Pith. sign in

REVIEW 5 major objections 4 minor 42 references

Navigating limitations with precision: A fine-grained ensemble approach to wrist pathology recognition on a limited x-ray dataset

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A fine-grained ensemble reaches 87.34% accuracy on a limited wrist X-ray test set, beating conventional and fine-grained baselines.

desk verdict Solid application paper with a real statistical overreach: the ensemble's claimed superiority is a tie on the hard test set and rests on test-set-driven choices. read the letter →

arxiv 2412.13884 v1 pith:HR2DGTPA submitted 2024-12-18 cs.CV

Ammar Ahmed , Ali Shariq Imran , Mohib Ullah , Zenun Kastrati , Sher Muhammad Daudpota This is my paper · ORCID
classification cs.CV
keywords fine-grainedvisualrecognitionwristpathologyclassificationX-rayimaginglimitedmedicaldatasetensemblelearningLIONoptimizerGrad-CAMexplainableAI
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 claims that wrist pathology recognition on a small, class-imbalanced pediatric X-ray dataset is best treated as a fine-grained visual recognition problem, because the differences between classes, such as subtle fractures, are small and localized. Its proposal is an ensemble of three configurations of a plug-in module for fine-grained recognition: the base model, the model with the LION optimizer, and the model with LION plus a 1024-capacity feature pyramid network, combined by majority voting. On its curated four-class dataset of 1,956 training images, the ensemble reaches 87.34% accuracy on a 474-image augmented test set and 83.75% on an 80-image unaugmented challenging test set, above the base plug-in module (84.38% and 82.50%) and above the strongest compared fine-grained baseline, HERBS (82.70% and 78.75%). The point of the exercise is practical: with only image-level labels and limited data, a machine-vision system can locate the discriminative regions, visualized by Grad-CAM heatmaps, without manual bounding-box annotation.

What carries the argument

The load-bearing machinery is the Plug-in Module (PIM), a fine-grained visual recognition component that treats each pixel of a backbone feature map as a separate feature. For each feature map $F_b \in \mathbb{R}^{C\times H\times W}$, a weakly supervised selector runs a linear classifier over pixels and keeps the $k$ highest-probability points, using softmax scores, a descending argsort, and a top-$k$ index set (equations 1-4). The selected points are fused by graph convolution and pooled into super-nodes before a linear classifier predicts the class. The SwinTransformer backbone supplies hierarchical feature maps, a Feature Pyramid Network handles multiple object scales, and a projection or FPN-size knob (chosen at 1024) controls the feature-map dimension feeding the graph network. The paper swaps the default SGD optimizer for LION, a sign-momentum optimizer whose constant-magnitude update adds a regularization-like noise, and finally combines three variants by majority voting.

What would settle it

Compute bootstrap confidence intervals or repeated-seed standard deviations for the ensemble and the base PIM on the 474-image and 80-image test sets; if the intervals overlap, the claimed superiority of the ensemble is not established.

Watch

Extended reading notes

Core claim

The paper's central claim is that a fine-grained ensemble built around the Plug-in Module (PIM) for fine-grained visual recognition outperforms both conventional convolutional networks and recent fine-grained visual recognition architectures on wrist pathology classification. It treats each pixel of the backbone feature map as an independent feature, uses a weakly supervised selector to keep only the pixels with the highest class-confidence scores, fuses the selected points with graph convolution in a feature pyramid, and then combines three model variants by majority voting: PIM with SGD, PIM with LION, and PIM with LION and an FPN size of 1024. On test set 1, this ensemble obtains 87.34% accuracy against 84.38% for the base PIM and 82.70% for the best alternative, HERBS; on test set 2, the original 80-image set, it obtains 83.75% against 82.50% for the base PIM and 78.75% for HERBS. The paper also reports per-class sensitivity, specificity, and precision on test set 2, alongside heatmaps showing that the ensemble highlights more focused discriminative regions than the base model.

