Pith. sign in

REVIEW 4 major objections 4 minor 23 references

Uncertainty-Aware Multi-Expert Knowledge Distillation for Imbalanced Disease Grading

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read UMKD distills several expert models into one compact student and reports the best mean accuracy on imbalanced disease grading.

desk verdict A decent multi-expert KD method for imbalanced disease grading, but the abstract overclaims SOTA and the uncertainty-weighting mechanism is not yet proven to be the cause of the gains. read the letter →

arxiv 2505.00592 v1 pith:TBB5N3OD submitted 2025-05-01 cs.CV cs.LG

Shuo Tong , Shangde Gao , Ke Liu , Zihang Huang , Hongxia Xu , Haochao Ying , Jian Wu This is my paper · ORCID
classification cs.CVcs.LG
keywords knowledgedistillationmulti-expertdiseaseimagegradingclassimbalanceuncertainty-awarediabeticretinopathyprostatecancerfeaturealignment
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 proposes UMKD, a multi-expert knowledge-distillation framework that compresses several expert models into one compact student for disease-image grading under class imbalance. Its central claim is that a student trained by UMKD reaches the best published trade-off among distillation baselines in both source-imbalanced and target-imbalanced settings on prostate histology (SICAPv2) and diabetic-retinopathy fundus (APTOS) grading. The key move is to treat teacher uncertainty as a signal for what to transfer: where a teacher's local prediction is ambiguous, the student is pushed to match the teacher's class distribution harder; where the teacher is confident, the student preserves exact logit alignment. On target-imbalanced APTOS the edge is metric-dependent, since a feature-based baseline keeps higher overall accuracy and F1 while UMKD leads in mean accuracy. If correct, this gives a practical recipe for deploying a small model in clinical settings without losing minority-class performance to the bias that imbalance injects into teachers.

What carries the argument

The load-bearing mechanism is the uncertainty-aware decoupled distillation (UDD) loss of Eq. (7): for each teacher $T_t$, scale $w$, and spatial cell $n$, the loss is $L_{\mathrm{UDD}}(w,n) = (2+U_{T_t}) L_{\mathrm{TCKD}} + (1-U_{T_t}) L_{\mathrm{NCKD}}$, where $U_{T_t} = 1 - \max(\sigma(\psi_{T_t}(w,n)))$ is the teacher's local ambiguity measured by how far its softmax output is from a one-hot vector. The intuition the paper relies on is that imbalance makes teachers unreliable precisely where their softmax is flat, so the student should intensify learning on those cells via $L_{\mathrm{TCKD}}$ and relax exact-logit matching via $L_{\mathrm{NCKD}}$. Two feature-space modules carry the rest: shallow feature alignment (SFA) uses multi-scale low-pass filtering to align task-agnostic structural features, and compact feature alignment (CFA) projects penultimate-layer features into a shared spherical space to align task-specific features; both are trained with maximum mean discrepancy plus a reconstruction loss that keeps the frozen experts unchanged. Together they are meant to handle architectural heterogeneity and source/target distribution mismatch while UDD handles imbalance bias.

What would settle it

Create a validation set with a known rare class, record the teacher's $U_{T_t}$ per image region and its per-region misclassification, and check whether high uncertainty concentrates where errors occur; alternatively, run UMKD with random or constant weights in place of $(2+U_{T_t})$ and $(1-U_{T_t})$ and compare accuracy. If uncertainty does not track errors, or the random-weight control matches UMKD, the uncertainty mechanism is not the source of the gains.

