Pith. sign in

REVIEW 3 major objections 5 minor 52 references

This paper claims that MedXplore, a frequency-domain attention module coupled with confidence-scaled angular margins, makes generalized category discovery in medical images more accurate and less biased toward known classes.

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 →

T0 review · deepseek-v4-flash

2026-08-01 04:28 UTC pith:PNW4NX4O

load-bearing objection MedXplore is a thoughtful method with believable internal ablations, but the headline SOTA claim is built on an uncontrolled baseline table and no code; worth refereeing, with revisions. the 3 major comments →

arxiv 2607.27620 v1 pith:PNW4NX4O submitted 2026-07-30 cs.CV

MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging

classification cs.CV
keywords generalized category discoverymedical imagingfrequency-domain filteringadaptive margin lossold-class biasopen-world recognitionlesion-sensitive representationclustering accuracy
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.

The paper argues that medical images, unlike natural images, contain subtle lesion cues and weak semantic separation, so standard generalized category discovery methods drift attention and over-assign ambiguous novel samples to old, known classes. To fix this, MedXplore couples a perception module (FAAC) that learns to filter the full frequency spectrum and activate local energy contrasts—highlighting the small abnormal regions that distinguish disease subtypes—with a decision module (ACAM) that adaptively sharpens class boundaries based on sample confidence. On four medical benchmarks, the combination raises average All-class clustering accuracy by 8.5 percentage points over the strongest competitor, and on the Kvasir endoscopy dataset it cuts false-old errors (novel samples mislabeled as known) from 14.5% to 0.8%. If correct, this means open-world medical image recognition can discover new disease categories reliably without retraining or manual annotation of those categories.

Core claim

MedXplore's central claim is that old-class bias in medical generalized category discovery is driven by two fixable bottlenecks: attention that overfits to familiar anatomical textures, and decision boundaries that ignore sample reliability. FAAC addresses the first by transforming features into the frequency domain, applying learnable per-channel filtering to the real and imaginary parts of the FFT, then normalizing local variance against global statistics so that only genuinely anomalous regions are amplified; the resulting top patches become semantic anchors for cross-view consistency. ACAM addresses the second by computing a per-sample confidence from attention strength, cross-view patch

What carries the argument

The key machinery is a two-part training-time addition: the FAAC perception branch, which performs learnable full-spectrum filtering via depth-wise convolutions on FFT components, followed by global-local energy contrast activation to select reliable semantic anchors for patch consistency; and the ACAM decision module, which maps a three-way confidence score into adaptive angular and cosine margins for the classification loss. The mechanism works by making the backbone learn texture-agnostic, lesion-sensitive features while constraining ambiguous samples from being collapsed into old classes.

Load-bearing premise

The claim of consistent state-of-the-art performance rests on the assumption that the baseline numbers being compared against were measured under the same backbone, training epochs, and hyperparameter settings as MedXplore; if those baselines were tuned differently or run with different architectures, the reported gains may not reflect the method itself.

What would settle it

Run every baseline method and MedXplore under identical conditions—same backbone, same 200-epoch schedule, same data splits, and multiple random seeds—then compute the mean and variance of All accuracy. If the average gain over the strongest baseline falls below about two points or overlaps within one standard deviation, the claim of consistent superiority is falsified. Separately, recompute the Kvasir false-old error across at least five seeds; if it does not reproduce near 0.80%, the bias-reduction claim is not robust.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Existing medical imaging pipelines could discover novel disease categories from unlabeled data without manual annotation, which matters for rare or emerging conditions.
  • The sharp reduction in false-old errors on endoscopy images implies that CAD systems could flag genuinely new mucosal patterns instead of forcing them into familiar classes.
  • Because FAAC and ACAM are discarded at inference, the approach adds no computational cost at deployment, making it feasible for clinical settings with limited compute.
  • The frequency-domain treatment suggests that subtle, localized lesion cues, not just global appearance, carry the discriminative signal for disease recognition, which could inform other medical representation learning tasks.
  • The combination of perceptual and decision-level corrections may be directly portable to other open-set recognition problems where old-class bias dominates.