Load-bearing premise

The ranking of models rests on accuracy measured on just 474 augmented test images and 80 original test images, so differences of a few percentage points, like the ensemble's edge over the base model, may be sampling noise rather than real improvement.

Editorial extensions

If this is right

  • If the ensemble result holds, wrist pathology classification can be done with image-level labels alone, removing the need for costly bounding-box or region annotations.
  • The gap over conventional CNNs on test set 1, such as 87.34% versus 79.96% for EfficientNet-b0, suggests that fine-grained architectures, not just bigger backbones, matter for subtle X-ray findings.
  • The two cheap modifications, replacing the optimizer and retuning the FPN projection size, are enough to lift the base method, and their combination by majority voting adds further accuracy.
  • The same recipe could be applied to other limited medical imaging datasets where classes differ by small localized regions.

Reading between the lines

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

  • A caution implied by the small test sets: the 1.25-point gap on the 80-image test set and the 2.96-point gap on the augmented set are reported without confidence intervals, so the superiority of the ensemble over the base PIM is not yet statistically established.
  • Because the FPN size of 1024 was chosen after looking at test set 1 performance in Table IV, part of the ensemble's advantage on that set may reflect selection on the test set rather than a general property of the architecture.
  • A natural next test, not run in the paper, would be cross-validation or repeated runs with different seeds to see whether the ensemble's gains are reproducible and whether they persist on an external wrist X-ray dataset.
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

5 major / 4 minor

Summary. The paper proposes a fine-grained visual recognition (FGVR) approach to wrist pathology classification on a limited, custom-curated subset of the GRAZPEDWRI dataset. It builds on the Plug-in Module (PIM) of Chou et al. with a Swin Transformer backbone, a weakly supervised selector, graph-convolution fusion, and an FPN, then introduces two variants: one using the LION optimizer and one additionally using an adjusted FPN size of 1024. The three configurations are combined by majority voting. The authors report 87.34% accuracy on an augmented 474-image test set and 83.75% on an 80-image original test set, claiming that the ensemble outperforms conventional SOTA CNNs and recent FGVR methods, and that it surpasses all three individual configurations. The paper includes ablations of the number of selected regions and FPN size, per-class sensitivity/specificity/precision on the original test set, and Grad-CAM heatmaps.

Significance. If the empirical claims were statistically supported, the paper would make a useful contribution by framing wrist pathology recognition as a fine-grained problem on a realistic limited dataset, by adding an XAI component, and by comparing against a broad set of 15 conventional and 9 FGVR baselines. The ablation study over selection counts and FPN sizes, as well as per-class metrics on a challenging original-image test set, are also useful. However, the central comparative claim rests on point estimates from very small test sets, with no confidence intervals, no significance tests, and hyperparameter choices made on the same test sets used for final evaluation. These issues currently prevent the claimed superiority of the ensemble from being established.