Watch

Extended reading notes

Core claim

The paper's central discovery is that class-imbalance bias in multi-expert distillation can be countered by a per-region, uncertainty-aware reweighting of decoupled knowledge. In the output space, UDD partitions each teacher's logit map into multi-scale cells, defines $U_{T_t} = 1 - \max(\sigma(\psi_{T_t}(w,n)))$, and scales the target-class distillation term $L_{\mathrm{TCKD}}$ by $(2+U_{T_t})$ while scaling the non-target term $L_{\mathrm{NCKD}}$ by $(1-U_{T_t})$; ambiguous cells therefore dominate training, while confident cells keep the fine-grained logit alignment that decoupled distillation is designed to protect. In feature space, shallow feature alignment (SFA) suppresses high-frequency details with multi-scale low-pass filtering to align structural, task-agnostic features, and compact feature alignment (CFA) maps penultimate-layer features into a shared spherical space to align task-specific semantic features; both use maximum mean discrepancy plus a reconstruction loss. The paper reports that this combination beats all compared baselines on SICAPv2 and APTOS for the source-imbalanced task, and on APTOS target-imbalanced it reports the best mean accuracy while noting that feature-based RKD still has higher overall accuracy and F1. On SICAPv2 target-imbalanced it reports the best performance on all four metrics.

Load-bearing premise

The method assumes that a teacher's flat softmax output, measured as one minus its top probability, really tracks the bias that class imbalance injects into that teacher's local predictions, so that weighting the distillation loss by this number is what improves the student.

Editorial extensions

If this is right

  • A compact student can replace a larger expert ensemble: across the paper's tables, UMKD's ResNet18 student hits or exceeds the per-expert ResNet50 results on the imbalanced tasks.
  • Uncertainty-aware weighting gives a rule for combining many teachers: instead of averaging all outputs, the student can trust each teacher where it is confident and learn class structure from it where it is uncertain.
  • Source-imbalanced distillation, where the teachers themselves learned from skewed data, is the harder setting and is where the UDD weighting shows its largest margin over SDD on SICAPv2.
  • The framework transfers across heterogeneous architectures without changing the experts, because SFA and CFA align features in normalized common spaces and the reconstruction loss keeps expert weights fixed, a property the paper ties to privacy constraints.
  • Reported gains are metric-dependent in one cell of the tables: on target-imbalanced APTOS, UMKD leads in mean accuracy and MAE, while RKD leads in overall accuracy and F1.

Reading between the lines

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

  • The authors do not test whether $U_{T_t}$ tracks actual teacher errors; a direct diagnostic would be to compare per-patch softmax uncertainty against per-patch misclassification on an artificially imbalanced validation set, and a calibrated confidence measure would be a natural upgrade if the correlation is weak.
  • The same $(2+U)/(1-U)$ reweighting idea transfers to other teacher ensembles, such as federated medical models or temporally averaged teachers, wherever domain shift creates spatially local ambiguity; nothing in the mechanism is specific to disease grading.
  • Because feature-based RKD and UMKD lead on complementary metrics in target-imbalanced APTOS, combining the uncertainty reweighting with relational knowledge distillation is a testable hybrid that might keep both high overall accuracy and high mean accuracy.
  • Before relying on the state-of-the-art claim, a reader should fix one primary metric per task and recompute the comparison tables, since no single method dominates every metric in both datasets.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes UMKD, a multi-expert knowledge distillation framework for imbalanced disease image grading, combining two feature-space alignment modules (SFA in the frequency domain and CFA in a spherical space) with an output-space uncertainty-aware decoupled distillation (UDD) loss. The UDD loss reweights DKD-style TCKD and NCKD terms per spatial cell using a teacher softmax-concentration uncertainty coefficient U_T, and the total loss adds a classification loss with weighting hyperparameters alpha and beta. The method is evaluated on SICAPv2 prostate grading and APTOS fundus grading under two scenarios: source-imbalanced distillation (experts trained on imbalanced data, distillation on balanced data) and target-imbalanced distillation (experts trained on balanced data, distillation on imbalanced data), reporting OA, mAcc, F1, and MAE against several KD baselines plus an ablation study on SICAPv2.

