Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

A Retrospective Systematic Study on Hierarchical Sparse Query Transformer-assisted Ultrasound Screening for Early Hepatocellular Carcinoma

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a hierarchical sparse-query transformer for ultrasound HCC screening reaches 95.38% multi-center AUC and matches senior radiologists while beating junior ones.

desk verdict A serious, unusually thorough clinical AI study whose headline multi-center AUC is undercut because the same test set was used for hyperparameter selection and patient-level deduplication is not documented. read the letter →

arxiv 2502.03772 v2 pith:FMVIR2MX submitted 2025-02-06 cs.CV cs.AI

Chaoyin She , Ruifang Lu , Danni He , Jiayi Lv , Yadan Lin , Meiqing Cheng , Hui Huang , Fengyu Ye
show 3 more authors
Lida Chen Wei Wang Qinghua Huang
This is my paper · ORCID
classification cs.CVcs.AI
keywords hepatocellularcarcinomaultrasoundscreeningvisiontransformermixtureofexpertscomputer-aideddiagnosishierarchicalsparsequeryCNN-ViThybridmedicalimageclassification
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 HSQformer, a hybrid architecture that combines convolutional local features with vision-transformer global context through a hierarchical set of learned query tokens and sparse mixture-of-experts routing. It claims this model outperforms state-of-the-art CNN and ViT baselines on ultrasound screening for early hepatocellular carcinoma, reaching 95.38% AUC in a multi-center test and 88.32% AUC on a high-risk hepatitis cohort. The authors also report that HSQformer matches the accuracy of senior radiologists and clearly outperforms junior radiologists, which matters because conventional ultrasound sensitivity depends heavily on reader experience. If these results hold, the model offers a route to standardizing HCC screening and reducing missed early cancers.

What carries the argument

The carrying mechanism is the Cross-Self-attention Mixed experts (CSM) block: cross-attention lets learned query tokens pull information from CNN and ViT feature maps, self-attention refines the pooled tokens, and a top-k Mixture-of-Experts layer routes each token to a small subset of specialist MLPs. CSM blocks are stacked in four stages that query feature maps at strides 4, 8, 16, and 32, so low-level texture and high-level semantic information are both retained. The learned query count and dimension act as a compression bottleneck that reduces redundancy in the long token sequences produced by the early stages.

What would settle it

Run the same 5-fold protocol with a test set drawn exclusively from hospitals that contributed no training images, with hyperparameters frozen before any test evaluation; if HSQformer-B then no longer beats ConvNext or SwinTransformer on AUC, the paper's central claim of superior generalization fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that CNN and ViT features can be integrated hierarchically, without changing the backbone structures, by projecting multi-scale feature maps into a shared latent space and letting a small set of learned query tokens extract diagnosis-relevant information at each stage. The resulting HSQformer-B reaches 95.38% AUC (88.29% accuracy, 90.38% recall) on eight-hospital multi-center data, outperforming all compared CNNs and ViTs, and 94.09% recall on high-risk patients, where it exceeds every radiologist in the comparison and matches senior-level diagnostic quality. The paper reads this as evidence that AI-assisted ultrasound screening can reach clinical-grade accuracy while being computationally sparse.

Load-bearing premise

The load-bearing assumption is that the multi-center test set is genuinely external to training, yet it includes data from the same hospital that supplied the training cohort and the paper does not document patient-level deduplication between the two.

Editorial extensions

If this is right

  • A deployed HSQformer could give non-specialist ultrasound operators senior-level diagnostic support, potentially raising early HCC detection in settings without expert radiologists.
  • The high recall (94.09%) on high-risk hepatitis patients directly targets the screening failure mode that matters most: missed early cancers in surveillance populations.
  • Sparse MoE routing means only a subset of experts runs at inference, so the accuracy gain does not require a proportional increase in per-image computation.
  • The four-stage query design should transfer to other lesion-classification tasks where both texture detail and global context matter, since it needs no changes to the backbone.
  • Matching senior radiologists on a multi-center set is evidence, if replicated prospectively, that AI-assisted screening could standardize HCC surveillance across institutions.