major comments (5)
  1. [Section IV-A, Tables II and V] The statement in Section IV-A that 'our ensemble approach surpasses the performance of all three individual configurations' is not supported by the reported numbers. On test set 2, the ensemble and PIM+LION both achieve 83.75%; with 80 images this is an exact tie (67/80 correct). On test set 1, the ensemble beats PIM+LION by 1.90 percentage points and PIM+LION+1024FPN by 1.64 points, which correspond to roughly 9 and 8 images out of 474, respectively. Without confidence intervals or a paired significance test, the 'surpasses' claim is not statistically justified.
  2. [Section IV-A, Tables III-V] The final accuracy claims are undermined by test-set-driven hyperparameter selection. Table III selects the default 'Number of Selections' and Table IV evaluates FPN sizes using accuracy on test sets 1 and 2; the ensemble composition and the FPN-size variant (1024) are then chosen based on that same evaluation, and the final accuracies in Table V are reported on the same test sets. This creates a circular validation loop and can inflate apparent gains. An independent validation split or nested cross-validation is needed to support the claim that the ensemble is superior.
  3. [Section III-A, Tables I and VIII] Test set 2 contains only 80 original images, with class sizes of 17, 25, 15, and 23 (Table I). On this set, a one-image difference changes accuracy by 1.25 percentage points, so the reported gaps between models are well within plausible sampling noise. In addition, Table I indicates that test set 1 is obtained by augmenting the original test images to about 120 per class, so test set 1 contains multiple augmented copies of the same images; scores on test set 1 are therefore not independent of test set 2 and may be optimistically influenced by augmentation overlap with training-style transformations. The paper should report confidence intervals, exact paired comparisons, or at minimum the number of images behind each key accuracy difference.
  4. [Section III-A and Table VI] The clinically most important class, fracture, does not support the ensemble claim. On test set 2, the ensemble's fracture sensitivity is 98%, while the base PIM attains 100% (Table VI). The ensemble's aggregate accuracy gain over the base is 1.25 percentage points (one image), and its accuracy is identical to PIM+LION. Thus the reported advantage is not concentrated in the fracture class and may reflect noise rather than a real improvement in diagnostic utility. The authors should discuss this explicitly and provide class-level uncertainty estimates.
  5. [Section III-A, dataset curation] The paper does not state whether the train/test split is performed at the patient level. GRAZPEDWRI contains multiple images per patient (20,327 images from 6,091 patients), and the text only says that '20% of data from each class (except fracture) is allocated for testing.' If images from the same patient appear in both training and testing, the reported accuracies could be inflated by patient-level leakage. This must be clarified, and if necessary the evaluation should be redone with a patient-exclusive split.
minor comments (4)
  1. [Abstract and Section IV-B] The phrase 'outperformed many conventional SOTA and FGVR techniques' is vague; the paper should state specifically which comparisons are statistically meaningful and report uncertainty measures for all headline numbers.
  2. [Section IV-A, Table VI] The class labels in Table VI are given only as 0, 1, 2, 3; for readability and clinical interpretation, the class names (Boneanomaly, Fracture, Metal, Softtissue) should appear in the table or its caption.
  3. [Section III-D] The experimental settings do not report the number of random seeds, the variance across runs, or the code/configuration release. Reporting mean and standard deviation over multiple runs, or at least the exact seed and code availability, would greatly improve reproducibility.
  4. [Section III-A] The description of the two test sets is difficult to follow, especially the sentence 'we keep the original images for each class (deemed as challenging test set) and reduce the fracture class to 120 and 25 images respectively.' Rewriting this to clearly state the construction of test set 1 and test set 2 would remove ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the central ensemble claim is an empirical comparison built on an external base architecture and external baselines; the main weakness is statistical test-set reuse, not a circular derivation.

full rationale

The paper makes no first-principles derivation; its central claim is an empirical accuracy comparison. The Plug-in Module is taken from Chou et al. [7], LION from Chen et al. [19], and all compared baselines are external work. The only self-citation ([5]) appears in related work and is not used to justify any result. The ablation in Tables III-IV selects 'Number of Selections' and FPN size using the same two test sets later used for final comparison, which is a statistical validity concern (test-set-driven model selection, a test-set-2 tie with PIM+LION, and very small test-set-2 sizes), but it does not make any reported number equivalent to an input by construction. The ensemble's 87.34% on test set 1 is not equal to the best component's 85.70%, and majority voting is not defined so as to guarantee superiority, so the claim is not forced by definition. No circular step can be exhibited under the required standard of quoting a specific reduction or fitted parameter renamed as a prediction.

