Pith. sign in

REVIEW 3 major objections 2 minor 37 references

Disorder-induced stress-flow misalignment in soft glassy materials revealed using multi-directional shear

T0 review · 3 major / 2 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper presents CT-GRAPH, a hierarchical graph attention model that pools fine-grained organ features from a frozen 3D encoder and reports an absolute 7.9% F1 improvement over prior CT report generators on CT-RATE.

desk verdict Abstract promises a solid soft-matter result, but the supplied full text is an unrelated CT-report paper, so the claims are unverifiable in this package. read the letter →

arxiv 2508.05379 v1 pith:2HBIARCY submitted 2025-08-07 cond-mat.soft cond-mat.mtrl-sci

Frédéric Blanc , Guillaume Ovarlez , Adam Trigui , Kirsten Martens , Romain Mari This is my paper
classification cond-mat.softcond-mat.mtrl-sci
keywords hierarchicalgraphattentionCTreportgenerationanatomicalmaskspretrained3DencodersclinicalentityF1chestradiologyautomationCT-RATE
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 tries to establish that radiology reports for 3D chest CT scans are generated more accurately when a model reasons about anatomy at multiple scales instead of looking at the whole volume at once. Its proposed pipeline, CT-GRAPH, pools features from a pretrained 3D encoder inside fine anatomical masks (individual lung lobes, heart chambers), assembles those organ features into a graph with coarse anatomical systems and a global patient node, and lets a graph attention network pass information up the hierarchy before a language model writes the report. On the CT-RATE dataset the method reports the best clinical-entity recall and F1 (0.296) among the compared systems, an absolute 7.9-point F1 gain over the previous state of the art, and gains are largest for localized pathologies such as consolidation and pericardial effusion. The paper matters because it isolates where the improvement comes from: fine region-level features and anatomical topology, not bigger language models.

What carries the argument

The central object is a three-level anatomical graph built from segmentation masks. Fine-level nodes are 34 anatomical structures, each represented by features pooled from the masks across several layers of a frozen 3D encoder; coarse-level nodes are unions such as 'lungs' or 'abdomen'; one global node represents the whole scan. Edges are fixed by the anatomy, and a graph attention network (GAT) aggregates fine features into coarse nodes and coarse nodes into the global node. The resulting node embeddings are projected and concatenated with a text prompt before being fed to LLaMA2-7B, trained with LoRA. The mechanism doing the work is the bottom-up hierarchical message passing, which lets lo

What would settle it

Replace the TotalSegmentator masks with random masks of the same volume while holding all else fixed: if the CT-GRAPH F1 advantage over the flat and global baselines persists nearly unchanged, the anatomical topology is not the causal driver. A second check is to re-evaluate on reports labeled by radiologists instead of the CT-CLIP extractor to see whether the 7.9-point gain survives independent ground truth.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the representation of the CT volume—how features are grouped and related—is the main driver of report quality. The authors find that global average pooling of the encoder feature map is near-useless (near-zero recall and F1 in pathology classification), that fine-grained mask-pooled features consistently beat coarse organ and global features, and that arranging these features in a fixed anatomical hierarchy with graph attention improves clinical-entity F1 to 0.296, above the region-based Reg2RG (0.217), CT2Rep with LLaMA (0.214), and its own unstructured baselines. A random-graph ablation performs worse, which the paper takes as evidenc

Load-bearing premise

The load-bearing premise is that the anatomical structures delineated by TotalSegmentator align with the regions where clinically relevant findings live, so that pooling encoder features inside those masks captures the information that matters for the report; if the masks or the automated label extraction miss the true pathology locations, the reported F1 gains reflect those alignments rather than the model's anatomical reasoning.

Editorial extensions

If this is right

  • Fine-grained mask-pooled features (e.g., individual lung lobes) are consistently better than whole-organ or global features for both pathology classification and report generation.
  • A fixed anatomical hierarchy beats flat combinations of local and global features, and random graph connectivity removes most of the gain.
  • The choice of pretrained 3D encoder changes clinical-entity accuracy much more than it changes surface language metrics; VoCo-160k gives the best overall results.
  • The two-stage frozen-feature pipeline is practical: with pre-extracted features, training completes in roughly 10 hours on a single A100, lowering the barrier for 3D report generation.
  • CT-GRAPH achieves the highest F1 on 15 of 18 pathology classes, with the largest gaps on spatially localized findings like consolidation, pericardial effusion, and arterial wall calcification.

