Pith. sign in

REVIEW 3 major objections 2 minor 1 cited by

DesignCLIP claims CLIP-based models boost design patent search, but the supplied full text is a different paper.

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

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

The submitted package describes DesignCLIP in metadata, but the full text is a different paper, so the central results cannot be verified.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection The submission's abstract is DesignCLIP but its full text is an unrelated fetal ultrasound paper; there is nothing to review. the 3 major comments →

arxiv 2508.15297 v1 pith:3LHPAGQ5 submitted 2025-08-21 cs.CV cs.AI

DesignCLIP: Multimodal Learning with CLIP for Design Patent Understanding

Zhu Wang , Homaira Huda Shomee , Sathya N. Ravi , Sourav Medya This is my paper
classification cs.CV cs.AI
keywords design patentsCLIPvision-language modelsmultimodal retrievalpatent classificationcontrastive learningimage retrieval
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 reading

DesignCLIP is proposed as a unified vision-language framework for design patent analysis. It claims that adapting CLIP with class-aware contrastive learning, generated detailed captions, and multi-view image learning improves patent classification, image retrieval, and multimodal retrieval over baselines. If correct, this would make prior-art search more reliable and could inspire design innovation. However, the full text attached to this submission is not the DesignCLIP manuscript; it is a separate paper about fetal congenital heart defect classification from ultrasound videos. Consequently, the central claims cannot be checked against any experimental results in this document.

Core claim

The DesignCLIP abstract asserts that a CLIP-based model, trained on a large-scale U.S. design patent dataset with class-aware contrastive learning, generated captions, and multi-view image learning, consistently outperforms baseline and state-of-the-art models on patent classification, patent image retrieval, and multimodal retrieval. The claimed mechanism is that fusing image and text semantics resolves ambiguities inherent to sketch-like patent images. But the full text provided is entirely a different work, so the manuscript itself does not present or support this discovery.

What carries the argument

The central mechanism claimed is class-aware contrastive learning applied to CLIP's dual-encoder architecture, augmented with generated detailed captions for patent images and multi-view image learning. The intended effect is to align patent image features with class-level text semantics, enabling classification, unimodal retrieval, and multimodal retrieval within one framework.

Load-bearing premise

The submission assumes that the full text provided is the DesignCLIP manuscript containing its experimental protocol, datasets, and results; in fact, the full text is a different paper on fetal ultrasound classification, so the central claim stands without its supporting evidence in this package.

What would settle it

Open the supplied full text and search for design patents, caption generation, multi-view learning, or retrieval benchmarks; none appear. A single matching experiment description or dataset reference would refute the mismatch, and its absence means the DesignCLIP claims are unsupported by the provided document.

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

If this is right

  • If DesignCLIP works as claimed, prior-art search over design patents could become more accurate by using text descriptions alongside sketches, reducing ambiguity.
  • Patent classification could be automated more reliably, helping examiners and applicants identify overlapping designs.
  • Multimodal retrieval could offer designers diverse visual and textual references, potentially fostering creativity and innovation.
  • A single vision-language model could replace separate classification and retrieval pipelines for patent repositories.

Where Pith is reading between the lines

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

  • A practical extension would test whether caption quality, rather than model architecture, drives the claimed gains: swapping generated captions for simpler labels would isolate the contribution.
  • The mismatch between the abstract and the full text suggests the submitted package may be a workflow error; a corrected submission would be needed for any scientific evaluation.
  • If DesignCLIP's approach transfers to other sketch-heavy domains, the same class-aware contrastive strategy could improve trademark or industrial-design retrieval.
Share X Bluesky LinkedIn Reddit HN

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 / 2 minor

Summary. The manuscript package consists of an abstract that claims DesignCLIP, a unified CLIP-based framework for U.S. design patent classification and retrieval, together with a full text that is, in fact, a different paper: 'TPA: Temporal Prompt Alignment for Fetal Congenital Heart Defect Classification' by Taratynova et al. (arXiv:2508.15298v5). The full text describes a method for fetal ultrasound video classification, including a margin-hinge contrastive loss and a CVAE-style modulation module, and reports F1/AUC/calibration results on a private CHD dataset and EchoNet-Dynamic. The full text contains no U.S. design patent dataset, no patent classification or retrieval experiments, no generated captions, no multi-view image learning, and no DesignCLIP method or architecture. The central claims of the abstract are therefore unsupported by any reviewable content in the submitted package.