Where Pith is reading between the lines

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

  • A critical test for this framework is whether the gains persist under distribution shift—e.g., CT scans from different scanners or histology from different stains—since the paper tunes hyperparameters on one dataset and reuses them elsewhere without cross-domain evaluation.
  • The adaptive margins depend on the confidence signals FAAC supplies; if the perception branch is noisy on diffuse or low-frequency pathologies (a limitation the paper acknowledges), the benefit of ACAM may shrink, suggesting a ceiling for this design on such cases.
  • One could directly test the FAAC hypothesis by measuring whether the learned frequency filters concentrate energy on radiologically-defined lesion regions and whether those regions align with clinician annotations.
  • The 0.8% false-old rate on Kvasir, if replicated across seeds, would make MedXplore a candidate for semi-automated case review: flagging samples that fall near decision boundaries for human inspection.

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

3 major / 5 minor

Summary. The paper proposes MedXplore, a framework for Generalized Category Discovery (GCD) in medical imaging. It consists of two modules: FAAC, which operates in the frequency domain (learnable full-spectrum filtering, global-local energy contrast activation, and patch consistency), and ACAM, which adaptively applies angular and cosine margins to classification logits based on a fused confidence score. The method is trained on top of a SimGCD/NGUF-style parametric GCD baseline with a ResNet-18 backbone and is evaluated on four medical datasets: Kvasir, OrganAMNIST, OrganCMNIST, and NCT-CRC-HE-100K. The authors report an average 8.5% improvement in All accuracy over the strongest competing methods and a reduction of false-old errors on Kvasir from 14.50% to 0.80%. They also provide ablations showing that both FAAC and ACAM contribute to the reported gains.

Significance. If the reported results hold under a controlled comparison, the paper makes a useful contribution to medical GCD. The perceptual/decision decomposition is sensible, and the paper gives credit for several good practices: FAAC and ACAM are training-time only, so inference matches the baseline; the ablation study isolates the contributions of each module; the Kvasir error decomposition explicitly targets old-class bias; and the hyperparameter sensitivity analysis is reported. The methodological novelty is moderate but, together with the strong empirical gains, would be of interest to the medical image analysis and GCD communities. However, the empirical claims currently rest on an uncontrolled baseline table and on point estimates without variance, so the significance is conditional on the authors being able to substantiate the comparison.

major comments (3)
  1. [§5.1, Table 1] The headline claim of an 8.5% average gain and 'state of the art in medical GCD' rests on an uncontrolled baseline table. Section 5.1 states only 'Following [11], we employ a ResNet-18 backbone ... fine-tune only its last stage,' and explicitly says the ViT-B/16 description in Sec. 3.2 is not the experimental backbone. However, SimGCD [43] and LegoGCD [2] appear as unstarred rows without any statement that they were re-run with ResNet-18, last-stage fine-tuning, 200 epochs, batch size 128, lr 0.1, and lambda=0.35. Both original methods use ViT-B/16 with DINO pretraining. Several other rows (RS+, UNO+, GCD, DCCL, CMS, NGUF) are imported from NGUF [11] without evidence that NGUF used the same backbone and training protocol. Without a controlled re-run or a clear demonstration of comparability, the average gain is not a fair comparison.
  2. [§5.1, Table 1] The 'all previous methods' claim is contradicted by the paper's own related-work discussion. MedGCD [5] and ClearGCD [28] are described as recent GCD methods but are absent from Table 1. If their numbers are not available, the claim should be narrowed or the methods should be included. Additionally, no results are reported with multiple seeds or error bars; the Kvasir false-old reduction (14.50% to 0.80%) and all ACC values are point estimates. Clustering accuracy is sensitive to initialization and data split, so the reported robustness and 'consistently outperforms' claims are not yet supported.
  3. [§5.1, 'Implementation details'; §5.4] Hyperparameters are set on OrganAMNIST and then reused on the other datasets, but OrganAMNIST is itself one of the four evaluation datasets. Thus the OrganAMNIST row in Table 1 and the hyperparameter studies in Fig. 6 are not independent of the hyperparameter selection. The authors should either select hyperparameters on a held-out validation split or explicitly acknowledge and justify this optimistic-bias risk. This does not invalidate the method, but it weakens the claim that the reported numbers transfer without dataset-specific tuning.
minor comments (5)
  1. [§5.4, Table 3] The caption for 'Table 3: Incremental internal ablation of FAAC' appears in the manuscript but the table body is missing from the provided text. Please include the actual table or provide a precise reference to the supplementary material with the numbers.
  2. [Fig. 6] The figure showing alpha/beta sensitivity and K/delta sensitivity is not legible in the current version; the labels appear garbled. Please replace it with a clean figure and ensure the axis labels and curves are readable.
  3. [§3.2] The disclaimer that the ViT-B/16 encoder is only a summary of SimGCD and not the experimental backbone is unusual and could be confusing. Consider moving this clarification to Sec. 5.1 or to a footnote.
  4. [Eq. (24)] In the ACAM loss, the symbol C (total number of classes) and the notation B_l are used without a definition in Sec. 4.2. Clarify whether B_l is the same as B_ell introduced in Sec. 3.2 and define C explicitly.
  5. [§5.1, Datasets] Please specify the old/new class split sizes and the labeled/unlabeled sample counts for each dataset in the main text, not only in the appendix. This would help readers assess the difficulty of each benchmark.