Reading between the lines

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

  • Because the headline hyperparameters were tuned on the same multi-center test set that produced the 95.38% AUC, a fair external evaluation on institutions never seen during development would be needed to know how much of the gain is architectural rather than selection.
  • The reported robustness to stage-ratio changes suggests the core design is insensitive to depth allocation; a natural test is freezing the architecture and tuning only query count and dimension on a new ultrasound dataset.
  • The attention heatmaps suggest the model focuses on perilesional tissue; adding radiologist lesion bounding boxes as supervision could test whether explicit localization improves both accuracy and clinical trust.
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 / 4 minor

Summary. The manuscript proposes HSQformer, a hybrid CNN–ViT architecture with hierarchical sparse querying and Mixture-of-Experts, for ultrasound-based screening of hepatocellular carcinoma. It reports experiments on three retrospective cohorts (single-center, multi-center, high-risk), compares against twelve SOTA models and six radiologists, and releases code. The headline result is HSQformer-B achieving 95.38% AUC on the multi-center test set, outperforming the best baseline (ConvNeXt, 94.45% AUC) and matching senior radiologists while exceeding junior radiologists.

Significance. If the reported multi-center performance were obtained from a genuinely held-out, pre-specified evaluation, the study would be a solid contribution to AI-assisted HCC screening: it uses a sizable retrospective dataset, patient-level cross-validation, three clinical scenarios, a human-machine comparison, and open-source code. The proposed architecture is modular and the ablation coverage is broad. However, the current evaluation protocol contains two load-bearing gaps — possible patient overlap between training and multi-center test sets and model selection directly on that same test set — that make the headline AUC and the claimed superiority over SOTA models unverified. The significance is therefore conditional on correcting the validation.

major comments (4)
  1. [Section 4.1 and Table 1] The multi-center test set is described as the UE-MICAP dataset, which per Section 4.1 includes the First Affiliated Hospital of Sun Yat-sen University — the same institution that provided the entire training set (Table 1). The paper does not document any patient-level or image-level de-duplication between the training cohort and the multi-center test cohort. Consequently, the multi-center test cannot be considered a genuinely external validation set, and a patient scanned at the First Affiliated Hospital could appear in both training and test splits. The same overlap applies to the high-risk patient test, whose data source is also the First Affiliated Hospital. Please either demonstrate explicit patient-level separation (e.g., by reporting the deduplication protocol and the number of excluded patients) or re-compute the results on a test set that excludes all patients from the training institution.
  2. [Section 4.3 and Table 4] Ablation results are explicitly obtained on the multi-center test set ('all ablation study results are obtained on the multi-center test set,' Section 4.3). Figures 5–8 and the accompanying text use that test set to select the stage ratio (2:2:6:2), query number (200), query dimension (384), number of experts (4), and Top-K strategy (Top-2) for HSQformer-B. The reported 95.38% AUC is therefore a post-selection estimate, not a pre-specified evaluation on an independent test set, and the comparison against baselines that did not receive the same test-set tuning is biased in favor of HSQformer. Please provide a nested or separate held-out evaluation (e.g., a validation split for hyperparameter selection and a final test set used only once), or at minimum report the performance of the final configuration on a truly untouched external cohort.
  3. [Section 4.2 and Table 2] The model's reported AUC is much higher on the multi-center test (95.38±0.33) than on the single-center test from a different institution (83.83±0.96). If the multi-center set were a clean external and more heterogeneous cohort, one would generally expect equal or lower performance than on the training-institution distribution, not an 11.5-point improvement. This discrepancy is consistent with the overlap and test-set tuning issues above and should be discussed explicitly. Reporting per-institution AUCs within the multi-center test (especially separating the First Affiliated Hospital from the other sites) would help determine whether the high multi-center number is driven by leakage or by genuine cross-site generalization.
  4. [Section 4.2.2 and Tables 2 and 3] The abstract and Section 4.2.2 state that HSQformer 'significantly surpasses' junior radiologists and outperforms SOTA models, but no statistical significance tests are reported for the pairwise comparisons. In Table 2, the multi-center AUC gap between HSQformer-B (95.38±0.33) and ConvNeXt (94.45±0.53) is smaller than the fold-to-fold variability, and in Table 3 the senior/junior averages are based on only three radiologists each. Please provide paired tests (e.g., DeLong test for AUC or a permutation test across folds) and, for the radiologist comparison, the number of readings and a suitable effect-size measure, before claiming statistical significance.
minor comments (4)
  1. [Figure 9] The legend spells the model name 'HSQfomer' in multiple places; this should be corrected to 'HSQformer'.
  2. [Section 4.2.2 and Table 3] The bullet points contain numeric inconsistencies with Table 3. For example, junior radiologists' average Precision is 85.47±2.17 in Table 3, but the text reports 88.62% as the junior value and 87.62% vs. 88.62% for the HSQformer-B comparison. Please reconcile the text with the table.
  3. [Section 4.3.1 and Figure 6] Several ablation numbers are read from plots; please include the exact values in a table so that the reported trends are reproducible.
  4. [Section 1 and Related Work] The claim of being the 'first systematic exploration of AI-assisted ultrasound screening for HCC' would benefit from a more detailed comparison with prior deep-learning HCC ultrasound studies in Related Work.