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

The paper contributes no new mathematical or physical entities. The main ledger items are hyperparameters tuned against the test sets, dataset curation choices, and the transferred fine-grained module assumptions; all affect the reliability of the reported accuracy advantage.

free parameters (4)
  • FPN size = 1536 (default; 1024 in ensemble variant)
    In Section IV-A Table IV, FPN size was varied from 512 to 3000; 1536 was retained because it gave the highest accuracy on test set 2, and 1024 was used in the ensemble because it gave the highest accuracy on test set 1.
  • Number of Selections per backbone block = (2048,512,128,32)
    Table III scans seven configurations; the default was retained because it achieved the highest accuracy on both test sets.
  • LION learning rate = 5e-6
    Set in Section III-D; no sensitivity analysis is reported, and accuracy gains from LION could depend on this specific value.
  • Data augmentation parameters = unspecified 'optimal specific values'
    Section III-A states Keras ImageDataGenerator settings for rotation, shift, zoom, and brightness but omits the actual numerical ranges, making exact replication impossible.
assumptions (4)
  • domain assumption The four curated GRAZPEDWRI class labels (bone anomaly, fracture, metal, soft tissue) are accurate and representative of wrist pathology.
    Used throughout as ground truth in Section III-A; no radiologist re-adjudication on the curated subset is reported.
  • domain assumption Test-set accuracy on 80 original images ranks models reliably.
    Section IV-B Table VIII uses test set 2 to select top models and report final superiority; differences of one image change accuracy by 1.25 points.
  • domain assumption The plug-in module's weakly supervised pixel selection, designed for natural images, transfers to X-ray radiographs without modification.
    Section III-B adopts the mechanism from Chou et al. [7]; no medical-image-specific adaptation is described.
  • domain assumption The class-balancing operations (downsampling fracture, augmenting all classes) preserve the real-world distribution.
    Section III-A describes the balancing; if real X-rays present multiple pathologies or different prevalence, the curated distribution differs from clinical practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Navigating limitations with precision: A fine-grained ensemble approach to wrist pathology recognition on a limited x-ray dataset." pith.science (2026). https://pith.science/paper/HR2DGTPA

@misc{pith2026241213884,
  author       = {Pith},
  title        = {Pith review of: Navigating limitations with precision: A fine-grained ensemble approach to wrist pathology recognition on a limited x-ray dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HR2DGTPA}},
  note         = {Machine review of arXiv:2412.13884}
}
read the original abstract

The exploration of automated wrist fracture recognition has gained considerable research attention in recent years. In practical medical scenarios, physicians and surgeons may lack the specialized expertise required for accurate X-ray interpretation, highlighting the need for machine vision to enhance diagnostic accuracy. However, conventional recognition techniques face challenges in discerning subtle differences in X-rays when classifying wrist pathologies, as many of these pathologies, such as fractures, can be small and hard to distinguish. This study tackles wrist pathology recognition as a fine-grained visual recognition (FGVR) problem, utilizing a limited, custom-curated dataset that mirrors real-world medical constraints, relying solely on image-level annotations. We introduce a specialized FGVR-based ensemble approach to identify discriminative regions within X-rays. We employ an Explainable AI (XAI) technique called Grad-CAM to pinpoint these regions. Our ensemble approach outperformed many conventional SOTA and FGVR techniques, underscoring the effectiveness of our strategy in enhancing accuracy in wrist pathology recognition.

Figures

Figures reproduced from arXiv: 2412.13884 by the authors.

Figure 1
Figure 1. Illustration of dataset curation steps. 20% of data from each class (except “fracture”) is allocated for testing. To address the class imbalance, we employ a downsampling strategy for the “fracture” class, and the extent of this downsampling is contingent on the number of augmen￾tations applied to each class. For testing purposes, we adopt two distinct approaches: in one testing set, we augment the existing images t… view at source ↗
Figure 2
Figure 2. Pipeline of plug-in module for fine-grained wrist pathology recognition. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Ensemble pipeline for the Plug-in module, incorporating three configurations: base, base + LION, and base + LION + [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: LION Optimization Pseudocode. D. Experimental Settings We complemented our fine-grained approach with a set of established models, each with distinct architectural principles and widely benchmarked in many studies. The models encom￾pass EfficientNetV2 [21], EfficientNe…
Figure 5
Figure 5. Figure 5: Heatmaps produced from the three employed configurations of the plug-in module for FGVR. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 27 canonical work pages

  1. [1]