Reading between the lines

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

  • A direct control the paper does not run would replace the anatomical masks with random masks of equal size; without it, how much of the gain comes from mask granularity versus graph topology is not fully separated.
  • Because the clinical-entity labels come from an automated extractor (CT-CLIP), the reported F1 gains may partly reflect alignment with that extractor's output rather than with radiologist judgment; an expert-annotation study would clarify this.
  • The near-zero performance of global pooling suggests a general trade-off for 3D foundation models: aggressive spatial compression destroys localized evidence, so anatomy-aware pooling may benefit dense 3D tasks beyond report generation.
  • Pretraining scale alone did not explain encoder performance (VoCo-10k and VoCo-160k are nearly tied), implying feature semantics or mask alignment matters more; varying mask granularity systematically would be a natural next experiment.
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

3 major / 2 minor

Summary. The submission as provided does not match the abstract under review. The abstract (arXiv:2508.05379, cond-mat.soft) claims an experimental and modeling study of soft glassy materials, reporting a transient shear response orthogonal to the applied shear direction and an anisotropic yield surface, rationalized by a mesoscopic elasto-plastic model with local mechanical disorder. The full text, however, is a completely different paper: 'CT-GRAPH: Hierarchical Graph Attention Network for Anatomy-Guided CT Report Generation' (arXiv:2508.05375v1, cs.CV). The full text contains no multi-axis shear apparatus, no rheological measurements, no yield surface characterization, no elasto-plastic model equations, and no disorder-parameter definition. The central claim of the abstract cannot be evaluated because none of its supporting content is present in the submitted manuscript.

Significance. If the claimed soft-matter result were properly documented, it would be a significant contribution: the observation of an orthogonal stress transient and an anisotropic yield surface after shear history would advance understanding of flow-history effects in soft glasses and could provide a new way to probe local yield stress distributions. The CT-GRAPH paper, considered on its own, appears to be a solid applied ML contribution with reproducible code and competitive results (e.g., absolute 7.9% F1 improvement over prior state of the art). However, the submitted full text contains none of the claimed soft-matter work, so the significance of the abstract's assertions cannot be assessed in this manuscript.

major comments (3)
  1. [Full text (all sections)] The manuscript body is an entirely different paper: CT-GRAPH (arXiv:2508.05375v1), on chest CT report generation. The abstract under review describes a multi-axis shear experiment on soft glassy materials and a mesoscopic elasto-plastic model. No section of the submitted full text addresses rheology, shear apparatus, stress measurements, yield surfaces, or elasto-plastic modeling. The claimed central result is therefore absent from the submitted materials.
  2. [§5, Tables 1–7] The only experimental results in the full text are pathology-classification and report-generation metrics for the CT-GRAPH model (e.g., Table 6: CE F1 = 0.296; Table 7: per-pathology F1 scores). These data cannot support—and are not cited in support of—the abstract's claims about transient orthogonal shear response or anisotropic yield surfaces. The abstract's assertion that 'local mechanical disorder governs the emergence of macroscopic stress-flow misalignment' has no accompanying model equations or parameter definitions anywhere in the manuscript.
  3. [Abstract / full-text consistency] The abstract promises a custom multi-axis shear apparatus and a mesoscopic elasto-plastic model, but the full text contains no apparatus description, no model formulation (no yield stress distribution, no disorder parameter), and no comparison between model and experiment. Without these elements, the reader cannot test the load-bearing assumption that the orthogonal response is a bulk material property rather than an apparatus artifact, nor whether the disorder distribution is measured or fitted. This is a missing-support issue of the first order and must be resolved before any scientific assessment.
minor comments (2)
  1. [Metadata] The full-text header states arXiv:2508.05375v1 [cs.CV], while the abstract under review corresponds to arXiv:2508.05379 (cond-mat.soft). This arXiv ID mismatch should be resolved; if the soft-matter paper exists, its correct full text must be supplied.
  2. [Administrative] If the CT-GRAPH submission is an upload error, the authors should be notified so that the correct manuscript can be submitted for review.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity demonstrable: the supplied full text is an unrelated CT report-generation paper, so the claimed soft-matter derivation chain is absent from the record.

full rationale

The claimed manuscript (arXiv:2508.05379) presents an abstract about multi-directional shear of soft glasses, but the supplied full text is CT-GRAPH, a computer-vision paper on CT report generation. There is therefore no accessible derivation chain: no model equations, no disorder-parameter definition, no fitting procedure, no experimental-misalignment calculation. The abstract alone asserts that a mesoscopic elasto-plastic model 'demonstrates' that local mechanical disorder governs the emergence of stress-flow misalignment, but it does not exhibit any equation or reduction that would let one check whether the prediction is equivalent to the model input by construction. Under the hard rule that circularity may only be claimed when the paper can be quoted and the specific reduction exhibited, no circular step can be identified. The mismatch between the claimed paper and the supplied text is a serious completeness/support issue, not a circularity, and should be resolved before any scientific assessment.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

