

    2024 , eprint=

    Weakly-Supervised Semantic Segmentation with Image-Level Labels: from Traditional Models to Foundation Models , author=. 2024 , eprint=

    2024

  74. [74]

    Proceedings of the IEEE international conference on computer vision , pages=

    Focal loss for dense object detection , author=. Proceedings of the IEEE international conference on computer vision , pages=

  75. [75]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Class-balanced loss based on effective number of samples , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  76. [76]

    2016 fourth international conference on 3D vision (3DV) , pages=

    V-net: Fully convolutional neural networks for volumetric medical image segmentation , author=. 2016 fourth international conference on 3D vision (3DV) , pages=. 2016 , organization=

    2016

  77. [77]

    International workshop on machine learning in medical imaging , pages=

    Tversky loss function for image segmentation using 3D fully convolutional deep networks , author=. International workshop on machine learning in medical imaging , pages=. 2017 , organization=

    2017

  78. [78]

    arXiv preprint arXiv:1910.09217 , year=

    Decoupling representation and classifier for long-tailed recognition , author=. arXiv preprint arXiv:1910.09217 , year=

    Pith/arXiv arXiv 1910

  79. [79]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Large-scale long-tailed recognition in an open world , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  80. [80]

    arXiv preprint arXiv:2007.07314 , year=

    Long-tail learning via logit adjustment , author=. arXiv preprint arXiv:2007.07314 , year=

    Pith/arXiv arXiv 2007

Showing first 80 references.