    Epi- demiology of fractures in children and adolescents,

    E. M. Hedstrom, O. Svensson, U. Bergstrom, and P. Michno, “Epi- demiology of fractures in children and adolescents,” Acta Orthopaedica, vol. 81, pp. 148–153, 2010

    work page 2010

  2. [2]

    Most frequently missed fractures in the emergency department,

    J. Mounts, J. Clingenpeel, E. McGuire, E. Byers, and Y . Kireeva, “Most frequently missed fractures in the emergency department,” Clin Pediatr (Phila), vol. 50, pp. 183–186, 2011

    work page 2011

  3. [3]

    Overlooked extremity fractures in the emergency department,

    E. Er, P. Kara, O. Oyar, and E. Unluer, “Overlooked extremity fractures in the emergency department,” Ulus Travma Acil Cerrahi Derg, vol. 19, pp. 25–28, 2013

    work page 2013

  4. [4]

    Artificial intelligence solutions for analysis of x-ray images,

    S. J. Adams, R. D. E. Henderson, X. Yi, and P. Babyn, “Artificial intelligence solutions for analysis of x-ray images,” Canadian Associ- ation of Radiologists journal = Journal l’Association canadienne des radiologistes, vol. 846537120941671, 2020

    work page 2020

  5. [5]

    Enhancing wrist abnormality detection with yolo: Analysis of state-of-the-art single-stage detection models,

    A. Ahmed, A. S. Imran, A. Manaf, Z. Kastrati, and S. M. Daudpota, “Enhancing wrist abnormality detection with yolo: Analysis of state-of-the-art single-stage detection models,” Biomedical Signal Processing and Control , vol. 93, July 2024. [Online]. Available: https://doi.org/10.1016/j.bspc.2024.106144

    arXiv 2024

  6. [6]

    D. Dhami. (2018) State of computer vision. [Online]. Available: https: //medium.com/@dhartidhami/state-of-computer-vision-c742a773bf2e

    work page 2018

  7. [7]

    A Novel Plug-in Module for Fine-Grained Visual Classification

    P.-Y . Chou, C.-H. Lin, and W.-C. Kao, “A novel plug-in module for fine-grained visual classification,” 2022. [Online]. Available: https://arxiv.org/abs/2202.03822

    work page Pith review arXiv 2022

  8. [8]

    A pediatric wrist trauma x-ray dataset (grazpedwri-dx) for machine learning,

    E. Nagy, M. Janisch, F. Hr ˇzi´c, E. Sorantin, and S. Tschauner, “A pediatric wrist trauma x-ray dataset (grazpedwri-dx) for machine learning,” Nature News , May 2022. [Online]. Available: https: //www.nature.com/articles/s41597-022-01328-z

    work page 2022

Show all 42 references
  1. [9]

    Arm fracture de- tection in x-rays based on improved deep convolutional neural network,

    B. Guan, G. Zhang, J. Yao, X. Wang, and M. Wang, “Arm fracture de- tection in x-rays based on improved deep convolutional neural network,” Computer and Electrical Engineering , vol. 81, p. 106530, 2020

    2020

  2. [10]

    Musculoskeletal images classification for detection of fractures using transfer learning,

    I. Kandel, M. Castelli, and A. Popovi ˇc, “Musculoskeletal images classification for detection of fractures using transfer learning,” Journal of Imaging , November 2020. [Online]. Available: https: //www.ncbi.nlm.nih.gov/pmc/articles/PMC8321195/

    2020

  3. [11]

    Parallelnet: Multiple backbone network for detection tasks on thigh bone fracture,

    M. Wang, J. Yao, G. Zhang, B. Guan, X. Wang, and Y . Zhang, “Parallelnet: Multiple backbone network for detection tasks on thigh bone fracture,” Multimedia Systems, vol. 27, pp. 1091–1100, 2021

    2021

  4. [12]

    Detection and localization of hand fractures based on ga faster r-cnn,

    L. Xue, W. Yan, P. Luo, X. Zhang, T. Chaikovska, K. Liu, W. Gao, and K. Yang, “Detection and localization of hand fractures based on ga faster r-cnn,” Alexandria Engineering Journal , vol. 60, pp. 4555– 4562, 2021

