Pith. sign in

REVIEW 4 major objections 6 minor 4 cited by

On dataset transferability in medical image classification

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a transferability score combining NCA label-prediction confidence with the ratio of conv2 to conv1 gradient magnitudes, and shows it ranks medical image sources better than existing feature-only metrics.

desk verdict A valuable benchmark and a promising NCA-based metric, but the cross-domain superiority claim is an artifact of post hoc sign reversal; the in-domain advantage is not statistically significant. read the letter →

arxiv 2412.20172 v1 pith:G7HL44M4 submitted 2024-12-28 cs.CV

Dovile Juodelyte , Enzo Ferrante , Yucheng Lu , Prabhant Singh , Joaquin Vanschoren , Veronika Cheplygina This is my paper · ORCID
classification cs.CV
keywords transferlearningmedicalimageclassificationtransferabilityestimationdatasetsimilarityMedMNISTneighborhoodcomponentanalysisgradient-basedrankingmodelselection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that choosing a pretrained source model for a medical image classification task should depend not only on how well the source's features already separate the target classes, but also on how easily those features can adapt to the target's subtle, texture-driven patterns. It proposes a transferability score $S = S_{LP} \times S_{FU}$ that multiplies a label-prediction probability computed after a Neighborhood Component Analysis projection of the target features (feature suitability) by the ratio of gradient norms in the second and first convolutional layers from a single triplet-loss backward pass (feature adaptability). Across 11 MedMNIST targets with 14 source datasets, the score ranks sources in closer agreement with true fine-tuned AUC than LogME, SFDA, PARC, NCTI, LEEP, and NLEEP, reaching the best average rank in both the source-dataset and cross-domain model-selection scenarios. The paper also reports that source dataset size alone does not predict transfer performance, that source diversity matters more than similarity, and that medical-specific sources beat ImageNet on 4 of 11 targets. A practitioner who trusts the result can skip exhaustive fine-tuning and select a source with one forward pass and one backward pass.

What carries the argument

The central object is the composite transferability score $S = S_{LP} \times S_{FU}$. $S_{LP}$ uses Neighborhood Component Analysis on the penultimate-layer features of the source model to approximate fine-tuning dynamics without inverting a covariance matrix, producing a label-prediction probability via a 5-NN classifier. $S_{FU}$ is the ratio of the $\ell^2$ norms of gradients with respect to the second and first convolutional layers, computed from a single backward pass of a triplet margin loss; the first layer acts as a normalizer because it is expected to change little, while the second indicates how much the source would adapt. The product is meant to reward sources whose features are both already separable and able to update toward new local texture patterns, and the final ranking is evaluated against true fine-tuned AUC using weighted Kendall's $\tau_w$.

What would settle it

Run the score on a held-out medical target outside MedMNIST and compare the rank correlation of the stated $S_{FU}$ with its sign-flipped version; if the sign-flipped version systematically ranks better, the fixed-direction premise is false.

Watch

Extended reading notes

Core claim

The central claim is that transferability for medical image classification decomposes into two measurable parts: suitability and adaptability. Suitability is captured by $S_{LP}$, the summed probability that a 5-nearest-neighbor classifier assigns the target labels after the source model's penultimate features are projected by Neighborhood Component Analysis, which mimics the class-separating effect of fine-tuning without matrix inversion. Adaptability is captured by $S_{FU}$, the ratio $\|\nabla_{\theta_{conv2}}L\|_2 / \|\nabla_{\theta_{conv1}}L\|_2$ obtained from one backward pass of a triplet margin loss through the source model; the paper argues a larger ratio means the source can more readily learn new local patterns. The product of these normalized terms, $S = S_{LP} \times S_{FU}$, is claimed to rank source datasets better than feature-only transferability metrics, and to avoid their self-source bias whereby a model pretrained on the target itself looks like the best source. The paper further claims the same score, with the gradient term's sign reversed, ranks ImageNet-pretrained architectures for cross-domain transfer to medical targets, and interprets the sign inversion as evidence that cross-domain transfer dynamics differ fundamentally from in-domain transfer.

Load-bearing premise

The ranking depends on the conv2/conv1 gradient ratio having a fixed sign, with larger early-layer updates meaning better transfer, yet the cross-domain experiment requires the opposite sign to achieve positive correlations.