Significance. If the DesignCLIP claims were substantiated, the work could provide a valuable vision-language benchmark and method for design patent prior-art search. However, the submission as provided contains none of the required evidence: there are no tables comparing against baselines or state of the art, no dataset description, no ablation, and no reproducible code for the claimed framework. The anonymous codebase link is not verifiable within the package. The only experimental content belongs to the unrelated TPA paper on fetal ultrasound, which, whatever its merits, does not address the claimed problem. Thus the scientific significance of the central claim cannot be assessed from this submission.

major comments (3)
  1. [Full text title and abstract] The full text is a different manuscript with a different title, author list, and topic: it is the TPA paper on fetal congenital heart defect classification (arXiv:2508.15298v5). It contains no mention of design patents, patent classification, patent image retrieval, or multimodal patent retrieval. Consequently the abstract's claim that DesignCLIP 'consistently outperforms baseline and SOTA models in the patent domain on all tasks' is not backed by any experiment in this document.
  2. [Sections 2 and 4] The proposed method equations (Eqs. 1-7) describe TPA's contrastive loss and CVAE-style modulation for ultrasound video embedding, not DesignCLIP's class-aware contrastive learning, generated captions, or multi-view image learning. The results in Tables 1-4 report F1, AUC, ECE, and AECE on fetal CHD and EchoNet-Dynamic; these are not patent-domain results. No section, table, or equation in the submitted text provides evidence for the claimed DesignCLIP performance.
  3. [Abstract (codebase and dataset)] The submitted package relies on an anonymous repository link and a 'large-scale dataset of U.S. design patents' for reproducibility, but neither the dataset nor the code is described or accessible in the manuscript. Since the full text does not even mention PATENTCLIP or the design-patent dataset, the evaluation protocol for the main claim cannot be reconstructed from the submission.
minor comments (2)
  1. [Submission integrity] The package should identify clearly which document is the manuscript under review. The current package appears to contain a different paper's main text, which makes even basic bibliographic verification impossible.
  2. [Reference consistency] The DesignCLIP abstract refers to an anonymous PATENTCLIP codebase, while the full text's code link points to a public GitHub repository for TPA. These should be aligned if the authors intend to submit both papers as a combined package.

Circularity Check

0 steps flagged

No circularity found: DesignCLIP's claimed experiments are absent from the supplied full text, but that is a missing-evidence mismatch, not a circular derivation; the TPA text that is present is self-contained.

full rationale

The submission pairs a DesignCLIP abstract with a full text that is actually the TPA paper (arXiv:2508.15298v5) on fetal congenital heart defect classification. The abstract's claim that "DesignCLIP consistently outperforms baseline and SOTA models in the patent domain on all tasks" cannot be checked because no DesignCLIP experiments, datasets, or derivations appear in the supplied text. This is a completeness/integrity problem, not a circularity problem: there is no equation or fitted parameter in the supplied material that secretly equals a claimed prediction. Within the TPA paper, the derivation chain is not circular. Classification is performed by computing cosine similarities between the video embedding h and projected class-prompt embeddings (pi_c)_proj, with a temperature-scaled softmax and cross-entropy loss. The class prompts are fixed text descriptions such as "Can an abnormality be detected in the fetal heart 4CH?" and class-specific defect descriptions. The label information enters through the standard supervised losses (L_cls and the margin-hinge contrastive loss L_ctr), and the reported F1/AUC numbers are measured on held-out folds and on the external EchoNet-Dynamic test set. Thus the results are not forced by construction; the prompts are inputs to a standard vision-language classification framework, not retrofitted predictions. The uncertainty module CV AESM is adapted from prior work and is evaluated through ECE/AECE calibration metrics; it does not generate the headline classification scores. The only apparent self-citation (e.g., reference [1], which shares a TPA author) appears in the related-work discussion of uncertainty estimation and is not load-bearing for TPA's central contribution. Therefore the supplied TPA text exhibits no significant circularity. The DesignCLIP abstract's strong empirical claims are unsupported by the provided full text, but that unsupported-status concern belongs to correctness/verification, not to circularity analysis.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The DesignCLIP paper itself introduces no visible free parameters, axioms, or entities because its full text is missing. The only content available is the TPA paper. The most important axiom is false: the supplied full text does not match the abstract. For the TPA paper, the margin, alpha, and beta are hand-tuned, and the style modulation module is an invented entity with evidence only from its own experiments.