    2021

  5. [13]

    Critical eval- uation of deep neural networks for wrist fracture detection,

    A. Raisuddin, E. Vaattovaara, M. Nevalainen, and et al., “Critical eval- uation of deep neural networks for wrist fracture detection,” Scientific Reports, vol. 11, p. 6006, 2021

    2021

  6. [14]

    Bone fracture detection through the two-stage system of cracksensitive convolutional neural network,

    Y . Ma and Y . Luo, “Bone fracture detection through the two-stage system of cracksensitive convolutional neural network,” Informatics in Medicine, vol. 236, pp. 24–40, 2021

    2021

  7. [15]

    Application of convolutional neural networks for distal radio-ulnar fracture detection on plain radiographs in the emergency room,

    M. W. Kim, J. Jung, S. J. Park, Y . S. Park, J. H. Yi, W. S. Yang, J. H. Kim, B.-J. Cho, and S. O. Ha, “Application of convolutional neural networks for distal radio-ulnar fracture detection on plain radiographs in the emergency room,” Clinical and Experimental Emergency Medic...

    2021

  8. [16]

    Deep learning-based localization and segmentation of wrist fractures on x-ray radiographs,

    D. Joshi, T. Singh, and A. Joshi, “Deep learning-based localization and segmentation of wrist fractures on x-ray radiographs,” Neural Comput & Applic, vol. 34, pp. 19 061–19 077, 2022

    2022

  9. [17]

    Fracture detection in wrist x- ray images using deep learning-based object detection models,

    F. Hardalac ¸, F. Uysal, O. Peker, M. C ¸ ic ¸eklida˘g, T. Tolunay, N. Tokg ¨oz, U. Kutbay, B. Demirciler, and F. Mert, “Fracture detection in wrist x- ray images using deep learning-based object detection models,” Sensors, vol. 22, no. 3, p. 1285, 2022

    2022

  10. [18]

    Fracture recognition in paediatric wrist radiographs: An object detection approach,

    F. Hr ˇzi’c et al. , “Fracture recognition in paediatric wrist radiographs: An object detection approach,” Mathematics, vol. 10, no. 16, p. 2939, 2022

    2022

  11. [19]

    Symbolic discovery of optimization algorithms,

    X. Chen, L. Chen, D. Huang, E. Real, K. Wang, Y . Liu, H. Pham, X. Dong, T. Luong, C.-L. Hsieh, Y . Lu, and Q. V . Le, “Symbolic discovery of optimization algorithms,” 2023. [Online]. Available: https://arxiv.org/abs/2302.06675

    2023 arXiv

  12. [20]

    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,” arXiv preprint arXiv:2103.14030 , 2021. [Online]. Available: https://arxiv.org/abs/2103.14030

    2021 arXiv

  13. [21]

    Efficientnetv2: Smaller models and faster training,

    M. Tan and Q. V . Le, “Efficientnetv2: Smaller models and faster training,” June 2021. [Online]. Available: https://arxiv.org/abs/2104. 00298

    2021

  14. [22]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    T. Mingxing and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” September 2020. [Online]. Available: https://arxiv.org/abs/1905.11946

    2020 arXiv

  15. [23]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” April 2015. [Online]. Available: https://arxiv.org/abs/1409.1556

    2015 arXiv

  16. [24]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” 2020. [Online]. Available: https://arxi...