Editorial extensions

If this is right

  • Source model selection for a medical target can be reduced to one forward pass and one backward pass, avoiding grid-searched fine-tuning of every candidate.
  • Public medical datasets, alone or combined as a leave-target-out set, can outperform ImageNet as transfer sources on 4 of 11 targets, so ImageNet should not be the default.
  • Dataset size is not a reliable guide: a small source like Breast can beat a much larger source like OrganS on most targets, so transferable sources must be ranked by content rather than scale.
  • Existing feature-quality transferability metrics suffer from self-source bias and weak or negative correlations on medical targets, while the proposed combined score maintains positive correlation across all in-domain targets.
  • Cross-domain transfer from natural to medical images appears to invert the adaptability signal: better transfer is associated with smaller early-layer updates and larger prediction deviations, pointing to a different mechanism than in-domain transfer.

Reading between the lines

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

  • The paper leaves open whether the sign of the gradient term is a stable property or an artifact of the setting; a natural extension is a domain-aware score that learns the sign of $S_{FU}$ from a small validation set rather than fixing it.
  • The released ground-truth benchmark (15 sources, 9 architectures, 11 targets) could be reused to train a lightweight rank learner that treats $S_{LP}$ and $S_{FU}$ as features, which the paper does not attempt.
  • The NCA approximation of fine-tuning may carry over to other low-data domains where feature dimension exceeds sample size, but the paper only demonstrates it on MedMNIST.
  • The negative correlation between feature suitability and transfer in cross-domain settings suggests suitability metrics may be rewarding shortcut alignment rather than true adaptability; a direct test would compare $S_{LP}$ with shortcut measures on the fine-tuned models.
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 / 6 minor

Summary. The paper proposes a transferability metric S = S_LP x S_FU for ranking pretrained models in medical image classification, where S_LP is an NCA-based label prediction probability and S_FU is the normalized ratio of conv2-to-conv1 gradient norms from a single triplet-loss backward pass. The authors introduce two evaluation scenarios: source-dataset transferability for 11 MedMNIST targets with 14 source datasets, and cross-domain model transferability for 9 ImageNet-pretrained architectures. They report that the proposed method achieves the best average weighted Kendall tau rank in both scenarios, and they release a large ground-truth benchmark from more than 20,000 fine-tuned models.

Significance. If the central claims were supported, the metric would have clear practical value: practitioners could select a source model with one forward and one backward pass instead of exhaustive fine-tuning. The paper also makes a useful empirical contribution by releasing ground-truth transfer performance tables for 15 source datasets, 9 architectures, and 11 medical targets, along with publicly available code. However, the cross-domain claim depends on a post hoc sign reversal of S_FU (Eq. 8 vs. Eq. 5), and the in-domain advantage is not statistically significant according to the paper's own Friedman test. The benchmarking data and the exploratory observation that cross-domain transfer from natural to medical images behaves differently remain valuable, but the stated conclusion of outperforming existing metrics in both settings is not supported as written.