Circularity Check

1 steps flagged · score 5.0 of 10

The headline multi-center AUC is partially circular: HSQformer-B's hyperparameters (stage ratio, query number/dimension, MoE experts, Top-K) were selected by ablations run on the same multi-center test set whose 95.38% AUC is then reported as the model's performance.

  1. fitted input called prediction [Section 4.3.1-4.3.2, Figures 5-8 and Table 4; Section 4.2.1, Table 2]
    "Unless otherwise specified, all ablation study results are obtained on the multi-center test set, as this approach better validates the generalizability of the model. ... based on the experimental data, we identified the most effective ratio configuration as 2:2:6:2, which optimizes the model's performance ... under the Top-2 strategy, MoE with four and eight experts shows similar performance, indicating that four experts might provide an optimal balance between performance and efficiency."

    The final HSQformer-B configuration in Table 4 (Query number 200, Query dimension 384, Stage ratio 2:2:6:2, Number of experts 4, Top-K Top-2) is exactly the configuration that the ablation curves in Figures 5-8 select using the multi-center test set. Table 2 then reports HSQformer-B's multi-center AUC as 95.38±0.33 on that same multi-center test set. The reported 'validation' number is therefore not a pre-specified measurement on an independent holdout; it is the result of selecting hyperparameters by maximizing performance on the very benchmark used to claim superiority. Part of the 0.93-point gap over ConvNext (94.45±0.53) reflects test-set selection rather than a purely out-of-sample prediction, making the central claim statistically forced rather than independently measured.

full rationale

The paper contains no self-definitional derivation: HSQformer's architecture is defined independently of the HCC labels, and the reported AUCs are empirical measurements rather than quantities computed from the model equations. There are no load-bearing self-citations, imported uniqueness theorems, or ansatz-smuggled-via-citation steps. The one genuine circularity is selection-on-the-test-set. Section 4.3 states that all ablations are obtained on the multi-center test set; Figures 5-8 and the accompanying text select the stage ratio 2:2:6:2, query number/dimension, four experts, and Top-2 strategy on the basis of that test set, and Table 4 hard-codes those choices into HSQformer-B. Table 2 then reports HSQformer-B's 95.38±0.33 multi-center AUC on the same set, so the headline multi-center result is not independent of the model-selection procedure. This is a partial circularity rather than a formal reduction: the single-center test and high-risk patient test provide some independent evidence for the architecture's utility, and the high-risk human-machine comparison does not rely on the multi-center selection. In addition, Table 1 indicates that the multi-center pool includes the First Affiliated Hospital of Sun Yat-sen University, the same institution that supplied the training data, and the paper does not document patient-level deduplication between training and test cohorts; this is a data-contamination concern that further weakens the external-validation claim, though it is not itself a definitional circularity. Overall, one partial circular step, with no self-citation load-bearing, warrants a score of 5.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

No new physical entities are introduced. The model hyperparameters above are tuned on the test data, which is the main circularity burden.