Circularity Check

0 steps flagged

No circular derivation; the central SOTA claim is empirical. Minor self-citations are not load-bearing and the flagged benchmark concerns are rigor issues, not circularity.

full rationale

MedXplore's derivation is self-contained as a method paper: FAAC (Eqs. 10-18) and ACAM (Eqs. 19-24) introduce new auxiliary losses, combined in Eq. 25, and their contributions are tested empirically by ablations and public benchmarks. No predicted quantity is defined as the output of a fitted parameter; the adaptive quantities (frequency-filter weights, gamma, and the margins m_a,b/m_c,b) are training-time parameters, not claimed predictions. ACAM's confidence u_b includes the model's own logit gap, but the margin is applied only to labeled samples with ground-truth labels and acts as a regularizer, so no result reduces to its input by construction. The self-citations [28, 29, 44] appear only in related-work/motivation passages and are not load-bearing. The manuscript itself flags its own experimental transparency: Sec. 3.2 explicitly isolates the ViT description as background, and Sec. 5.1 declares hyperparameters selected on OrganAMNIST. These, together with the absence of error bars and the imported NGUF baseline numbers, are empirical-rigor concerns rather than circularity.

Axiom & Free-Parameter Ledger

8 free parameters · 5 axioms · 0 invented entities

The central contribution is an empirical training-time method, so the ledger records hand-set loss weights and thresholds plus assumptions inherited from the GCD setup or introduced by the FAAC design. No new physical or ontological entities are introduced; 'semantic anchors' are a methodological label, not an entity.