Significance. If the central claims held, the paper would make a useful contribution to an important practical problem: enabling a compact student model to inherit multi-expert knowledge while mitigating class-imbalance bias in medical image grading. The use of public datasets, external baselines, and ablation components is a strength, and the reported SICAPv2 gains are large. However, the current evidence does not fully support the abstract's state-of-the-art claim, and the paper's central mechanism—uncertainty-aware weighting—is not directly validated as the cause of the observed gains. The missing control experiments and the inconsistency in the 'balanced' subset sizes are load-bearing issues for the paper's main narrative.

major comments (4)
  1. [Abstract and Table 2 (APTOS target-imbalanced)] The abstract and Section 3.2 claim 'a new state-of-the-art in both source-imbalanced and target-imbalanced scenarios,' but Table 2 shows that in the target-imbalanced APTOS task RKD achieves higher OA (85.00 vs. 83.91) and higher F1 (84.38 vs. 84.03) than UMKD. The claim therefore holds only for mAcc and MAE, and the 'state-of-the-art' statement is overstated as written. Please either revise the claim to be metric-specific or provide a clearly defined aggregated criterion under which UMKD is best.
  2. [§2, Eq. (7) and §3.3, Table 3 (UDD mechanism)] The core novelty is the UDD weighting in Eq. (7), where U_T = 1 - max(softmax(teacher logits)) is asserted to quantify expert bias from class imbalance and to justify the weights (2+U_T) and (1-U_T). The paper provides no evidence that this softmax-concentration score tracks per-region teacher unreliability or minority-class error; the only support is end-task accuracy. The ablation in Table 3 removes UDD entirely, so it cannot distinguish 'uncertainty-aware weighting helps' from 'any spatially varying reweighting of TCKD/NCKD helps' or 'the logit-map partitioning helps.' To substantiate the mechanistic claim, please add controls that fix the functional form, e.g., constant weights, random per-cell weights, U_T computed from the student instead of the teacher, and, ideally, a direct analysis correlating U_T with teacher misclassification on minority classes.
  3. [§3.1, Implementation details (balanced subsets)] The text says the authors 'generate balanced subsets from original imbalanced datasets' and lists SICAPv2-balanced (2500, 2222, 2500, 948) and APTOS-balanced (600, 370, 300, 193, 295). These class counts are not equal across classes, so the datasets are not class-balanced as the term is normally used. This ambiguity affects the interpretation of both the source-imbalanced and target-imbalanced setups. Please clarify the exact class distribution used in each scenario, or relabel the subsets with their true counts.
  4. [§2, Eq. (2) and surrounding text] The text states that the reconstruction loss L_MSE is used to 'ensure that the expert models remain unchanged due to privacy constraints,' but Eq. (2) defines L_MSE as a loss that measures the change between original and decoded expert features. If the expert models are frozen and unchanged, it is unclear why a reconstruction loss on teacher features is needed or what it constrains. Please clarify whether the experts are trainable in the alignment modules, and how the reconstruction loss is computed without modifying the experts.
minor comments (4)
  1. [§3.2, Results on APTOS Grading] The phrase 'As shown in 7-th row' is unclear; the row numbers are not labeled in Table 2, and the sentence would be clearer with the method name (RKD) and metric values stated directly.
  2. [§3.3, Ablation Study] The ablation study is reported only on SICAPv2, with APTOS results deferred to 'space limitations.' Since APTOS is the dataset with the strongest imbalance and where the target-imbalanced claim is weakest, providing the APTOS ablation table in a supplement would substantially strengthen the paper.
  3. [Tables 1 and 2] None of the tables report standard deviations, confidence intervals, or statistical significance tests. Given that many of the reported gaps between methods are small (e.g., below 1 point), adding variance estimates from multiple runs would materially improve the reliability of the comparison.
  4. [Throughout] The phrase 'sources-imbalanced' is used inconsistently ('sources-imbalanced' versus 'source-imbalanced'); please standardize the terminology. Also, the caption of Fig. 1 describes 'sources-imbalanced' and 'target-imbalanced KD tasks' but the plot itself is not fully explained in the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: UDD weights are computed from teacher logits and evaluated against external baselines.