free parameters (5)
  • Query number Q = 200 (Base), 400 (Large)
    Chosen by ablations on the multi-center test set; affects token capacity.
  • Query dimension D = 384 (Base), 768 (Large)
    Chosen by ablations on the multi-center test set; larger dimension improves F1/AUC but saturates.
  • Stage ratio L1:L2:L3:L4 = 2:2:6:2
    Selected from six tested ratios; variation in AUC was small but this ratio was chosen as optimal.
  • Number of MoE experts E = 4 (Base), 8 (Large)
    Ablations on multi-center test showed 4 or 8 experts with Top-2 are best.
  • Top-K routing = Top-2 (Base/Large)
    Chosen based on multi-center ablation; Top-2 outperformed Top-1.
assumptions (3)
  • domain assumption Patient-level 5-fold cross-validation prevents data leakage within the training set
    Assumed to ensure no patient appears in both training and validation within the training set; not verified at test-set level (multi-center test may contain training-hospital patients).
  • domain assumption Ground truth labels (benign/malignant) are correct and consistent
    Labels from clinical/pathological diagnosis; no central review or inter-observer agreement is reported.
  • domain assumption The multi-center test set is independent of the training set
    The paper does not document patient-level deduplication between the training cohort and the UE-MICAP multi-center test set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Retrospective Systematic Study on Hierarchical Sparse Query Transformer-assisted Ultrasound Screening for Early Hepatocellular Carcinoma." pith.science (2026). https://pith.science/paper/FMVIR2MX

@misc{pith2026250203772,
  author       = {Pith},
  title        = {Pith review of: A Retrospective Systematic Study on Hierarchical Sparse Query Transformer-assisted Ultrasound Screening for Early Hepatocellular Carcinoma},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FMVIR2MX}},
  note         = {Machine review of arXiv:2502.03772}
}
read the original abstract

Hepatocellular carcinoma (HCC), ranking as the third leading cause of cancer-related mortality worldwide, demands urgent improvements in early detection to enhance patient survival. While ultrasound remains the preferred screening modality due to its cost-effectiveness and real-time capabilities, its sensitivity (59%-78%) heavily relies on radiologists' expertise, leading to inconsistent diagnostic outcomes and operational inefficiencies. Recent advancements in AI technology offer promising solutions to bridge this gap. This study introduces the Hierarchical Sparse Query Transformer (HSQformer), a novel hybrid architecture that synergizes CNNs' local feature extraction with Vision Transformers' global contextual awareness through latent space representation and sparse learning. By dynamically activating task-specific experts via a Mixture-of-Experts (MoE) framework, HSQformer achieves hierarchical feature integration without structural redundancy. Evaluated across three clinical scenarios: single-center, multi-center, and high-risk patient cohorts, HSQformer outperforms state-of-the-art models (e.g., 95.38% AUC in multi-center testing) and matches senior radiologists' diagnostic accuracy while significantly surpassing junior counterparts. These results highlight the potential of AI-assisted tools to standardize HCC screening, reduce dependency on human expertise, and improve early diagnosis rates. The full code is available at https://github.com/Asunatan/HSQformer.

Figures

Figures reproduced from arXiv: 2502.03772 by the authors.

Figure 1
Figure 1. The improvement brought by simply combining CNN [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. model architecture. Overview of the HSQformer, integrating CNN and ViT features through a hierarchical sparse querying framework for efficient diagnosis. Given an input image X ∈ R C×H×W , we first apply two distinct feature extractors to capture multi-scale rep￾resentations. The resulting feature maps are denoted as {fc1, fc2, fc3, fc4} for ConvNeXt and {fs1, fs2, fs3, fs4} for SwinTransformer. Each set of feature … view at source ↗
Figure 3
Figure 3. Visualization of Human-Machine Diagnostic Efficacy [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Assessing the Impact of Stage Schemes on Model Performance. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Influence of proportional parameters on model perfor [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Systematic Analysis of Query Numbers and Dimensions [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Serial vs. Parallel Attention Mechanisms: Structural Diagram and Performance Comparison [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Trend Curves of Number of Experts and Top-K Selection [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Systematic Comparative Analysis of SOTA Models Using 5-Fold Cross-Validation on Three Test Datasets [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Heatmap Analysis of Attention Maps for Ultrasound HCC Diagnosis. [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding

    eess.IV 2026-01 conditional novelty 5.0 of 10

    A medical-image benchmark that scores the step-by-step reasoning chains of multimodal LLMs shows current models explain poorly and chain-of-thought prompting frequently reduces diagnostic accuracy.