major comments (4)
  1. [Table 3 and Eq. (8)] The cross-domain experiment does not support the claim that the proposed metric outperforms existing methods. The authors first observe predominantly negative weighted Kendall tau for Ours (and for PARC, NCTI, LEEP, and NLEEP), and then apply Eq. (8), which reverses the S_FU normalization relative to Eq. (5). Because Kendall's tau is antisymmetric under score reversal, this is equivalent to selecting, after seeing the data, the orientation that yields positive correlation. With the original sign, Ours would have mostly negative tau values and would rank near the bottom of Table 3. The text acknowledges that the transformation 'contradicts intuition' but does not provide independent evidence for the hypothesized direction of S_FU. The Abstract's claim of outperformance in cross-domain transfer is therefore unsupported; the authors should either validate the sign on held-out targets before computing correlations or present the analysis as exploratory rather than as a metric comparison.
  2. [Table 2, Friedman test] Even in the in-domain source-dataset scenario, the stated outperformance is not statistically supported. The Friedman test rejects the null hypothesis (p = 0.002), but with a critical difference of 2.792 and the best average rank 1.91, all methods with average rank up to 4.70 are within the critical difference. This includes NLEEP (3.45), LEEP (3.91), LogME (4.00), and PARC (4.00). The descriptive statement that Ours outperforms NLEEP on six targets is correct, but the general conclusion that the method 'outperforms existing transferability metrics' goes beyond what the statistical test supports. The sentence that additional datasets would make the difference significant is speculative and should not be used as evidence; either temper the claim or provide additional experiments.
  3. [Section 3.2, Eq. (4), and Table 3] The gradient ratio S_FU is intended to make gradients comparable across source models, but in the cross-domain experiment the models differ not only in pretraining data but also in architecture: ResNet18, DenseNet121, EfficientNetV2-S, MobileNetV3-Small, GoogleNet, MnasNet-1.0, VGG11, ConvNeXt-Tiny, and ShuffleNetV2-0.5x have different first-layer kernel sizes, strides, channel widths, and depths. A raw ratio of conv2/conv1 gradient norms has no common scale across these architectures, and the paper does not justify why this ratio is comparable. This is a load-bearing issue for the cross-domain rankings, because the same metric is applied to models that are not directly comparable through a simple gradient-norm ratio.
  4. [Section 5.2 and Table 3] The 1 - S transformation is applied selectively to PARC, NCTI, LEEP, NLEEP, and Ours, but not to LogME or SFDA. If the transformation is meant to correct for systematically negative correlations, it should be applied to all methods under a prespecified rule, or to none. Applying it to some methods but not others makes the comparison in Table 3 non-principled and can invert the relative ordering of methods. The authors should either justify a single decision rule for when the transformation is applied or compare all methods under the same convention.
minor comments (6)
  1. [Introduction and Related Work] There are several typographical errors, including 'Chaves el al.' (should be 'Chaves et al.') and 'transferability estation' (should be 'transferability estimation').
  2. [Eqs. (5) and (8)] The paper uses two different min-max normalizations for S_FU with opposite ordering. Please define a single normalization convention and state explicitly how the cross-domain normalization in Eq. (8) is computed across models, since it is not the same as Eq. (5).
  3. [Table 2] SFDA has missing tau values on binary classification tasks, and the paper assigns the lowest rank for these missing values. This imputation should be justified, and a sensitivity analysis (e.g., excluding SFDA from the Friedman test) should be reported because it affects the average ranks.
  4. [Figure 3] Please specify which source model and which target subset are shown in the t-SNE projection. The caption currently only says 'binary Pneumonia classification,' so the reader cannot determine how representative the figure is.
  5. [Section 5.2, Table 3] For targets where the proposed method's predictions are negatively correlated, the ground-truth AUC differences between the best and worst source models are very small (e.g., 0.004 for Blood and 0.007 for Path). The paper should discuss whether the correlation values are practically meaningful in such compressed ranges, because small ranking differences may not correspond to clinically relevant performance differences.
  6. [Eq. (6)] The final score is a product of two normalized terms, but the choice of product over sum is not justified. Since S_LP values can be very small, the product may be dominated by one term; a sensitivity analysis using an additive combination or different normalization would be informative.

Circularity Check

1 steps flagged · score 6.0 of 10

Cross-domain claim is generated by post hoc sign reversal (Eq. 8 vs Eq. 5), so 'outperforms in both settings' is partly fitted; the in-domain scenario remains independent.

  1. fitted input called prediction [Section 5.2, 'Model transferability in cross-domain transfer', Eq. 8 (vs. Eq. 5)]
    "In this scenario, none of the evaluated transferability metrics demonstrate a positive rank correlation across all target datasets. In fact, PARC, NCTI, LEEP, NLEEP, and our proposed method predominantly have negative rank correlations. To address this, we transform the predictions of these methods to 1−S(ϕm,T ). For our method specifically, we normalize the feature quality and feature update terms before combining them, as follows:"

    Eq. 8 reverses the min-max normalization of Eq. 5 by using denominator min−max instead of max−min, which is equivalent to ranking by −S_FU. Reversing the predicted ranking flips concordant and discordant pairs, so the sign of the correlation with ground truth is flipped. The paper selects this orientation only after observing that the original metric predominantly correlates negatively with the fine-tuning benchmark in Table 3, i.e., the sign is fitted to the target data. The paper itself states the adjustment 'contradicts intuition' and leaves 'a potential gap in our understanding.' Without Eq. 8, Ours would have predominantly negative τw and would not outperform the baselines in the cross-domain scenario, so the abstract's 'outperforms ...