free parameters (8)
  • Mmax_a (max angular margin) = not specified in main text
    ACAM Eq. 22 scales angular margin by Mmax_a * u_b; the maximum margin strength must be chosen by hand or tuned, and its value is not reported.
  • Mmax_c (max cosine suppression margin) = not specified in main text
    Eq. 22 scales cosine suppression on hard negatives; the value is required to instantiate ACAM and is not reported in the main text.
  • delta' (negative gating threshold) = not specified in main text
    Eq. 23 uses delta' to decide which negatives receive suppression; the threshold is a hand-set hyperparameter not reported in the main text.
  • s (logit scaling factor) = not specified in main text
    Eq. 23 uses a unified scaling factor s, standard in margin losses; its value is not specified.
  • alpha (patch-consistency loss weight) = 1.0
    Eq. 25; selected by a hyperparameter sweep on OrganAMNIST (Fig. 6a).
  • beta (ACAM loss weight) = 0.5
    Eq. 25; selected by a hyperparameter sweep on OrganAMNIST (Fig. 6b).
  • K (Top-K patch count) = 8
    Selected by sweep on OrganAMNIST (Fig. 6c); larger K degrades performance.
  • delta (patch match similarity threshold) = 0.65
    Selected by sweep on OrganAMNIST (Fig. 6d); controls the strictness of mutual-nearest-neighbor patch matches.
axioms (5)
  • domain assumption Total number of categories K is known a priori
    Sec. 3.1: 'Following prior work, we assume that this total number of categories is known a priori.' This is inherited from the GCD setup and avoids the harder unknown-K setting.
  • ad hoc to paper Local energy variance after learnable frequency filtering indicates lesion saliency
    FAAC/GLEC (Eqs. 13-14) assumes that regions whose local energy deviates strongly from the global filtered context are abnormal lesion regions; no independent evidence is given beyond qualitative visualizations.
  • domain assumption Depthwise convolution on real and imaginary FFT components followed by inverse FFT gives useful frequency filtering
    Eqs. 10-11; the paper does not provide a theoretical justification for this specific frequency-domain operation, only qualitative comparisons in Fig. 3.
  • standard math SimGCD's parametric classification formulation is a valid base objective in medical GCD
    Sec. 3.2 and Eq. 9; MedXplore is built directly on SimGCD's representation and classification losses without re-deriving their validity.
  • domain assumption The 50%/50% category and sample split is representative of clinical GCD conditions
    Sec. 5.1 follows the protocol of [38,43]; real clinical label scarcity may be more extreme or imbalanced.

pith-pipeline@v1.3.0-daily-deepseek · 15928 in / 11684 out tokens · 119043 ms · 2026-08-01T04:28:31.760580+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging." pith.science (2026). https://pith.science/paper/PNW4NX4O

@misc{pith2026260727620,
  author       = {Pith},
  title        = {Pith review of: MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNW4NX4O}},
  note         = {Machine review of arXiv:2607.27620}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Deep learning has shown strong potential in medical image analysis, but most existing methods rely on large-scale annotations and a closed-world assumption that rarely holds in clinical practice. Although Generalized Category Discovery (GCD) has advanced rapidly on natural images, it remains underexplored in medical imaging. To address this issue, we propose MedXplore, a unified framework for reliable and unbiased medical GCD, optimizing from both perceptual and decision levels. Specifically, at the perceptual level, taking a frequency domain perspective, Frequency-SNR Adaptive Attention and Consistency (FAAC) performs learnable full-spectrum filtering and global-local energy contrast activation to not only highlight local abnormal signals relative to the global context, but also provide reliable semantic anchors for patch consistency learning. At the decision level, Adaptive Cosine-Angular Margin (ACAM) adjusts angular margins using semantic difficulty and feature confidence to balance intra-class compactness and inter-class separability. Together, the two modules improve lesion-sensitive representation learning and mitigate old-class bias. Experiments on multiple benchmarks show an average \textbf{8.5\%} gain in \textit{All} accuracy over the strongest competing methods. On Kvasir, MedXplore reduces false-old errors from 14.50\% to 0.80\%, demonstrating strong robustness under severe old-new ambiguity.

Figures

Figures reproduced from arXiv: 2607.27620 by Di Wu, Jianwei He, Jie Hao, Jingze Lu, Junhao Dong, Kailin Lyu, Lin Shu, Long Xiao, WenJie Hou.

Figure 1
Figure 1. Figure 1: Qualitative and quantitative comparison of MedXplore (Ours) against state-of-the-art (SOTA) methods. (a) Attention [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Overview of our MedXplore framework. MedXplore is mainly composed of representation learning, parametric [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison of lesion activation maps under [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Conceptual illustration of how our proposed Adap [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Kvasir ambiguity and bias analysis. (a) Represen [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Effect of different weights of 𝛼 and 𝛽 on OrganAMNIST [45]. The sensitivity to 𝐾 values and similarity threshold 𝛿 on OrganAMNIST [45] [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualisation of the embedding space with t-SNE [PITH_FULL_IMAGE:figures/full_fig_p008_7.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

52 extracted references · 1 canonical work pages

  1. [1]

    Kaidi Cao, Maria Brbic, and Jure Leskovec. 2021. Open-world semi-supervised learning.arXiv preprint arXiv:2102.03526(2021)