Reference graph

Works this paper leans on

52 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Global cancer statistics 2022: Globocan estimates of incidence and mortality world- wide for 36 cancers in 185 countries,

    F. Bray, M. Laversanne, H. Sung, J. Ferlay, R. L. Siegel, I. Soerjomataram, and A. Jemal, “Global cancer statistics 2022: Globocan estimates of incidence and mortality world- wide for 36 cancers in 185 countries,” CA: a cancer journal for clinicians, vol. 74, no. 3, pp. 229–263, 2024. 1

    work page 2022

  2. [2]

    Comparative 13-year meta- analysis of the sensitivity and positive predictive value of ultrasound, ct, and mri for detecting hepatocellular carci- noma,

    R. F. Hanna, V . Z. Miloushev, A. Tang, L. A. Finklestone, S. Z. Brejt, R. S. Sandhu, C. S. Santillan, T. Wolfson, A. Gamst, and C. B. Sirlin, “Comparative 13-year meta- analysis of the sensitivity and positive predictive value of ultrasound, ct, and mri for detecting hepatocellular carci- noma,” Abdominal radiology, vol. 41, pp. 71–90, 2016. 1

    work page 2016

  3. [3]

    Imaging techniques for the diagnosis of hepatocellular car- cinoma: a systematic review and meta-analysis,

    R. Chou, C. Cuevas, R. Fu, B. Devine, N. Wasson, A. Gins- burg, B. Zakher, M. Pappas, E. Graham, and S. D. Sullivan, “Imaging techniques for the diagnosis of hepatocellular car- cinoma: a systematic review and meta-analysis,” Annals of internal medicine, vol. 162, no. 10, pp. 697–711, 2015. 1

    work page 2015

  4. [4]

    An image is worth 16x16 words: Trans- formers for image recognition at scale,

    A. Dosovitskiy, “An image is worth 16x16 words: Trans- formers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020. 1, 7

    arXiv 2010

  5. [5]

    A convnet for the 2020s,

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 11 976–11 986. 2, 7, 13

    work page 2022

  6. [6]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 10 012–10 022. 2, 3, 7, 13

    work page 2021

  7. [7]

    Exploring the synergies of hybrid cnns and vits architectures for computer vision: A survey,

    H. Yunusa, S. Qin, A. H. A. Chukkol, A. A. Yusuf, I. Bello, and A. Lawan, “Exploring the synergies of hybrid cnns and vits architectures for computer vision: A survey,” arXiv preprint arXiv:2402.02941, 2024. 2

    arXiv 2024

  8. [8]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE confer- ence on computer vision and pattern recognition , 2016, pp. 770–778. 2, 6, 7

    work page 2016

Show all 52 references
  1. [9]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in Proceed- ings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 4700–4708. 2, 6, 7

    2017

  2. [10]

    Scaling up your kernels to 31x31: Revisiting large kernel design in cnns,

    X. Ding, X. Zhang, J. Han, and G. Ding, “Scaling up your kernels to 31x31: Revisiting large kernel design in cnns,” in Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2022, pp. 11 963–11 975. 2

    2022

  3. [11]

    More convnets in the 2020s: Scaling up kernels beyond 10 51x51 using sparsity,

    S. Liu, T. Chen, X. Chen, X. Chen, Q. Xiao, B. Wu, T. K¨arkk¨ainen, M. Pechenizkiy, D. Mocanu, and Z. Wang, “More convnets in the 2020s: Scaling up kernels beyond 10 51x51 using sparsity,” arXiv preprint arXiv:2207.03620 ,

    arXiv

  4. [12]

    Pelk: Parameter-efficient large kernel convnets with peripheral convolution,

    H. Chen, X. Chu, Y . Ren, X. Zhao, and K. Huang, “Pelk: Parameter-efficient large kernel convnets with peripheral convolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 5557–5567. 2