full rationale

The source-dataset transferability scenario is self-contained: S_LP is computed from an NCA projection of source features and a 5-NN likelihood (Eq. 2), S_FU from a triplet-loss backward pass (Eq. 4), and neither term is fit to the fine-tuning AUC used as ground truth. The in-domain rankings are therefore independent evidence. The cross-domain scenario, however, is not independent: after observing that Ours and several baselines predominantly correlate negatively with ground truth, the paper reverses the feature-update normalization (Eq. 8 vs. Eq. 5) and applies the same 1−S reversal to the baselines. Since reversing a ranking flips the sign of the correlation, this is equivalent to choosing the sign of the metric after seeing the benchmark; the paper concedes it 'contradicts intuition' and calls it 'a potential gap in our understanding.' With the original orientation, Ours would not show the reported positive cross-domain rankings, so the abstract's 'outperforms in both settings' is supported only by the post hoc orientation. The self-citations [6,30] motivating the gradient term are not load-bearing: the ablation and external fine-tuning benchmark carry the argument. Overall, the circularity is partial and confined to the cross-domain claim, while the in-domain result retains independent content.

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

The metric involves hand-chosen design choices: triplet margin alpha, k=5 for the neighbor classifier, the product combination rule, the conv1/conv2 layer pair, and the normalization scheme. The cross-domain experiment introduces a further hand-chosen sign flip. The core domain assumptions are that NCA approximates fine-tuning dynamics and that the gradient ratio indicates adaptability; the latter is contradicted by the cross-domain sign inversion. No invented entities are introduced.

free parameters (5)
  • Triplet margin alpha = not reported
    Eq (3) uses triplet margin loss with an unspecified margin alpha; gradient magnitudes in Eq (4) depend on this hand-chosen value.
  • k in 5-NN label probability = 5
    S_LP in Eq (2) uses a 5-NN classifier; the number of neighbors is chosen without ablation.
  • Combination rule = product
    Final score in Eq (6) multiplies normalized S_LP and S_FU; no comparison with additive or weighted combinations is provided.
  • Gradient layer pair = conv1, conv2
    The adaptability term uses the ratio of conv2 to conv1 gradient norms; the choice of these layers is justified only by a hypothesis about early-layer updates.
  • Sign of feature-update term = +1 in dataset scenario, -1 in cross-domain
    In the dataset-transfer scenario S_FU is used positively (Eq 5); in the cross-domain scenario it is inverted (Eq 8) after observing negative correlations, effectively fitting the sign to the benchmark.
assumptions (4)
  • domain assumption NCA projection on penultimate-layer features approximates the feature dynamics of fine-tuning.
    Section 3.2 and Figure 3 claim NCA matches fine-tuning behavior better than regularized FDA, but no quantitative evidence links NCA projections to actual fine-tuned representations.
  • ad hoc to paper The conv2/conv1 gradient norm ratio from a single triplet-loss backward pass indicates adaptability to local texture patterns in the target task.
    Eq (4) is motivated by a hypothesis about early layers, not derived; the paper's own cross-domain experiment later inverts this term.
  • domain assumption Ground-truth transfer performance measured by full fine-tuning AUC with grid search is a stable proxy for transferability.
    Section 4.2 benchmarks with one protocol; no repeated runs or seeds are reported, so ranking stability is unknown.
  • standard math Standard mathematical tools: NCA optimization, triplet loss gradients, k-NN classification, and Kendall's tau are correctly applied.
    Used in Eqs (2)-(6); no formal verification is provided but these are standard methods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On dataset transferability in medical image classification." pith.science (2026). https://pith.science/paper/G7HL44M4

@misc{pith2026241220172,
  author       = {Pith},
  title        = {Pith review of: On dataset transferability in medical image classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G7HL44M4}},
  note         = {Machine review of arXiv:2412.20172}
}
read the original abstract

Current transferability estimation methods designed for natural image datasets are often suboptimal in medical image classification. These methods primarily focus on estimating the suitability of pre-trained source model features for a target dataset, which can lead to unrealistic predictions, such as suggesting that the target dataset is the best source for itself. To address this, we propose a novel transferability metric that combines feature quality with gradients to evaluate both the suitability and adaptability of source model features for target tasks. We evaluate our approach in two new scenarios: source dataset transferability for medical image classification and cross-domain transferability. Our results show that our method outperforms existing transferability metrics in both settings. We also provide insight into the factors influencing transfer performance in medical image classification, as well as the dynamics of cross-domain transfer from natural to medical images. Additionally, we provide ground-truth transfer performance benchmarking results to encourage further research into transferability estimation for medical image classification. Our code and experiments are available at https://github.com/DovileDo/transferability-in-medical-imaging.

Figures

Figures reproduced from arXiv: 2412.20172 by the authors.