Only the abstract was available, so the ledger is inferred. The elasto-plastic model likely contains free parameters governing the local yield stress distribution; the paper's treatment of these parameters as fitted or derived is unknown.

free parameters (1)
  • local yield stress distribution parameters
    The mesoscopic elasto-plastic model relies on a distribution of local yield stresses to produce macroscopic misalignment; whether these are fitted to the experimental data or derived is not stated in the abstract.
assumptions (2)
  • domain assumption Mesoscopic elasto-plastic models capture the essential physics of soft glassy flow.
    The abstract uses such a model to rationalize experimental observations; this assumes the model class is appropriate for the system.
  • domain assumption The custom multi-axis shear apparatus produces homogeneous simple shear and reliably measures orthogonal stress components.
    The abstract reports measurements from a custom apparatus; any interpretation assumes the apparatus does not introduce artifacts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Disorder-induced stress-flow misalignment in soft glassy materials revealed using multi-directional shear." pith.science (2026). https://pith.science/paper/2HBIARCY

@misc{pith2026250805379,
  author       = {Pith},
  title        = {Pith review of: Disorder-induced stress-flow misalignment in soft glassy materials revealed using multi-directional shear},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HBIARCY}},
  note         = {Machine review of arXiv:2508.05379}
}
read the original abstract

Controlling the mechanical response of soft glassy materials, such as emulsions, foams, and colloidal suspensions, is key for many industrial processes. While their steady-state flow behavior is reasonably well understood, their response to complex flow histories, as encountered in operations like pumping or mixing, remains poorly known. Using a custom multi-axis shear apparatus that enables arbitrary changes in flow direction, we investigate how shear history influences the mechanical behavior of a model soft glassy system. We uncover a transient shear response orthogonal to the applied shear direction, together with an anisotropic yield surface. These effects point to an underlying anisotropic distribution of internal stresses imprinted by previous deformation. To rationalize this behavior, we use a mesoscopic elasto-plastic model, demonstrating that local mechanical disorder governs the emergence of macroscopic stress-flow misalignment. Our findings offer a new route to experimentally probe the distribution of local yield stresses in soft glassy materials.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 19 canonical work pages

  1. [1]

    Layer normalization

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hin- ton. Layer normalization. arXiv preprint arXiv:1607.06450,

    arXiv

  2. [2]

    M3d: Advancing 3d medical image analysis with multi-modal large language models

    Fan Bai, Yuxin Du, Tiejun Huang, Max Q-H Meng, and Bo Zhao. M3d: Advancing 3d medical image analysis with multi-modal large language models. arXiv preprint arXiv:2404.00578, 2024. 1, 2

    arXiv 2024

  3. [3]

    Meteor: An automatic metric for mt evaluation with improved correlation with hu- man judgments

    Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with hu- man judgments. In Proceedings of the acl workshop on in- trinsic and extrinsic evaluation measures for machine trans- lation and/or summarization, pages 65–72, 2005. 5

    work page 2005

  4. [4]

    Dia-LLaMA: Towards Large Language Model-driven CT Report Generation

    Zhixuan Chen, Luyang Luo, Yequan Bie, and Hao Chen. Dia-llama: Towards large language model-driven ct report generation. arXiv preprint arXiv:2403.16386, 2024. 1, 2

    work page Pith review arXiv 2024

  5. [5]

    Large language model with region-guided referring and ground- ing for ct report generation

    Zhixuan Chen, Yequan Bie, Haibo Jin, and Hao Chen. Large language model with region-guided referring and ground- ing for ct report generation. IEEE Transactions on Medical Imaging, 2025. 1, 2, 7, 8

    work page 2025

  6. [6]

    Ct-agrg: Automated abnormality-guided report generation from 3d chest ct volumes

    Theo Di Piazza. Ct-agrg: Automated abnormality-guided report generation from 3d chest ct volumes. arXiv preprint arXiv:2408.11965, 2024. 2

    arXiv 2024

  7. [7]

    Trialing u-net train- ing modifications for segmenting gliomas using open source deep learning framework

    David G Ellis and Michele R Aizenberg. Trialing u-net train- ing modifications for segmenting gliomas using open source deep learning framework. In International MICCAI Brainle- sion Workshop, pages 40–49. Springer, 2020. 5

    work page 2020

  8. [8]

    Evidence-based guideline for the written radiology report: Methods, recommendations and implementation challenges