full rationale

The paper's central derivation is self-contained with respect to the model definition. The uncertainty coefficient U_T = 1 - max(softmax(teacher logits)) is computed directly from the teacher's own output distribution for each spatial cell, not fitted to any target label or test-set statistic. The UDD loss in Eq. (7) is a weighted combination of two DKD terms whose weights depend on this coefficient; the paper does not derive its empirical claim from that formula, and the formula is not an identity with any input quantity. The feature alignment losses (MMD and reconstruction) are standard distribution-matching objectives, and the final classification loss trains the student on the same labels, but this is ordinary supervised distillation rather than a circular prediction. All reported gains are measured against external baselines (FitNet, RKD, KD, DKD, SDD) on public datasets (SICAPv2, APTOS), so the central claim is not a definitional tautology and is falsifiable. The paper does cite several prior works by overlapping authors (e.g., [5], [7], [8]), but those citations are used as background or as sources of standard techniques such as MMD; none is load-bearing for the novelty claim, and the uncertainty mechanism does not rest on a self-citation chain. The ablation in Table 3 removes UDD entirely and therefore does not isolate whether the gain comes from the specific uncertainty weighting or from any dynamic reweighting; that is an experimental-validity limitation, not circularity. Likewise, the statement that UMKD achieves state-of-the-art is metric-dependent in the target-imbalanced APTOS case (RKD has higher OA and F1), but this is an overclaim, not a circular derivation. No equation in the paper reduces to another by construction, and no fitted parameter is renamed as a prediction.

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

The central result rests on empirical benchmarks and on assumptions that frozen expert features can be aligned without distortion, that low-pass filtering separates structure from semantics, and that softmax uncertainty captures imbalance-induced bias. No external calibration is provided for the uncertainty measure, and multiple loss weights and architectural scales are unreported.

free parameters (5)
  • alpha (α) = not reported
    Loss weight for feature-alignment terms in Eq. (8); not disclosed, so the reported results depend on an unreported value chosen by hand or validation.
  • beta (β) = not reported
    Loss weight for UDD terms in Eq. (8); same issue as alpha, and no sensitivity analysis is provided.
  • UDD weighting constants = 2 and 1
    Eq. (7) fixes TCKD weight at 2+U_T and NCKD at 1-U_T; the 2 and 1 are hand-chosen design constants without derivation or ablation.
  • SFA filter scales and strides = not reported
    Multi-scale average-pooling kernel sizes, strides, the downsampling factor s, and the 3x3 convolution in Eq. (5) are not specified.
  • UDD partition scales = not reported
    The set W = {1,2,4,...,w_max} and w_max are unspecified, yet they determine which local regions contribute to distillation.
assumptions (5)
  • domain assumption MMD over projected features is a reliable measure of distribution mismatch between heterogeneous teacher and student encoders.
    Used in Eq. (1) for both SFA and CFA; standard in domain adaptation but not validated for cross-architecture medical features.
  • domain assumption Multi-scale average-pooling low-pass filtering preserves task-agnostic structural information while removing high-frequency details that are irrelevant or harmful.
    Motivates SFA in Section 2 and Eq. (4); no direct evidence that the frequency split separates structure from grading semantics.
  • ad hoc to paper The teacher's softmax uncertainty U_T = 1 - max(softmax(logits)) reflects imbalance-induced error and is a suitable basis for weighting distillation terms.
    Introduced for UDD in Eq. (7); not calibrated against teacher misclassification, so it is an unverified modeling assumption.
  • domain assumption Expert models can remain frozen under privacy constraints while decoded expert features are used for alignment, with the reconstruction loss keeping the projection near-identity.
    Stated in Section 2 around Eq. (2); if the projection materially changes expert features, the privacy-preserving story is unsupported.
  • standard math MMD and softmax are standard mathematical tools used as defined in Eqs. (1) and (7).
    Background formalism, not the source of the empirical claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncertainty-Aware Multi-Expert Knowledge Distillation for Imbalanced Disease Grading." pith.science (2026). https://pith.science/paper/TBB5N3OD

@misc{pith2026250500592,
  author       = {Pith},
  title        = {Pith review of: Uncertainty-Aware Multi-Expert Knowledge Distillation for Imbalanced Disease Grading},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TBB5N3OD}},
  note         = {Machine review of arXiv:2505.00592}
}
read the original abstract