Figure 1
Figure 1. Illustration of the transferability estimation problem: Given a model [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our method. We use Neighborhood Component Analysis (NCA) on feature representations obtained from a forward pass of the target dataset [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. t-SNE projections of feature representations [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Transfer performance (AUC) of source datasets (y-axis) evaluated on target test sets. Source datasets are sorted by size, from smallest to largest. The grey [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Ground-truth transfer performance P(ϕm,T) (test AUC) on the x-axis versus transferability score S (ϕm,T) on the y-axis. The predicted transferability scores are shown for LogME, LEEP, SFDA, PARC, NCTI, NLEEP, and our method (columns) across 11 medical target datasets (…
Figure 6
Figure 6. Figure 6: Contribution of the feature quality S LP(ϕm,T) and feature update S FU (ϕm,T) terms to the overall transferability score. LEEP and NLEEP–among the earliest proposed transferabil￾ity metrics–outperform more recent, sophisticated methods like NCTI and SFDA that explicitl…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Topology-Driven Transferability Estimation for 3D Medical Vision Foundation Models

    cs.CV 2026-07 conditional novelty 6.5 of 10

    MST-based local boundary leakage and global topology divergence, fused by task complexity, rank SSL 3D medical encoders for segmentation without fine-tuning, beating prior TE metrics by 0.36 weighted Kendall τ at 56× speed.

  2. Intuitions of Machine Learning Researchers about Transfer Learning for Medical Image Classification

    cs.CV 2025-10 conditional novelty 6.0 of 10

    Source-dataset selection for medical transfer learning is driven by community practice and perceived similarity, and 'more similar is better' does not consistently hold.

  3. Robustness of transferability estimation metrics for medical imaging

    eess.IV 2026-08 conditional novelty 5.0 of 10

    Transferability estimation metric rankings in medical imaging are unstable to target resampling and to the evaluation metric used for the reference ranking.

  4. Analysis of Transferability Estimation Metrics for Surgical Phase Recognition

    eess.IV 2025-08 conditional novelty 5.0 of 10

    LogME, aggregated by its minimum per-subset score, best matches fine-tuning accuracy for surgical phase recognition across two datasets, while TransRate reverses true model rankings.

Reference graph

Works this paper leans on

59 extracted references · 55 canonical work pages · cited by 4 Pith papers

  1. [1]

    Cheplygina, Cats or cat scans: Transfer learning from natural or med- ical image source data sets?, Current Opinion in Biomedical Engineering 9 (2019) 21–27

    V . Cheplygina, Cats or cat scans: Transfer learning from natural or med- ical image source data sets?, Current Opinion in Biomedical Engineering 9 (2019) 21–27

    work page 2019

  2. [2]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2009, pp. 248–255

    work page 2009

  3. [3]

    Raghu, C

    M. Raghu, C. Zhang, J. Kleinberg, S. Bengio, Transfusion: Understand- ing transfer learning for medical imaging, Advances in neural information processing systems 32 (2019)

    work page 2019

  4. [4]

    Mensink, J

    T. Mensink, J. Uijlings, A. Kuznetsova, M. Gygli, V . Ferrari, Factors of influence for transfer learning across diverse appearance domains and task types, IEEE Transactions on Pattern Analysis and Machine Intelligence 44 (12) (2021) 9298–9314

    work page 2021

  5. [5]

    Y . Cui, Y . Song, C. Sun, A. Howard, S. Belongie, Large scale fine-grained categorization and domain-specific transfer learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 4109–4118

    work page 2018

  6. [6]

    Juodelyte, Y

    D. Juodelyte, Y . Lu, A. Jim ´enez-S´anchez, S. Bottazzi, E. Ferrante, V . Cheplygina, Source matters: Source dataset impact on model robust- ness in medical imaging, in: MICCAI Workshop on Applications of Med- ical AI (AMAI), Springer, 2024 (In press)

    work page 2024

  7. [7]

    Y . Lu, D. Juodelyte, J. D. Victor, V . Cheplygina, Exploring connections of spectral analysis and transfer learning in medical imaging, in: Medical Imaging 2025: Image Processing, SPIE, 2025 (In press)

    work page 2025

  8. [8]

    J. W. Gichoya, I. Banerjee, A. R. Bhimireddy, J. L. Burns, L. A. Celi, L.- C. Chen, R. Correa, N. Dullerud, M. Ghassemi, S.-C. Huang, P.-C. Kuo, M. P. Lungren, L. J. Palmer, B. J. Price, S. Purkayastha, A. T. Pyrros, L. Oakden-Rayner, C. Okechukwu, L. Seyyed-Kalantari, H. Trivedi, R. Wang, Z. Zaiman, H. Zhang, AI recognition of patient race in med- ical i...