  2. [2]

    Xinzi Cao, Xiawu Zheng, Guanhong Wang, Weijiang Yu, Yunhang Shen, Ke Li, Yutong Lu, and Yonghong Tian. 2024. Solving the catastrophic forgetting problem in generalized category discovery. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 16880–16889

  3. [3]

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. 2021. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF international conference on computer vision. 9650–9660

  4. [4]

    Sua Choi, Dahyun Kang, and Minsu Cho. 2024. Contrastive mean-shift learning for generalized category discovery. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 23094–23104

  5. [5]

    Ankit Das, Chandan Gautam, Pritee Agrawal, Feng Yang, Yong Liu, and Ra- masamy Savitha. 2025. MedGCD: Generalized Category Discovery in Medical Imaging. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention. Springer, 427–437

  6. [6]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Im- agenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition. Ieee, 248–255

  7. [7]

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. 2019. Arcface: Additive angular margin loss for deep face recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 4690–4699

  8. [8]

    Ming Deng, Sijin Sun, Zihao Li, Xiaochuan Hu, and Xing Wu. 2025. Fmnet: Frequency-assisted mamba-like linear attention network for camouflaged object detection. In2025 IEEE International Conference on Multimedia and Expo (ICME). IEEE, 1–6

  9. [9]

    Arjun D Desai, Beliz Gunel, Batu M Ozturkler, Harris Beg, Shreyas Vasanawala, Brian A Hargreaves, Christopher Ré, John M Pauly, and Akshay S Chaudhari

  10. [10]

    Alexey Dosovitskiy. 2020. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020)

  11. [11]

    Wei Feng, Sijin Zhou, Yiwen Jiang, Feilong Tang, and Zongyuan Ge. 2025. Neighbor-Guided Unbiased Framework for Generalized Category Discovery in Medical Image Classification.IEEE Journal of Biomedical and Health Informat- ics(2025)

  12. [12]

    Enrico Fini, Enver Sangineto, Stéphane Lathuilière, Zhun Zhong, Moin Nabi, and Elisa Ricci. 2021. A unified objective for novel class discovery. InProceedings of the IEEE/CVF international conference on computer vision. 9284–9292

  13. [13]

    Minghan Fu, Na Zhang, Zhenxing Huang, Chao Zhou, Xu Zhang, Jianmin Yuan, Qiang He, Yongfeng Yang, Hairong Zheng, Dong Liang, et al. 2023. Oif-net: An optical flow registration-based pet/mr cross-modal interactive fusion network for low-count brain pet image denoising.IEEE Transactions on Medical Imaging 43, 4 (2023), 1554–1567

  14. [14]

    Chuanxing Geng, Sheng-jun Huang, and Songcan Chen. 2020. Recent advances in open set recognition: A survey.IEEE transactions on pattern analysis and machine intelligence43, 10 (2020), 3614–3631

  15. [15]

    Hao Guan, Pew-Thian Yap, Andrea Bozoki, and Mingxia Liu. 2024. Federated learning for medical image analysis: A survey.Pattern recognition151 (2024), 110424

  16. [16]

    Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, Andrea Vedaldi, and An- drew Zisserman. 2021. Autonovel: Automatically discovering and learning novel visual categories.IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 10 (2021), 6767–6781

  17. [17]

    Kai Han, Andrea Vedaldi, and Andrew Zisserman. 2019. Learning to discover novel visual categories via deep transfer clustering. InProceedings of the IEEE/CVF international conference on computer vision. 8401–8409

  18. [18]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition. 770–778

  19. [19]

    Yen-Chang Hsu, Zhaoyang Lv, and Zsolt Kira. 2017. Learning to cluster in order to transfer across domains and tasks.arXiv preprint arXiv:1711.10125(2017)

  20. [20]

    Alex Pappachen James and Belur V Dasarathy. 2014. Medical image fusion: A survey of the state of the art.Information fusion19 (2014), 4–19

  21. [21]