free parameters (2)
  • TPA loss weights alpha and beta = alpha = 0.5, beta = 0.2
    In the TPA paper, which is the only full text supplied, alpha and beta are set by hand. They are not fitted to the DesignCLIP target.
  • margin m in contrastive loss = 0.5 or 1.0
    The TPA paper tunes the margin for its contrastive loss. This belongs to the TPA text, not to DesignCLIP, which has no derivation in the supplied full text.
axioms (2)
  • ad hoc to paper The full text of arXiv 2508.15297 corresponds to DesignCLIP as described in the abstract.
    Unstated and false: the full text supplied is titled TPA and is about fetal congenital heart defect classification. All experimental claims in the abstract are unverifiable as a result.
  • domain assumption A design patent image with generated captions can be aligned with text prompts for classification and retrieval.
    This is the core modeling assumption of DesignCLIP from the abstract. The evidence for it is not in the supplied full text, and it cannot be checked.
invented entities (1)
  • CV AESM style modulation no independent evidence
    purpose: In the TPA paper, this module injects a learned style vector into video embeddings to estimate uncertainty.
    This is a module from the TPA text. It is a new component, but its evidence is only in the TPA experiments, and those experiments are for fetal ultrasound videos, not for patent drawings.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of DesignCLIP: Multimodal Learning with CLIP for Design Patent Understanding." pith.science (2026). https://pith.science/paper/3LHPAGQ5

@misc{pith2026250815297,
  author       = {Pith},
  title        = {Pith review of: DesignCLIP: Multimodal Learning with CLIP for Design Patent Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3LHPAGQ5}},
  note         = {Machine review of arXiv:2508.15297}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In the field of design patent analysis, traditional tasks such as patent classification and patent image retrieval heavily depend on the image data. However, patent images -- typically consisting of sketches with abstract and structural elements of an invention -- often fall short in conveying comprehensive visual context and semantic information. This inadequacy can lead to ambiguities in evaluation during prior art searches. Recent advancements in vision-language models, such as CLIP, offer promising opportunities for more reliable and accurate AI-driven patent analysis. In this work, we leverage CLIP models to develop a unified framework DesignCLIP for design patent applications with a large-scale dataset of U.S. design patents. To address the unique characteristics of patent data, DesignCLIP incorporates class-aware classification and contrastive learning, utilizing generated detailed captions for patent images and multi-views image learning. We validate the effectiveness of DesignCLIP across various downstream tasks, including patent classification and patent retrieval. Additionally, we explore multimodal patent retrieval, which provides the potential to enhance creativity and innovation in design by offering more diverse sources of inspiration. Our experiments show that DesignCLIP consistently outperforms baseline and SOTA models in the patent domain on all tasks. Our findings underscore the promise of multimodal approaches in advancing patent analysis. The codebase is available here: https://anonymous.4open.science/r/PATENTCLIP-4661/README.md.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

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

  1. Contribution of Globular Clusters to Diffuse Gamma-ray Emission from Galactic Plane

    astro-ph.HE 2025-08 unverdicted novelty 5.0

    Unresolved globular clusters contribute only ~2% (TeV) and <1% (GeV) of the diffuse Galactic gamma-ray emission, effectively negligible.