    2024

  5. [13]

    Acnet: Strengthening the kernel skeletons for powerful cnn via asymmetric con- volution blocks,

    X. Ding, Y . Guo, G. Ding, and J. Han, “Acnet: Strengthening the kernel skeletons for powerful cnn via asymmetric con- volution blocks,” in Proceedings of the IEEE/CVF interna- tional conference on computer vision, 2019, pp. 1911–1920. 2

    2019

  6. [14]

    Repvgg: Making vgg-style convnets great again,

    X. Ding, X. Zhang, N. Ma, J. Han, G. Ding, and J. Sun, “Repvgg: Making vgg-style convnets great again,” in Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 13 733–13 742. 2

    2021

  7. [15]

    Repvit: Re- visiting mobile cnn from vit perspective,

    A. Wang, H. Chen, Z. Lin, J. Han, and G. Ding, “Repvit: Re- visiting mobile cnn from vit perspective,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 15 909–15 920. 2

    2024

  8. [16]

    Multi-scale context aggregation by dilated convolu- tions,

    F. Yu, “Multi-scale context aggregation by dilated convolu- tions,” arXiv preprint arXiv:1511.07122, 2015. 2

    2015 arXiv

  9. [17]

    Deformable convolutional networks,

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, and Y . Wei, “Deformable convolutional networks,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 764–773. 2

    2017

  10. [18]

    Deformable convnets v2: More deformable, better results,

    X. Zhu, H. Hu, S. Lin, and J. Dai, “Deformable convnets v2: More deformable, better results,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 9308–9316. 2

    2019

  11. [19]

    Internimage: Exploring large-scale vision foundation models with deformable convolutions,

    W. Wang, J. Dai, Z. Chen, Z. Huang, Z. Li, X. Zhu, X. Hu, T. Lu, L. Lu, H. Liet al., “Internimage: Exploring large-scale vision foundation models with deformable convolutions,” in Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2023, pp. 14...

    2023

  12. [20]

    Squeeze-and-excitation net- works,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation net- works,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141. 3

    2018

  13. [21]

    Cbam: Convo- lutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convo- lutional block attention module,” in Proceedings of the Eu- ropean conference on computer vision (ECCV), 2018, pp. 3–

    2018

  14. [22]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017. 3

    2017

  15. [23]

    Cvt: Introducing convolutions to vision trans- formers,

    H. Wu, B. Xiao, N. Codella, M. Liu, X. Dai, L. Yuan, and L. Zhang, “Cvt: Introducing convolutions to vision trans- formers,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 22–31. 3

    2021

  16. [24]

    Escaping the big data paradigm with compact trans- formers,

    A. Hassani, S. Walton, N. Shah, A. Abuduweili, J. Li, and H. Shi, “Escaping the big data paradigm with compact trans- formers,” arXiv preprint arXiv:2104.05704, 2021. 3

    2021 arXiv

  17. [25]

    Slide- transformer: Hierarchical vision transformer with local self- attention,

    X. Pan, T. Ye, Z. Xia, S. Song, and G. Huang, “Slide- transformer: Hierarchical vision transformer with local self- attention,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2023, pp. 2082–

    2023

  18. [26]

    Cswin transformer: A general vi- sion transformer backbone with cross-shaped windows,

    X. Dong, J. Bao, D. Chen, W. Zhang, N. Yu, L. Yuan, D. Chen, and B. Guo, “Cswin transformer: A general vi- sion transformer backbone with cross-shaped windows,” in Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2022, pp. 12 124–12 134. 3, 6, 7, 13

    2022

  19. [27]

    P2t: Pyramid pooling transformer for scene understanding,

    Y .-H. Wu, Y . Liu, X. Zhan, and M.-M. Cheng, “P2t: Pyramid pooling transformer for scene understanding,”IEEE transac- tions on pattern analysis and machine intelligence , vol. 45, no. 11, pp. 12 760–12 771, 2022. 3

    2022

  20. [28]