    Debesh Jha, Pia H Smedsrud, Michael A Riegler, Pål Halvorsen, Thomas de Lange, Dag Johansen, and Håvard D Johansen. 2020. Kvasir-seg: A segmented polyp dataset. InMultiMedia Modeling: 26th International Conference, MMM 2020, Daejeon, South Korea, January 5–8, 2020, Proceedings, Part II 26. Springer, 451–462

  22. [22]

    Jakob Nikolas Kather, Johannes Krisam, Pornpimol Charoentong, Tom Luedde, Esther Herpel, Cleo-Aron Weis, Timo Gaiser, Alexander Marx, Nektarios A Valous, Dyke Ferber, et al . 2019. Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study.PLoS medicine16, 1 (2019), e1002730

  23. [23]

    Philippe Lambin, Emmanuel Rios-Velazquez, Ralph Leijenaar, Sara Carvalho, Ruud GPM Van Stiphout, Patrick Granton, Catharina ML Zegers, Robert Gillies, Ronald Boellard, André Dekker, et al. 2012. Radiomics: extracting more informa- tion from medical images using advanced feature analysis.European journal of cancer48, 4 (2012), 441–446

  24. [24]

    Junghwan Lee, Cong Liu, Junyoung Kim, Zhehuan Chen, Yingcheng Sun, James R Rogers, Wendy K Chung, and Chunhua Weng. 2022. Deep learning for rare disease: A scoping review.Journal of biomedical informatics135 (2022), 104227

  25. [25]

    Haotang Li, Shengtao Guo, Kailin Lyu, Xiao Yang, Tianchen Chen, Jianqing Zhu, and Huanqiang Zeng. 2022. A challenging benchmark of anime style recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4721–4730

  26. [26]

    Mingqian Li, Zhiqian Yan, Miaoning Yan, Yaodong Liang, Qingmao Zhang, and Qiongxiong Ma. 2025. PFESA: FFT-Based Parameter-Free Edge and Structure Attention for Medical Image Segmentation. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 470–480

  27. [27]

    Shuhui Liu, Xinlong Liu, Xiaowei Liu, Chaobin Zeng, and Wei Liu. 2026. Fre- quency domain Attention Network for S/TEM Image Segmentation.Materials Today Nano(2026), 100772

  28. [28]

    Kailin Lyu, Jianwei He, Long Xiao, Jianing Zeng, Liang Fan, Lin Shu, and Jie Hao

  29. [29]

    Kailin Lyu, Long Xiao, Jianing Zeng, Junhao Dong, Xuexin Liu, Zhuojun Zou, Haoyue Yang, Lin Shu, and Jie Hao. 2026. TouchFormer: A Robust Transformer- Based Framework for Multimodal Material Perception. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 18496–18504. doi:10.1609/aaai.v40i22. 38915

  30. [30]

    Maria Ines Meyer, Ezequiel de la Rosa, Nuno Pedrosa de Barros, Roberto Paolella, Koen Van Leemput, and Diana M Sima. 2021. A contrast augmentation approach to improve multi-scanner generalization in MRI.Frontiers in neuroscience15 (2021), 708196

  31. [31]

    Abhishek Midya, Jayasree Chakraborty, Mithat Gönen, Richard KG Do, and Amber L Simpson. 2018. Influence of CT acquisition and reconstruction parame- ters on radiomic feature reproducibility.Journal of Medical Imaging5, 1 (2018), 011020–011020

  32. [32]

    Konstantin Pogorelov, Kristin Ranheim Randel, Carsten Griwodz, Sigrun Losada Eskeland, Thomas de Lange, Dag Johansen, Concetto Spampinato, Duc-Tien Dang-Nguyen, Mathias Lux, Peter Thelin Schmidt, et al. 2017. Kvasir: A multi- class image dataset for computer aided gastrointestinal disease detection. In Proceedings of the 8th ACM on Multimedia Systems Conf...

  33. [33]

    Nan Pu, Zhun Zhong, and Nicu Sebe. 2023. Dynamic conceptional contrastive learning for generalized category discovery. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7579–7588

  34. [34]

    Michael Roberts, Derek Driggs, Matthew Thorpe, Julian Gilbey, Michael Yeung, Stephan Ursprung, Angelica I Aviles-Rivero, Christian Etmann, Cathal McCague, Lucian Beer, et al. 2021. Common pitfalls and recommendations for using ma- chine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans.Nature Machine Intelligence3, 3 (...