Reference graph

Works this paper leans on

50 extracted references · 35 canonical work pages · cited by 1 Pith paper · 10 internal anchors

  1. [1]

    Edue: Expert disagreement-guided one-pass un- certainty estimation for medical image segmentation, 2024

    Kudaibergen Abutalip, Numan Saeed, Ikboljon Sobirov, Vincent Andrearczyk, Adrien Depeursinge, and Mohammad Yaqub. Edue: Expert disagreement-guided one-pass un- certainty estimation for medical image segmentation, 2024. URL https://arxiv. org/abs/2403.16594

    Pith/arXiv arXiv 2024

  2. [2]

    A category attention in- stance segmentation network for four cardiac chambers segmentation in fetal echocar- diography

    Shan An, Haogang Zhu, Yuanshuai Wang, Fangru Zhou, Xiaoxue Zhou, Xu Yang, Yingying Zhang, Xiangyu Liu, Zhicheng Jiao, and Yihua He. A category attention in- stance segmentation network for four cardiac chambers segmentation in fetal echocar- diography. Computerized Medical Imaging and Graphics , 93:101983, October 2021. doi: 10.1016/j.compmedimag.2021.101...

    work page arXiv 2021

  3. [3]

    Levine, Ellen Chinn, and Anita J

    Rima Arnaout, Lindsey Curran, Yao Zhao, Jennifer C. Levine, Ellen Chinn, and Anita J. Moon-Grady. An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease. Nature Medicine, 27(5):882–891, May

    work page

  4. [4]

    xlstm: Extended long short-term memory, 2024

    Maximilian Beck, Korbinian Pöppel, Markus Spanring, Andreas Auer, Oleksandra Prudnikova, Michael Kopp, Günter Klambauer, Johannes Brandstetter, and Sepp Hochreiter. xlstm: Extended long short-term memory, 2024. URLhttps://arxiv. org/abs/2405.04517

    Pith/arXiv arXiv 2024

  5. [5]

    ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts

    Mu Cai, Haotian Liu, Dennis Park, Siva Karthik Mustikovela, Gregory P. Meyer, Yun- ing Chai, and Yong Jae Lee. Vip-llava: Making large multimodal models understand arbitrary visual prompts, 2024. URL https://arxiv.org/abs/2312.00784

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    J. S. Carvalho, E. Mavrides, E. A. Shinebourne, S. Campbell, and B. Thilaganathan. Improving the effectiveness of routine prenatal screening for major congenital heart defects. Heart, 88(4):387–391, October 2002. doi: 10.1136/heart.88.4.387. TARA TYNOV A ET. AL: TPA 11

    work page doi:10.1136/heart.88.4.387 2002

  7. [7]

    Rojas Chaves and Subarna Tripathi

    Jose M. Rojas Chaves and Subarna Tripathi. Videosage: Video summarization with graph representation learning, 2024. URL https://arxiv.org/abs/2404. 10539

    work page 2024

  8. [8]

    Vision– language foundation model for echocardiogram interpretation

    Mads Christensen, Milan Vukadinovic, Ning Yuan, and David Ouyang. Vision– language foundation model for echocardiogram interpretation. Nature Medicine, 30 (5):1481–1488, 2024. doi: 10.1038/s41591-024-02959-y

    doi:10.1038/s41591-024-02959-y 2024

  9. [9]

    Improving Zero-shot Generalization and Robustness of Multi-modal Models

    Yunhao Ge, Jie Ren, Andrew Gallagher, Yuxiao Wang, Ming-Hsuan Yang, Hartwig Adam, Laurent Itti, Balaji Lakshminarayanan, and Jiaping Zhao. Improving zero-shot generalization and robustness of multi-modal models, 2023. URLhttps://arxiv. org/abs/2212.01758

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  10. [10]