    Stacy K Goergen, Felicity J Pool, Tari J Turner, Jane E Grimm, Mark N Appleyard, Carmel Crock, Michael C Fa- hey, Michael F Fay, Nicholas J Ferris, Susan M Liew, et al. Evidence-based guideline for the written radiology report: Methods, recommendations and implementation challenges. Journal of medical imaging and radiation oncology , 57(1): 1–7, 2013. 1

    work page 2013

Show all 37 references
  1. [9]

    vox2vec: a framework for self-supervised contrastive learning of voxel-level repre- sentations in medical images

    Mikhail Goncharov, Vera Soboleva, Anvar Kurmukov, Maxim Pisov, and Mikhail Belyaev. vox2vec: a framework for self-supervised contrastive learning of voxel-level repre- sentations in medical images. In International Conference on Medical Image Computing and Computer-Assisted In...

    2023

  2. [10]

    Complex organ mask guided radiology report generation

    Tiancheng Gu, Dongnan Liu, Zhiyuan Li, and Weidong Cai. Complex organ mask guided radiology report generation. In Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision, pages 7995–8004, 2024. 2

    2024

  3. [11]

    Orid: Organ-regional in- formation driven framework for radiology report generation

    Tiancheng Gu, Kaicheng Yang, Xiang An, Ziyong Feng, Dongnan Liu, and Weidong Cai. Orid: Organ-regional in- formation driven framework for radiology report generation. arXiv preprint arXiv:2411.13025, 2024. 1, 2

    2024 arXiv

  4. [12]

    Transferable visual words: Exploiting the semantics of anatomical patterns for self-supervised learning.IEEE trans- actions on medical imaging, 40(10):2857–2868, 2021

    Fatemeh Haghighi, Mohammad Reza Hosseinzadeh Taher, Zongwei Zhou, Michael B Gotway, and Jianming Liang. Transferable visual words: Exploiting the semantics of anatomical patterns for self-supervised learning.IEEE trans- actions on medical imaging, 40(10):2857–2868, 2021. 2, 5, 6, 7

    2021

  5. [13]

    Developing general- ist foundation models from a multimodal dataset for 3d com- puted tomography

    Ibrahim Ethem Hamamci, Sezgin Er, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esirgun, Irem Dogan, Muhammed Furkan Dasdelen, Omer Faruk Durugol, Bastian Wittmann, Tamaz Amiranashvili, et al. Developing general- ist foundation models from a multimodal dataset for 3d com- put...

    2024

  6. [14]

    Ct2rep: Automated radiology report generation for 3d medi- cal imaging

    Ibrahim Ethem Hamamci, Sezgin Er, and Bjoern Menze. Ct2rep: Automated radiology report generation for 3d medi- cal imaging. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 476–

  7. [15]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models. ICLR, 1(2):3, 2022. 5

    2022

  8. [16]

    Kiut: Knowledge-injected u-transformer for radiology re- port generation

    Zhongzhen Huang, Xiaofan Zhang, and Shaoting Zhang. Kiut: Knowledge-injected u-transformer for radiology re- port generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 19809– 19818, 2023. 1

    2023

  9. [17]

    Read like a radiologist: Efficient vision-language model for 3d medical imaging interpretation

    Changsun Lee, Sangjoon Park, Cheong-Il Shin, Woo Hee Choi, Hyun Jeong Park, Jeong Eun Lee, and Jong Chul Ye. Read like a radiologist: Efficient vision-language model for 3d medical imaging interpretation. arXiv preprint arXiv:2412.13558, 2024. 1

    2024 arXiv

  10. [18]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out , pages 74–81, 2004. 5

    2004

  11. [19]

    Benchmarking and boosting radiology report generation for 3d high-resolution medical images

    Che Liu, Zhongwei Wan, Yuqi Wang, Hui Shen, Haozhe Wang, Kangyu Zheng, Mi Zhang, and Rossella Arcucci. Benchmarking and boosting radiology report generation for 3d high-resolution medical images. arXiv preprint arXiv:2406.07146, 2024. 1, 2

    2024 arXiv

  12. [20]

    Clinically accurate chest x-ray report generation

    Guanxiong Liu, Tzu-Ming Harry Hsu, Matthew McDermott, Willie Boag, Wei-Hung Weng, Peter Szolovits, and Marzyeh Ghassemi. Clinically accurate chest x-ray report generation. In Machine Learning for Healthcare Conference, pages 249–

  13. [21]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021. 5

    2021

  14. [22]