  35. [35]

    Saeed Shurrab and Rehab Duwairi. 2022. Self-supervised learning methods and applications in medical imaging analysis: A survey.PeerJ Computer Science8 (2022), e1045

  36. [36]

    Fan Song, Xiao Song, Youdan Feng, Guangda Fan, Yangyang Sun, Peng Zhang, Jinkai Li, Fei Liu, and Guanglei Zhang. 2023. Radiomics feature analysis and model research for predicting histopathological subtypes of non-small cell lung cancer on CT images: A multi-dataset study.Medical physics50, 7 (2023), 4351– 4365

  37. [37]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

  38. [38]

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. 2022. Generalized category discovery. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7492–7501

  39. [39]

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. 2022. Open-Set Recognition: a Good Closed-Set Classifier is All You Need? arXiv:2110.06207 [cs.CV] https://arxiv.org/abs/2110.06207

  40. [40]

    Hongjun Wang, Sagar Vaze, and Kai Han. 2024. Sptnet: An efficient alternative framework for generalized category discovery with spatial prompt tuning.arXiv preprint arXiv:2403.13684(2024)

  41. [41]

    Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu. 2018. Cosface: Large margin cosine loss for deep face recognition. InProceedings of the IEEE conference on computer vision and pattern recognition. 5265–5274

  42. [42]

    Xiu-Shen Wei, Yi-Zhe Song, Oisin Mac Aodha, Jianxin Wu, Yuxin Peng, Jinhui Tang, Jian Yang, and Serge Belongie. 2021. Fine-grained image analysis with deep learning: A survey.IEEE transactions on pattern analysis and machine intelligence 44, 12 (2021), 8927–8948

  43. [43]

    Xin Wen, Bingchen Zhao, and Xiaojuan Qi. 2023. Parametric classification for generalized category discovery: A baseline study. InProceedings of the IEEE/CVF international conference on computer vision. 16590–16600

  44. [44]

    Long Xiao, Kailin Lyu, Jianwei He, Jianing Zeng, Lin Shu, and Jie Hao. 2026. Tac- Expert: A Pseudo-Temporal Mixture-of-Experts Framework for Open-Set Tactile Object Recognition. In2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 3761–3765. doi:10.1109/ICASSP55912.2026. 11462662

  45. [45]

    Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, and Bingbing Ni. 2023. Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification.Scientific Data10, 1 (2023), 41

  46. [46]

    Xiaohan Yu, Yang Zhao, Yongsheng Gao, Xiaohui Yuan, and Shengwu Xiong

  47. [47]

    Fan Zhang, Zhiwei Gu, and Hua Wang. 2026. Decoding with structured aware- ness: integrating directional, frequency-spatial, and structural attention for med- ical image segmentation. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 12421–12429

  48. [48]

    Bingchen Zhao and Kai Han. 2021. Novel visual category discovery with dual ranking statistics and mutual knowledge distillation.Advances in Neural Infor- mation Processing Systems34 (2021), 22982–22994

  49. [49]

    InProceedings of the IEEE/CVF international conference on computer vision

    Benchmark platform for ultra-fine-grained visual categorization beyond human performance. InProceedings of the IEEE/CVF international conference on computer vision. 10285–10295

  50. [50]

    Zhun Zhong, Linchao Zhu, Zhiming Luo, Shaozi Li, Yi Yang, and Nicu Sebe. 2021. Openmix: Reviving known knowledge for discovering novel visual categories in an open world. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 9462–9470

  51. [52]

    Zhun Zhong, Enrico Fini, Subhankar Roy, Zhiming Luo, Elisa Ricci, and Nicu Sebe. 2021. Neighborhood contrastive learning for novel class discovery. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 10867–10875

  52. [2021]

    Vortex: Physics-driven data augmentations using consistency training for robust accelerated mri reconstruction.arXiv preprint arXiv:2111.02549(2021)