    Increasing Textual Context Size Boosts Medical Image-Text Matching

    Idan Glassberg and Tom Hope. Increasing textual context size boosts medical image- text matching, 2023. URL https://arxiv.org/abs/2303.13340

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  11. [11]

    Framewise phoneme classification with bidirec- tional lstm and other neural network architectures

    Alex Graves and Jürgen Schmidhuber. Framewise phoneme classification with bidirec- tional lstm and other neural network architectures. Neural Networks, 18(5-6):602–610,

    work page

  12. [12]

    On calibration of modern neural networks

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q Weinberger. On calibration of modern neural networks. In International conference on machine learning, pages 1321–1330. PMLR, 2017

    2017

  13. [13]

    Mmsummary: Multimodal summary generation for fetal ultrasound video

    Xiaoqing Guo, Qianhui Men, and J Alison Noble. Mmsummary: Multimodal summary generation for fetal ultrasound video. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 678–688. Springer, 2024

    work page 2024

  14. [14]

    Long short-term memory

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Compu- tation, 9(8):1735–1780, 11 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL https://doi.org/10.1162/neco.1997.9.8.1735

    doi:10.1162/neco.1997.9.8.1735 1997

  15. [15]

    Severe aortic stenosis detection by deep learning applied to echocardiography

    Gregory Holste, Evangelos K Oikonomou, Bobak J Mortazavi, Andreas Coppi, Kamil F Faridi, Edward J Miller, John K Forrest, Robert L McNamara, Lucila Ohno- Machado, Neal Yuan, Aakriti Gupta, David Ouyang, Harlan M Krumholz, Zhangyang Wang, and Rohan Khera. Severe aortic stenosis detection by deep learning applied to echocardiography. European Heart Journal,...

    doi:10.1093/eurheartj/ehad456 2023

  16. [16]

    Oikonomou, Márton Tokodi, Attila Kovács, Zhangyang Wang, and Rohan Khera

    Gregory Holste, Evangelos K. Oikonomou, Márton Tokodi, Attila Kovács, Zhangyang Wang, and Rohan Khera. Panecho: Complete ai-enabled echocardiography inter- pretation with multi-task deep learning. medRxiv, 2025. doi: 10.1101/2024.11.16. 24317431. URL https://www.medrxiv.org/content/early/2025/04/ 16/2024.11.16.24317431

    work page doi:10.1101/2024.11.16 2025

  17. [17]

    Efficient Uncertainty Estimation for Semantic Segmentation in Videos

    Po-Yu Huang, Wan-Ting Hsu, Chun-Yueh Chiu, Ting-Fan Wu, and Min Sun. Efficient uncertainty estimation for semantic segmentation in videos, 2018. URL https:// arxiv.org/abs/1807.11037. 12 TARA TYNOV A ET. AL: TPA

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  18. [18]

    Perceiver: General perception with iterative attention, 2021

    Andrew Jaegle, Felix Gimeno, Andrew Brock, Andrew Zisserman, Oriol Vinyals, and Joao Carreira. Perceiver: General perception with iterative attention, 2021. URL https://arxiv.org/abs/2103.03206

    Pith/arXiv arXiv 2021

  19. [19]

    Khalil and K

    A. Khalil and K. H. Nicolaides. Fetal heart defects: potential and pitfalls of first- trimester detection. Seminars in Fetal & Neonatal Medicine, 18(5):251–260, October

    work page

  20. [20]

    Maple: Multi-modal prompt learning, 2023

    Muhammad Uzair Khattak, Hanoona Rasheed, Muhammad Maaz, Salman Khan, and Fahad Shahbaz Khan. Maple: Multi-modal prompt learning, 2023. URL https: //arxiv.org/abs/2210.03117

    Pith/arXiv arXiv 2023

  21. [21]