    Pvt v2: Improved baselines with pyramid vision transformer,

    W. Wang, E. Xie, X. Li, D.-P. Fan, K. Song, D. Liang, T. Lu, P. Luo, and L. Shao, “Pvt v2: Improved baselines with pyramid vision transformer,” Computational Visual Media , vol. 8, no. 3, pp. 415–424, 2022. 3, 7

    2022

  21. [29]

    Metaformer is actually what you need for vi- sion,

    W. Yu, M. Luo, P. Zhou, C. Si, Y . Zhou, X. Wang, J. Feng, and S. Yan, “Metaformer is actually what you need for vi- sion,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , 2022, pp. 10 819– 10 829. 3

    2022

  22. [30]

    Focal self-attention for local-global interactions in vision transformers,

    J. Yang, C. Li, P. Zhang, X. Dai, B. Xiao, L. Yuan, and J. Gao, “Focal self-attention for local-global interactions in vision transformers,” arXiv preprint arXiv:2107.00641 ,

    arXiv

  23. [31]

    Mobilevit: light-weight, general-purpose, and mobile-friendly vision transformer,

    S. Mehta and M. Rastegari, “Mobilevit: light-weight, general-purpose, and mobile-friendly vision transformer,” arXiv preprint arXiv:2110.02178, 2021. 3

    2021 arXiv

  24. [32]

    Medical transformer: Gated axial-attention for medi- cal image segmentation,

    J. M. J. Valanarasu, P. Oza, I. Hacihaliloglu, and V . M. Pa- tel, “Medical transformer: Gated axial-attention for medi- cal image segmentation,” in Medical image computing and computer assisted intervention–MICCAI 2021: 24th inter- national conference, Strasbourg, France, Sep...

    2021

  25. [33]

    Co- ordination among neural modules through a shared global workspace,

    A. Goyal, A. Didolkar, A. Lamb, K. Badola, N. R. Ke, N. Ra- haman, J. Binas, C. Blundell, M. Mozer, and Y . Bengio, “Co- ordination among neural modules through a shared global workspace,” arXiv preprint arXiv:2103.01197, 2021. 3

    2021 arXiv

  26. [34]

    Set transformer: A framework for attention-based permutation-invariant neural networks,

    J. Lee, Y . Lee, J. Kim, A. Kosiorek, S. Choi, and Y . W. Teh, “Set transformer: A framework for attention-based permutation-invariant neural networks,” in International conference on machine learning. PMLR, 2019, pp. 3744–

    2019

  27. [35]

    Perceiver: General perception with iterative attention,

    A. Jaegle, F. Gimeno, A. Brock, O. Vinyals, A. Zisserman, and J. Carreira, “Perceiver: General perception with iterative attention,” in International conference on machine learning. PMLR, 2021, pp. 4651–4664. 3

    2021

  28. [36]

    Involution: Inverting the inherence of convolution for visual recognition,

    D. Li, J. Hu, C. Wang, X. Li, Q. She, L. Zhu, T. Zhang, and Q. Chen, “Involution: Inverting the inherence of convolution for visual recognition,” inProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , 2021, pp. 12 321–12 330. 3 11

    2021

  29. [37]

    V olo: Vi- sion outlooker for visual recognition,

    L. Yuan, Q. Hou, Z. Jiang, J. Feng, and S. Yan, “V olo: Vi- sion outlooker for visual recognition,” IEEE transactions on pattern analysis and machine intelligence, vol. 45, no. 5, pp. 6575–6586, 2022. 3

    2022

  30. [38]

    Learned queries for efficient local attention,

    M. Arar, A. Shamir, and A. H. Bermano, “Learned queries for efficient local attention,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 10 841–10 852. 3, 4

    2022

  31. [39]

    Modeling task relationships in multi-task learning with multi-gate mixture-of-experts,

    J. Ma, Z. Zhao, X. Yi, J. Chen, L. Hong, and E. H. Chi, “Modeling task relationships in multi-task learning with multi-gate mixture-of-experts,” in Proceedings of the 24th ACM SIGKDD international conference on knowledge dis- covery & data mining, 2018, pp. 1930–1939. 3, 4

    2018

  32. [40]

    Deepseekmoe: Towards ul- timate expert specialization in mixture-of-experts language models,