    work page 2022

Show all 59 references
  1. [9]

    Bassignana, M

    E. Bassignana, M. M ¨uller-Eberstein, M. Zhang, B. Plank, Evidence> in- tuition: Transferability estimation for encoder selection, in: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Pro- cessing, 2022, pp. 4218–4227

    2022

  2. [10]

    Chaves, A

    L. Chaves, A. Bissoto, E. Valle, S. Avila, The performance of transfer- ability metrics does not translate to medical tasks, in: MICCAI Workshop on Domain Adaptation and Representation Transfer, Springer, 2023, pp. 105–114

    2023

  3. [11]

    J. Yang, R. Shi, B. Ni, Medmnist classification decathlon: A lightweight automl benchmark for medical image analysis, in: IEEE 18th Interna- tional Symposium on Biomedical Imaging (ISBI), 2021, pp. 191–195

    2021

  4. [12]

    J. Yang, R. Shi, D. Wei, Z. Liu, L. Zhao, B. Ke, H. Pfister, B. Ni, Medm- nist v2-a large-scale lightweight benchmark for 2d and 3d biomedical im- age classification, Scientific Data 10 (1) (2023) 41

    2023

  5. [13]

    A. R. Zamir, A. Sax, W. Shen, L. J. Guibas, J. Malik, S. Savarese, Taskonomy: Disentangling task transfer learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 3712–3722

    2018

  6. [14]

    Achille, M

    A. Achille, M. Lam, R. Tewari, A. Ravichandran, S. Maji, C. C. Fowlkes, S. Soatto, P. Perona, Task2vec: Task embedding for meta-learning, in: Proceedings of the IEEE/CVF International Conference on Computer Vi- sion (ICCV), 2019, pp. 6430–6439

    2019

  7. [15]

    X. Peng, Y . Li, K. Saenko, Domain2vec: Domain embedding for unsu- pervised domain adaptation, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Proceedings of the European conference on computer vi- sion (ECCV), V ol. 12351, Springer International Publishing, Cham, 202...

    2020

  8. [16]

    Alvarez-Melis, N

    D. Alvarez-Melis, N. Fusi, Geometric dataset distances via optimal transport, Advances in neural information processing systems 33 (2020) 21428–21439

    2020

  9. [17]

    Nguyen, T

    C. Nguyen, T. Hassner, M. Seeger, C. Archambeau, Leep: A new mea- sure to evaluate transferability of learned representations, in: Proceedings of the International Conference on Machine Learning, PMLR, 2020, pp. 7294–7305

    2020

  10. [18]

    Y . Li, X. Jia, R. Sang, Y . Zhu, B. Green, L. Wang, B. Gong, Ranking neu- ral checkpoints, in: Proceedings of the IEEE /CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2021, pp. 2663–2673

    2021

  11. [19]

    P ´andy, A

    M. P ´andy, A. Agostinelli, J. Uijlings, V . Ferrari, T. Mensink, Transferabil- ity estimation using bhattacharyya class separability, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 9172–9182

    2022

  12. [20]

    Bolya, R

    D. Bolya, R. Mittapalli, J. Ho ffman, Scalable diverse model selection for accessible transfer learning, Advances in neural information processing systems 34 (2021) 19301–19312

    2021

  13. [21]

    K. You, Y . Liu, J. Wang, M. Long, Logme: Practical assessment of pre- trained models for transfer learning, in: Proceedings of the International Conference on Machine Learning, PMLR, 2021, pp. 12133–12143

    2021

  14. [22]

    W. Shao, X. Zhao, Y . Ge, Z. Zhang, L. Yang, X. Wang, Y . Shan, P. Luo, Not all models are equal: Predicting model transferability in a self- challenging fisher space, in: Proceedings of the European conference on computer vision (ECCV), Springer, 2022, pp. 286–302

    2022

  15. [23]

    Z. Wang, Y . Luo, L. Zheng, Z. Huang, M. Baktashmotlagh, How far pre- trained models are from neural collapse on the target dataset informs their transferability, in: Proceedings of the IEEE /CVF International Confer- ence on Computer Vision (ICCV), 2023, pp. 5549–5558

    2023

  16. [24]