    Vividmed: Vision language model with versatile visual grounding for medicine

    Lingxiao Luo, Bingda Tang, Xuanzhong Chen, Rong Han, and Ting Chen. Vividmed: Vision language model with versatile visual grounding for medicine. arXiv preprint arXiv:2410.12694, 2024. 1, 2

    2024 arXiv

  15. [23]

    3d mri brain tumor segmentation using autoencoder regularization

    Andriy Myronenko. 3d mri brain tumor segmentation using autoencoder regularization. In International MICCAI brain- lesion workshop, pages 311–320. Springer, 2018. 5

    2018

  16. [24]

    Vision foundation models for computed tomography

    Suraj Pai, Ibrahim Hadzic, Dennis Bontempi, Keno Bressem, Benjamin H Kann, Andriy Fedorov, Raymond H Mak, and Hugo JWL Aerts. Vision foundation models for computed tomography. arXiv preprint arXiv:2501.09001, 2025. 5, 6, 7

    2025 arXiv

  17. [25]

    Bleu: a method for automatic evaluation of machine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages 311–318,

  18. [26]

    De- tailed annotations of chest x-rays via ct projection for report understanding

    Constantin Seibold, Simon Reiß, Saquib Sarfraz, Matthias A Fink, Victoria Mayer, Jan Sellner, Moon Sung Kim, Klaus H Maier-Hein, Jens Kleesiek, and Rainer Stiefelhagen. De- tailed annotations of chest x-rays via ct projection for report understanding. arXiv preprint arXiv:2210...

    2022 arXiv

  19. [27]

    Self-supervised pre-training of swin trans- formers for 3d medical image analysis

    Yucheng Tang, Dong Yang, Wenqi Li, Holger R Roth, Bennett Landman, Daguang Xu, Vishwesh Nath, and Ali Hatamizadeh. Self-supervised pre-training of swin trans- formers for 3d medical image analysis. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogn...

    2022

  20. [28]

    Interactive and explainable region-guided radiol- ogy report generation

    Tim Tanida, Philip M ¨uller, Georgios Kaissis, and Daniel Rueckert. Interactive and explainable region-guided radiol- ogy report generation. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 7433–7442, 2023. 1, 2

    2023

  21. [29]

    Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023. 5

    2023 arXiv

  22. [30]

    Graph at- tention networks

    Petar Veli ˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. Graph at- tention networks. arXiv preprint arXiv:1710.10903 , 2017. 4

    2017 arXiv

  23. [31]

    To- talsegmentator: robust segmentation of 104 anatomic struc- tures in ct images

    Jakob Wasserthal, Hanns-Christian Breit, Manfred T Meyer, Maurice Pradella, Daniel Hinck, Alexander W Sauter, Tobias Heye, Daniel T Boll, Joshy Cyriac, Shan Yang, et al. To- talsegmentator: robust segmentation of 104 anatomic struc- tures in ct images. Radiology: Artificial In...

    2023

  24. [32]

    Towards generalist foundation model for radi- ology by leveraging web-scale 2d&3d medical data

    Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, and Weidi Xie. Towards generalist foundation model for radi- ology by leveraging web-scale 2d&3d medical data. arXiv preprint arXiv:2308.02463, 2023. 1, 2

    2023 arXiv

  25. [33]

    Large-scale 3d medical image pre-training with geometric context priors

    Linshan Wu, Jiaxin Zhuang, and Hao Chen. Large-scale 3d medical image pre-training with geometric context priors. arXiv preprint arXiv:2410.09890, 2024. 2, 5

    2024 arXiv

  26. [34]

    V oco: A simple- yet-effective volume contrastive learning framework for 3d medical image analysis

    Linshan Wu, Jiaxin Zhuang, and Hao Chen. V oco: A simple- yet-effective volume contrastive learning framework for 3d medical image analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 22873–22882, 2024. 2, 5, 6, 7

    2024

  27. [35]

    Empirical evaluation of rectified activations in convolutional network

    Bing Xu, Naiyan Wang, Tianqi Chen, and Mu Li. Empirical evaluation of rectified activations in convolutional network. arXiv preprint arXiv:1505.00853, 2015. 4

    2015 arXiv

  28. [36]

    Radgenome-chest ct: A grounded vision-language dataset for chest ct analysis

    Xiaoman Zhang, Chaoyi Wu, Ziheng Zhao, Jiayu Lei, Ya Zhang, Yanfeng Wang, and Weidi Xie. Radgenome-chest ct: A grounded vision-language dataset for chest ct analysis. arXiv preprint arXiv:2404.16754, 2024. 4

    2024 arXiv

  29. [486]

    1, 2, 7, 8

    Springer, 2024. 1, 2, 7, 8

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.