    Dudes: Deep uncertainty distillation using ensembles for semantic segmentation

    Steven Landgraf, Kira Wursthorn, Markus Hillemann, and Markus Ulrich. Dudes: Deep uncertainty distillation using ensembles for semantic segmentation. PFG – Journal of Photogrammetry, Remote Sensing and Geoinformation Science , 92(2): 101–114, March 2024. ISSN 2512-2819. doi: 10.1007/s41064-024-00280-4. URL http://dx.doi.org/10.1007/s41064-024-00280-4

    work page doi:10.1007/s41064-024-00280-4 2024

  22. [22]

    A Critical Synthesis of Uncertainty Quantification and Foundation Models in Monocular Depth Estimation

    Steven Landgraf, Rongjun Qin, and Markus Ulrich. A critical synthesis of uncer- tainty quantification and foundation models in monocular depth estimation, 2025. URL https://arxiv.org/abs/2501.08188

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    Colin Lea, Rene Vidal, Austin Reiter, and Gregory D. Hager. Temporal convolutional networks: A unified approach to action segmentation, 2016

    2016

  24. [24]

    Uncertainty Modeling in Ultrasound Image Segmentation for Precise Fetal Biometric Measurements

    Shuge Lei. Uncertainty modeling in ultrasound image segmentation for precise fetal biometric measurements, 2024. URL https://arxiv.org/abs/2401.09639

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    Black, Mun-kit Choy, Ningxiu Li, and Bernard D

    Yingjuan Liu, Sen Chen, Liesl Zühlke, Graeme C. Black, Mun-kit Choy, Ningxiu Li, and Bernard D. Keavney. Global birth prevalence of congenital heart defects 1970–2017: updated systematic review and meta-analysis of 260 studies.International Journal of Epidemiology, 48(2):455–463, 2019. doi: 10.1093/ije/dyz009

    work page doi:10.1093/ije/dyz009 1970

  26. [26]

    Task-Oriented Multi-Modal Mutual Leaning for Vision-Language Models

    Sifan Long, Zhen Zhao, Junkun Yuan, Zichang Tan, Jiangjiang Liu, Luping Zhou, Shengsheng Wang, and Jingdong Wang. Task-oriented multi-modal mutual leaning for vision-language models, 2023. URL https://arxiv.org/abs/2303.17169

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  27. [27]

    A yolox-based deep instance segmentation neural network for cardiac anatomical structures in fetal ultrasound im- ages

    Yuhuan Lu, Kenli Li, Bin Pu, Ying Tan, and Ningbo Zhu. A yolox-based deep instance segmentation neural network for cardiac anatomical structures in fetal ultrasound im- ages. IEEE/ACM Transactions on Computational Biology and Bioinformatics , 21(4): 1007–1018, Jul-Aug 2024. doi: 10.1109/TCBB.2022.3222356. Epub 2024 Aug 8; PMID: 36378800

    arXiv 2024

  28. [28]

    Fetalclip: A visual-language foundation model for fetal ultrasound image analysis — arxiv.org

    Saeed Maani, Farooq Saleem, Diehl Alasmawi, Waring Mohammed, and Bricker Valappi. Fetalclip: A visual-language foundation model for fetal ultrasound image analysis — arxiv.org. https://arxiv.org/html/2502.14807v1, 2025

    arXiv 2025

  29. [29]

    Congenital heart disease: types, pathophysiology, diagnosis, and treat- ment options

    Xiangyu Meng, Mengdi Song, Kai Zhang, Wenhao Lu, Yujia Li, Cheng Zhang, and Yuhan Zhang. Congenital heart disease: types, pathophysiology, diagnosis, and treat- ment options. MedComm (2020), 5(7):e631, Jul 2024. doi: 10.1002/mco2.631. TARA TYNOV A ET. AL: TPA 13

    work page doi:10.1002/mco2.631 2020

  30. [30]

    Measuring calibration in deep learning

    Jeremy Nixon, Michael W Dusenberry, Linchuan Zhang, Ghassen Jerfel, and Dustin Tran. Measuring calibration in deep learning. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019

    work page 2019

  31. [31]