    X. Li, Z. Hu, Y . Ge, Y . Shan, L.-Y . Duan, Exploring model transferability through the lens of potential energy, in: Proceedings of the IEEE /CVF International Conference on Computer Vision (ICCV), 2023, pp. 5429– 5438

    2023

  17. [25]

    Papyan, X

    V . Papyan, X. Han, D. L. Donoho, Prevalence of neural collapse during the terminal phase of deep learning training, Proceedings of the National Academy of Sciences 117 (40) (2020) 24652–24663

    2020

  18. [26]

    Y . Yang, M. Wei, J. He, J. Yang, J. Ye, Y . Gu, Pick the best pre-trained model: Towards transferability estimation for medical image segmen- tation, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2023, pp. 674–683

    2023

  19. [27]

    Molina-Moreno, M

    M. Molina-Moreno, M. P. Schilling, M. Reischl, R. Mikut, Automated style-aware selection of annotated pre-training databases in biomedical imaging, in: 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), IEEE, 2023, pp. 1–5

    2023

  20. [28]

    X. Du, Z. Liu, Z. Feng, H. Deng, Datamap: Dataset transferability map for medical image classification, Pattern Recognition 146 (2024) 110044

    2024

  21. [29]

    Vigna, A weighted correlation index for rankings with ties, in: Pro- ceedings of the 24th international conference on World Wide Web, 2015, pp

    S. Vigna, A weighted correlation index for rankings with ties, in: Pro- ceedings of the 24th international conference on World Wide Web, 2015, pp. 1166–1176

    2015

  22. [30]

    Juodelyte, A

    D. Juodelyte, A. Jim ´enez-S´anchez, V . Cheplygina, Revisiting hidden rep- resentations in transfer learning for medical imaging, Transactions on Machine Learning Research (2023)

    2023

  23. [31]

    Goldberger, G

    J. Goldberger, G. E. Hinton, S. Roweis, R. R. Salakhutdinov, Neighbour- hood components analysis, Advances in neural information processing systems 17 (2004)

    2004

  24. [32]

    H. Park, S. Baek, J. Park, High-dimensional linear discriminant analy- sis using nonparametric methods, Journal of Multivariate Analysis 188 (2022) 104836

    2022

  25. [33]

    J. Wang, Y . Song, T. Leung, C. Rosenberg, J. Wang, J. Philbin, B. Chen, Y . Wu, Learning fine-grained image similarity with deep ranking, in: Pro- ceedings of the IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2014, pp. 1386–1393

    2014

  26. [34]

    Musgrave, S

    K. Musgrave, S. J. Belongie, S.-N. Lim, Pytorch metric learning, ArXiv abs/2008.09164 (2020)

    2020 arXiv

  27. [35]

    J. N. Kather, J. Krisam, P. Charoentong, T. Luedde, E. Herpel, C.-A. Weis, T. Gaiser, A. Marx, N. A. Valous, D. Ferber, et al., Predicting survival from colorectal cancer histology slides using deep learning: A retrospec- tive multicenter study, PLoS medicine 16 (1) (2019) e1002730

    2019

  28. [36]

    Tschandl, C

    P. Tschandl, C. Rosendahl, H. Kittler, The ham10000 dataset, a large col- lection of multi-source dermatoscopic images of common pigmented skin lesions, Scientific data 5 (1) (2018) 1–9

    2018

  29. [37]

    Codella, V

    N. Codella, V . Rotemberg, P. Tschandl, M. Emre Celebi, S. Dusza, D. Gutman, B. Helba, A. Kalloo, K. Liopyris, M. Marchetti, et al., Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic), arXiv e-prints (...

    2019

  30. [38]

    D. S. Kermany, M. Goldbaum, W. Cai, C. C. Valentim, H. Liang, S. L. Baxter, A. McKeown, G. Yang, X. Wu, F. Yan, et al., Identifying med- ical diagnoses and treatable diseases by image-based deep learning, cell 172 (5) (2018) 1122–1131

    2018

  31. [39]

    11 URL https://isbi.deepdr.org/data.html

    The 2nd diabetic retinopathy grading and image quality estimation chal- lenge, DeepDR Diabetic Retinopathy Image Dataset (DeepDRiD) (2020). 11 URL https://isbi.deepdr.org/data.html

    2020

  32. [40]

    Al-Dhabyani, M

    W. Al-Dhabyani, M. Gomaa, H. Khaled, A. Fahmy, Dataset of breast ultrasound images, Data in brief 28 (2020) 104863

    2020

  33. [41]