Automatic disease image grading is a significant application of artificial intelligence for healthcare, enabling faster and more accurate patient assessments. However, domain shifts, which are exacerbated by data imbalance, introduce bias into the model, posing deployment difficulties in clinical applications. To address the problem, we propose a novel \textbf{U}ncertainty-aware \textbf{M}ulti-experts \textbf{K}nowledge \textbf{D}istillation (UMKD) framework to transfer knowledge from multiple expert models to a single student model. Specifically, to extract discriminative features, UMKD decouples task-agnostic and task-specific features with shallow and compact feature alignment in the feature space. At the output space, an uncertainty-aware decoupled distillation (UDD) mechanism dynamically adjusts knowledge transfer weights based on expert model uncertainties, ensuring robust and reliable distillation. Additionally, UMKD also tackles the problems of model architecture heterogeneity and distribution discrepancies between source and target domains, which are inadequately tackled by previous KD approaches. Extensive experiments on histology prostate grading (\textit{SICAPv2}) and fundus image grading (\textit{APTOS}) demonstrate that UMKD achieves a new state-of-the-art in both source-imbalanced and target-imbalanced scenarios, offering a robust and practical solution for real-world disease image grading.

Figures

Figures reproduced from arXiv: 2505.00592 by the authors.

Figure 1
Figure 1. Domain shifts between source and target data (left) and the performance [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Model of uncertainty-aware multi-expert knowledge distillation. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 14 canonical work pages

  1. [1]

    Nature medicine 28(1), 154–163 (2022)

    Bulten, W., Kartasalo, K., Chen, P.H.C., Str¨ om, P., Pinckaers, H., Nagpal, K., Cai, Y., Steiner, D.F., Van Boven, H., Vink, R., et al.: Artificial intelligence for diagnosis and gleason grading of prostate cancer: the panda challenge. Nature medicine 28(1), 154–163 (2022)

    work page 2022

  2. [2]

    Robust Image Ordinal Regression with Controllable Image Generation

    Cheng, Y., Ying, H., Hu, R., Wang, J., Zheng, W., Zhang, X., Chen, D., Wu, J.: Robust image ordinal regression with controllable image generation. arXiv preprint arXiv:2305.04213 (2023)

    work page Pith review arXiv 2023

  3. [3]

    Nature Communications 15(1), 4596 (2024)

    Claudio Quiros, A., Coudray, N., Yeaton, A., Yang, X., Liu, B., Le, H., Chiriboga, L., Karimkhan, A., Narula, N., Moore, D.A., et al.: Mapping the landscape of his- tomorphological cancer phenotypes using self-supervised learning on unannotated pathology slides. Nature Communications 15(1), 4596 (2024)

    work page 2024

  4. [4]

    Nature communications 12(1), 3242 (2021)

    Dai, L., Wu, L., Li, H., Cai, C., Wu, Q., Kong, H., Liu, R., Wang, X., Hou, X., Liu, Y., et al.: A deep learning system for detecting diabetic retinopathy across the disease spectrum. Nature communications 12(1), 3242 (2021)

    work page 2021

  5. [5]

    Information Sciences 669, 120564 (2024)

    Gao, S., Fu, Y., Liu, K., Gao, W., Xu, H., Wu, J., Han, Y.: Collaborative knowl- edge amalgamation: Preserving discriminability and transferability in unsupervised learning. Information Sciences 669, 120564 (2024)

    work page 2024

  6. [6]

    In: International Conference on Artificial Neural Networks

    Gao, S., Fu, Y., Liu, K., Han, Y.: Contrastive knowledge amalgamation for un- supervised image classification. In: International Conference on Artificial Neural Networks. pp. 192–204. Springer (2023)

    work page 2023

  7. [7]

    KA$^2$ER: Knowledge Adaptive Amalgamation of ExpeRts for Medical Images Segmentation

    Gao, S., Fu, Y., Liu, K., Xu, H., Wu, J.: Ka 2 er: Knowledge adaptive amalgama- tion of experts for medical images segmentation. arXiv preprint arXiv:2410.21085 (2024)

    work page Pith review arXiv 2024

  8. [8]

    Medical Image Analysis 89, 102889 (2023)