    Lungren, William S

    Daniel Ouyang, Bowen He, Amirata Ghorbani, Matthew P. Lungren, William S. Colucci, Dinggang Shen, Euan A. Ashley, and David H. Liang. Video-based ai for beat-to-beat assessment of cardiac function. Nature, 580:252–256, March

    work page

  32. [32]

    Arijit Patra and Julia A. Noble. Hierarchical class incremental learning of anatomical structures in fetal echocardiography videos. IEEE Journal of Biomedical and Health Informatics, 24(4):1046–1058, April 2020. doi: 10.1109/JBHI.2020.2973372

    work page arXiv 2020

  33. [33]

    Perperidis, D

    A. Perperidis, D. Cusack, A. White, N. McDicken, T. MacGillivray, and T. Anderson. Dynamic enhancement of b-mode cardiac ultrasound image sequences. Ultrasound in Medicine & Biology , 43(7):1533–1548, jul 2017. doi: 10.1016/j.ultrasmedbio.2017. 03.006

    work page doi:10.1016/j.ultrasmedbio.2017 2017

  34. [34]

    Fetal cardiac cycle detection in multi- resource echocardiograms using hybrid classification framework

    Bin Pu, Ningbo Zhu, Kenli Li, and Shengli Li. Fetal cardiac cycle detection in multi- resource echocardiograms using hybrid classification framework. Future Generation Computer Systems, 115:825–836, 02 2021. doi: 10.1016/j.future.2020.09.014

    work page doi:10.1016/j.future.2020.09.014 2021

  35. [35]

    Flds: An intelligent feature learning detection system for visualizing medical images supporting fetal four-chamber views

    Sibo Qiao, Shanchen Pang, Gang Luo, Silin Pan, Taotao Chen, and Zhihan Lv. Flds: An intelligent feature learning detection system for visualizing medical images supporting fetal four-chamber views. IEEE Journal of Biomedical and Health Informatics , 26 (10):4814–4825, October 2022. doi: 10.1109/JBHI.2021.3091579. Epub 2022 Oct 4; PMID: 34156957

    work page arXiv 2022

  36. [36]

    U-net: Convolutional net- works for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional net- works for biomedical image segmentation. InMedical image computing and computer- assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, pages 234–241. Springer, 2015

    work page 2015

  37. [37]

    Self-supervised Normality Learning and Divergence Vector-guided Model Merging for Zero-shot Congenital Heart Disease Detection in Fetal Ultrasound Videos

    Pramit Saha, Divyanshu Mishra, Netzahualcoyotl Hernandez-Cruz, Olga Patey, Aris Papageorghiou, Yuki M. Asano, and J. Alison Noble. Self-supervised normality learn- ing and divergence vector-guided model merging for zero-shot congenital heart dis- ease detection in fetal ultrasound videos, 2025. URL https://arxiv.org/abs/ 2503.07799

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    Papa- georghiou, and J

    Harshita Sharma, Lior Drukker, Pierre Chatelain, Richard Droste, Aris T. Papa- georghiou, and J. Alison Noble. Knowledge representation and learning of opera- tor clinical workflow from full-length routine fetal ultrasound scan videos. Medical Image Analysis , 69:101973, 2021. ISSN 1361-8415. doi: https://doi.org/10.1016/ j.media.2021.101973. URL https://...

    work page arXiv 2021

  39. [39]

    CV AE-SM: A Conditional Variational Autoen- coder with Style Modulation for Efficient Uncertainty Quantification

    Amin Ullah, Taiqing Yan, and Li Fuxin. CV AE-SM: A Conditional Variational Autoen- coder with Style Modulation for Efficient Uncertainty Quantification. In Proceedings of the 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 10786–10792, 2024. doi: 10.1109/ICRA57147.2024.10611160. 14 TARA TYNOV A ET. AL: TPA

    arXiv 2024

  40. [40]