    Acevedo, A

    A. Acevedo, A. Merino, S. Alf ´erez, ´A. Molina, L. Bold ´u, J. Rodellar, A dataset of microscopic peripheral blood cell images for development of automatic recognition systems, Data in brief 30 (2020) 105474

    2020

  34. [42]

    Ljosa, K

    V . Ljosa, K. L. Sokolnicki, A. E. Carpenter, Annotated high-throughput microscopy image sets for validation., Nature methods 9 (7) (2012) 637– 637

    2012

  35. [43]

    X. Xu, F. Zhou, B. Liu, D. Fu, X. Bai, E fficient multiple organ localiza- tion in ct image using 3d region proposal network, IEEE transactions on medical imaging 38 (8) (2019) 1885–1898

    2019

  36. [44]

    Bilic, P

    P. Bilic, P. Christ, H. B. Li, E. V orontsov, A. Ben-Cohen, G. Kaissis, A. Szeskin, C. Jacobs, G. E. H. Mamani, G. Chartrand, et al., The liver tumor segmentation benchmark (lits), Medical Image Analysis 84 (2023) 102680

    2023

  37. [45]

    X. Wang, Y . Peng, L. Lu, Z. Lu, M. Bagheri, R. M. Summers, Chestx- ray8: Hospital-scale chest x-ray database and benchmarks on weakly- supervised classification and localization of common thorax diseases, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- ...

    2017

  38. [46]

    K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recog- nition, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  39. [47]

    X. Mei, Z. Liu, P. M. Robson, B. Marinelli, M. Huang, A. Doshi, A. Ja- cobi, C. Cao, K. E. Link, T. Yang, et al., Radimagenet: an open radiologic deep learning research dataset for e ffective transfer learning, Radiology: Artificial Intelligence 4 (5) (2022) e210315

    2022

  40. [48]

    H. Li, P. Chaudhari, H. Yang, M. Lam, A. Ravichandran, R. Bhotika, S. Soatto, Rethinking the hyperparameters for fine-tuning, arXiv preprint arXiv:2002.11770 (2020)

    2020 arXiv

  41. [49]

    Robroek, A

    T. Robroek, A. Duane, E. Yousefzadeh-Asl-Miandoab, P. Tozun, Data management and visualization for benchmarking deep learning training systems, in: Proceedings of the Seventh Workshop on Data Management for End-to-End Machine Learning, 2023, pp. 1–5

    2023

  42. [50]

    Dem ˇsar, Statistical comparisons of classifiers over multiple data sets, The Journal of Machine learning research 7 (2006) 1–30

    J. Dem ˇsar, Statistical comparisons of classifiers over multiple data sets, The Journal of Machine learning research 7 (2006) 1–30

    2006

  43. [51]

    Huang, Z

    G. Huang, Z. Liu, L. Van Der Maaten, K. Q. Weinberger, Densely con- nected convolutional networks, in: Proceedings of the IEEE /CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 4700–4708

    2017

  44. [52]

    M. Tan, Q. Le, Efficientnetv2: Smaller models and faster training, in: Pro- ceedings of the International Conference on Machine Learning, PMLR, 2021, pp. 10096–10106

    2021

  45. [53]

    Howard, M

    A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y . Zhu, R. Pang, V . Vasudevan, et al., Searching for mobilenetv3, in: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 1314–1324

    2019

  46. [54]

    Szegedy, W

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, A. Rabinovich, Going deeper with convolutions, in: Pro- ceedings of the IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1–9

    2015

  47. [55]

    M. Tan, B. Chen, R. Pang, V . Vasudevan, M. Sandler, A. Howard, Q. V . Le, Mnasnet: Platform-aware neural architecture search for mobile, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), 2019, pp. 2820–2828

    2019

  48. [56]

    Simonyan, A

    K. Simonyan, A. Zisserman, Very deep convolutional networks for large- scale image recognition, arXiv preprint arXiv:1409.1556 (2014)

    2014 arXiv

  49. [57]

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, S. Xie, A convnet for the 2020s, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 11976–11986

    2022

  50. [58]

    N. Ma, X. Zhang, H.-T. Zheng, J. Sun, Shu fflenet v2: Practical guidelines for efficient cnn architecture design, in: Proceedings of the European con- ference on computer vision (ECCV), 2018, pp. 116–131

    2018

  51. [59]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high- perfor...

    2019

Pith tools

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