    D. Dai, C. Deng, C. Zhao, R. Xu, H. Gao, D. Chen, J. Li, W. Zeng, X. Yu, Y . Wuet al., “Deepseekmoe: Towards ul- timate expert specialization in mixture-of-experts language models,” arXiv preprint arXiv:2401.06066, 2024. 3

    2024 arXiv

  33. [41]

    Llama-moe: Building mixture-of-experts from llama with continual pre-training,

    T. Zhu, X. Qu, D. Dong, J. Ruan, J. Tong, C. He, and Y . Cheng, “Llama-moe: Building mixture-of-experts from llama with continual pre-training,” in Proceedings of the 2024 Conference on Empirical Methods in Natural Lan- guage Processing, 2024, pp. 15 913–15 923. 3

    2024

  34. [42]

    Llama-moe v2: Exploring sparsity of llama from perspec- tive of mixture-of-experts with post-training,

    X. Qu, D. Dong, X. Hu, T. Zhu, W. Sun, and Y . Cheng, “Llama-moe v2: Exploring sparsity of llama from perspec- tive of mixture-of-experts with post-training,”arXiv preprint arXiv:2411.15708, 2024. 3

    2024 arXiv

  35. [43]

    A survey on mixture of experts,

    W. Cai, J. Jiang, F. Wang, J. Tang, S. Kim, and J. Huang, “A survey on mixture of experts,” arXiv preprint arXiv:2407.06204, 2024. 3

    2024 arXiv

  36. [44]

    Masked autoencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick, “Masked autoencoders are scalable vision learners,” in Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 16 000–16 009. 4

    2022

  37. [45]

    Aggre- gated residual transformations for deep neural networks,

    S. Xie, R. Girshick, P. Doll ´ar, Z. Tu, and K. He, “Aggre- gated residual transformations for deep neural networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1492–1500. 7

    2017

  38. [46]

    Deep learning-based ar- tificial intelligence model to assist thyroid nodule diagno- sis and management: a multicentre diagnostic study,

    S. Peng, Y . Liu, W. Lv, L. Liu, Q. Zhou, H. Yang, J. Ren, G. Liu, X. Wang, X. Zhang et al., “Deep learning-based ar- tificial intelligence model to assist thyroid nodule diagno- sis and management: a multicentre diagnostic study,” The Lancet Digital Health, vol. 3, no. 4, pp....

    2021

  39. [47]

    Hiera: A hierarchical vision transformer without the bells-and-whistles,

    C. Ryali, Y .-T. Hu, D. Bolya, C. Wei, H. Fan, P.-Y . Huang, V . Aggarwal, A. Chowdhury, O. Poursaeed, J. Hoffman et al., “Hiera: A hierarchical vision transformer without the bells-and-whistles,” in International Conference on Machine Learning. PMLR, 2023, pp. 29 441–29 454. 7

    2023

  40. [48]

    Focal modulation net- works,

    J. Yang, C. Li, X. Dai, and J. Gao, “Focal modulation net- works,” Advances in Neural Information Processing Sys- tems, vol. 35, pp. 4203–4217, 2022. 7

    2022

  41. [49]

    Mpvit: Multi- path vision transformer for dense prediction,

    Y . Lee, J. Kim, J. Willette, and S. J. Hwang, “Mpvit: Multi- path vision transformer for dense prediction,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 7287–7296. 7

    2022

  42. [50]

    Yolov3: An incremental im- provement,

    A. Farhadi and J. Redmon, “Yolov3: An incremental im- provement,” inComputer vision and pattern recognition, vol

  43. [51]

    Davit: Dual attention vision transformers,

    M. Ding, B. Xiao, N. Codella, P. Luo, J. Wang, and L. Yuan, “Davit: Dual attention vision transformers,” in European conference on computer vision. Springer, 2022, pp. 74–92. 7 12 Appendix A. Architecture Details The HSQformer models (HSQformer-S, HSQformer-B, and HSQformer-L)...

    2022

  44. [1804]

    Springer Berlin/Heidelberg, Germany, 2018, pp. 1–6. 6

    2018

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.