    ProbVLM: Probabilistic Adapter for Frozen Vision-Language Models

    Uddeshya Upadhyay, Shyamgopal Karthik, Massimiliano Mancini, and Zeynep Akata. Probvlm: Probabilistic adapter for frozen vision-language models, 2023. URL https://arxiv.org/abs/2307.00398

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  41. [41]

    van der Linde, E

    D. van der Linde, E. E. E. Konings, M. A. Slager, M. Witsenburg, W. A. Helbing, J. J. M. Takkenberg, and J. W. Roos-Hesselink. Birth prevalence of congenital heart disease worldwide: a systematic review and meta-analysis. Journal of the American College of Cardiology, 58(21):2241–2247, November 2011. doi: 10.1016/j.jacc.2011. 08.025. PMID: 22078432

    work page doi:10.1016/j.jacc.2011 2011

  42. [42]

    Ogunbona

    Zhen Wang, Guangxu Li, Jingjie Zhou, and Philip O. Ogunbona. Optical flow net- works for heartbeat estimation in 4d ultrasound images. InProceedings of the 2021 7th International Conference on Computing and Artificial Intelligence , ICCAI ’21, page 127–131, New York, NY , USA, 2021. Association for Computing Machinery. ISBN 9781450389501. doi: 10.1145/346...

    work page arXiv 2021

  43. [43]

    Semantic Alignment for Multimodal Large Language Models

    Tao Wu, Mengze Li, Jingyuan Chen, Wei Ji, Wang Lin, Jinyang Gao, Kun Kuang, Zhou Zhao, and Fei Wu. Semantic alignment for multimodal large language models, 2024. URL https://arxiv.org/abs/2408.12867

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    Segmentation of ten fetal heart components with coarse-to-fine cascading and dynamic feature powering

    Tingyang Yang, Ye Zhang, Mengxiao Zhu, Yan Wang, Shan An, Xiaoyan Gu, Xiaowei Liu, Jiancheng Han, Yihua He, and Haogang Zhu. Segmentation of ten fetal heart components with coarse-to-fine cascading and dynamic feature powering. IET Image Processing, 16:n/a–n/a, 07 2022. doi: 10.1049/ipr2.12597

    work page doi:10.1049/ipr2.12597 2022

  45. [45]

    Sunhee W. Yun. Congenital heart disease in the newborn requiring early intervention. Korean Journal of Pediatrics, 54(5):183–191, May 2011. doi: 10.3345/kjp.2011.54.5

    work page doi:10.3345/kjp.2011.54.5 2011

  46. [46]

    Moderately/Severely Decreased

    Michael Zhang and Christopher Ré. Contrastive adapters for foundation model group robustness, 2022. URL https://arxiv.org/abs/2207.07180. TARA TYNOV A ET. AL: TPA 15 Supplementary Material In the Supplementary Material, we first present in Section A detailed descriptions of private CHD dataset and of the EchoNet-Dynamic dataset, including how we converted...

    Pith/arXiv arXiv 2022

  47. [2005]

    doi: 10.1016/j.neunet.2005.06.042

    doi:10.1016/j.neunet.2005.06.042 2005

  48. [2013]

    Epub 2013 Jun 7; PMID: 23751926

    doi: 10.1016/j.siny.2013.05.004. Epub 2013 Jun 7; PMID: 23751926

    work page doi:10.1016/j.siny.2013.05.004 2013

  49. [2020]

    URL https://doi.org/10.1038/ s41586-020-2145-8

    doi: 10.1038/s41586-020-2145-8. URL https://doi.org/10.1038/ s41586-020-2145-8 . Received: 11 November 2019; Accepted: 20 February 2020; Published: 25 March 2020; Issue Date: 09 April 2020

    doi:10.1038/s41586-020-2145-8 2019

  50. [2021]

    URL https://doi.org/10.1038/ s41591-021-01342-5

    doi: 10.1038/s41591-021-01342-5. URL https://doi.org/10.1038/ s41591-021-01342-5

    work page doi:10.1038/s41591-021-01342-5

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.