    Gao, S., Zhou, H., Gao, Y., Zhuang, X.: Bayeseg: Bayesian modeling for medical image segmentation with interpretable generalizability. Medical Image Analysis 89, 102889 (2023)

    work page 2023

Show all 23 references
  1. [9]

    Advances in Neural Information Processing Systems 36 (2024)

    Hao, Z., Guo, J., Han, K., Tang, Y., Hu, H., Wang, Y., Xu, C.: One-for-all: Bridge the gap between heterogeneous architectures in knowledge distillation. Advances in Neural Information Processing Systems 36 (2024)

    2024

  2. [10]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    2016

  3. [11]

    arXiv preprint arXiv:1503.02531 (2015)

    Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015)

    2015 arXiv

  4. [12]

    https://kaggle.com/ competitions/aptos2019-blindness-detection (2019), kaggle

    Karthik, Maggie, Dane, S.: Aptos 2019 blindness detection. https://kaggle.com/ competitions/aptos2019-blindness-detection (2019), kaggle

    2019

  5. [13]

    In: Forty- first International Conference on Machine Learning (2024)

    Li, L., Li, X.C., Ye, H.J., Zhan, D.C.: Enhancing class-imbalanced learning with pre-trained guidance through class-conditional knowledge distillation. In: Forty- first International Conference on Machine Learning (2024)

    2024

  6. [14]

    Medical image analysis 42, 60–88 (2017)

    Litjens, G., Kooi, T., Bejnordi, B.E., Setio, A.A.A., Ciompi, F., Ghafoorian, M., Van Der Laak, J.A., Van Ginneken, B., S´ anchez, C.I.: A survey on deep learning in medical image analysis. Medical image analysis 42, 60–88 (2017)

    2017

  7. [15]

    IEEE Transactions on Parallel and Dis- tributed Systems 34(6), 1789–1801 (2023)

    Mohan, N.J., Murugan, R., Goel, T., Roy, P.: Drfl: federated learning in diabetic retinopathy grading using fundus images. IEEE Transactions on Parallel and Dis- tributed Systems 34(6), 1789–1801 (2023)

    2023

  8. [16]

    In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Park, W., Kim, D., Lu, Y., Cho, M.: Relational knowledge distillation. In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 3967–3976 (2019)

    2019

  9. [17]

    Medical image analysis 59, 101561 (2020)

    Porwal, P., Pachade, S., Kokare, M., Deshmukh, G., Son, J., Bae, W., Liu, L., Wang, J., Liu, X., Gao, L., et al.: Idrid: Diabetic retinopathy–segmentation and grading challenge. Medical image analysis 59, 101561 (2020)

    2020

  10. [18]

    arXiv preprint arXiv:1412.6550 (2014)

    Romero, A., Ballas, N., Kahou, S.E., Chassang, A., Gatta, C., Bengio, Y.: Fitnets: Hints for thin deep nets. arXiv preprint arXiv:1412.6550 (2014)

    2014 arXiv

  11. [19]

    Computer methods and programs in biomedicine 195, 105637 (2020)

    Silva-Rodr´ ıguez, J., Colomer, A., Sales, M.A., Molina, R., Naranjo, V.: Going deeper through the gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection. Computer methods and programs in biomedicine 195, 105637 (2020)

    2020

  12. [20]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Wang, J., Cheng, Y., Chen, J., Chen, T., Chen, D., Wu, J.: Ord2seq: Regarding ordinal regression as label sequence prediction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 5865–5875 (2023)

    2023

  13. [21]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wei, S., Luo, C., Luo, Y.: Scaled decoupled distillation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 15975– 15983 (2024)

    2024

  14. [22]

    Medical Image Analysis 69, 101985 (2021)

    Xie, X., Niu, J., Liu, X., Chen, Z., Tang, S., Yu, S.: A survey on incorporating domain knowledge into deep learning for medical image analysis. Medical Image Analysis 69, 101985 (2021)

    2021

  15. [23]

    In: Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition

    Zhao, B., Cui, Q., Song, R., Qiu, Y., Liang, J.: Decoupled knowledge distillation. In: Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition. pp. 11953–11962 (2022)

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.