    2020 arXiv

  17. [25]

    Regnet: Self- regulated network for image classification,

    J. Xu, Y . Pan, X. Pan, S. Hoi, Z. Yi, and Z. Xu, “Regnet: Self- regulated network for image classification,” 2021. [Online]. Available: https://arxiv.org/abs/2101.00590

    2021 arXiv

  18. [26]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” 2018. [Online]. Available: https://arxiv.org/abs/1608.06993

    2018 arXiv

  19. [27]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” March 2019. [Online]. Available: https://arxiv.org/abs/1801.04381

    2019 arXiv

  20. [28]

    Rexnet: Diminishing representa- tional bottleneck on convolutional neural network,

    D. Han, S. Yun, B. Heo, and Y . Yoo, “Rexnet: Diminishing representa- tional bottleneck on convolutional neural network,” 2020

    2020

  21. [29]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” December 2015. [Online]. Available: https: //arxiv.org/abs/1512.03385

    2015 arXiv

  22. [30]

    Resnest: Split-attention networks,

    H. Zhang, C. Wu, Z. Zhang, Y . Zhu, H. Lin, Z. Zhang, Y . Sun et al. , “Resnest: Split-attention networks,” December 2020. [Online]. Available: https://arxiv.org/abs/2004.08955

    2020 arXiv

  23. [31]

    Going deeper with convolutions,

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” September 2014. [Online]. Available: https://arxiv.org/abs/1409.4842

    2014 arXiv

  24. [32]

    [Online]

    (2023) Ultralytics. [Online]. Available: https://github.com/ultralytics/ ultralytics

    2023

  25. [33]

    SIM-Trans: Structure Information Modeling Transformer for Fine-Grained Visual Categorization,

    H. Sun, X. He, and Y . Peng, “SIM-Trans: Structure Information Modeling Transformer for Fine-Grained Visual Categorization,” arXiv.org, August 2022. [Online]. Available: https://arxiv.org/abs/2208.14607v1

    2022 arXiv

  26. [34]

    Learn from each other to classify better: Cross-layer mutual attention learning for fine-grained visual classification,

    D. Liu, L. Zhao, Y . Wang, and J. Kato, “Learn from each other to classify better: Cross-layer mutual attention learning for fine-grained visual classification,” Pattern Recognition, vol. 140, p. 109550, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/...

    2023

  27. [35]

    ViT-NeT: Interpretable Vision Transformers with Neural Tree Decoder,

    S. Kim, J. Nam, and B. C. Ko, “ViT-NeT: Interpretable Vision Transformers with Neural Tree Decoder,” ser. PMLR, June 2022. [Online]. Available: https://proceedings.mlr.press/v162/kim22g.html

    2022

  28. [36]

    Fine-grained visual classification via internal ensemble learning transformer,

    Q. Xu, J. Wang, B. Jiang, and B. Luo, “Fine-grained visual classification via internal ensemble learning transformer,” IEEE Transactions on Multimedia, pp. 1–14, 2023

    2023

  29. [37]

    TransFG: A Transformer Architecture for Fine- Grained Recognition,

    J. He, J.-N. Chen, S. Liu, A. Kortylewski, C. Yang, Y . Bai, and C. Wang, “TransFG: A Transformer Architecture for Fine- Grained Recognition,” arXiv.org, December 2021. [Online]. Available: https://arxiv.org/abs/2103.07976v5

    2021 arXiv

  30. [38]

    MetaFormer: A Unified Meta Framework for Fine-Grained Recognition,

    Q. Diao, Y . Jiang, B. Wen, J. Sun, and Z. Yuan, “MetaFormer: A Unified Meta Framework for Fine-Grained Recognition,” arXiv.org, March 2022. [Online]. Available: https://arxiv.org/abs/2203.02751v1

    2022 arXiv

  31. [39]

    Context-Aware Attentional Pooling (CAP) for Fine-Grained Visual Classification,

    A. Behera, Z. Wharton, P. Hewage, and A. Bera, “Context-Aware Attentional Pooling (CAP) for Fine-Grained Visual Classification,” arXiv.org, 2021. [Online]. Available: https://arxiv.org/abs/2101.06635v1

    2021 arXiv

  32. [40]

    Feature Fusion Vision Transformer for Fine-Grained Visual Categorization,

    J. Wang, X. Yu, and Y . Gao, “Feature Fusion Vision Transformer for Fine-Grained Visual Categorization,” arXiv.org, February 2022. [Online]. Available: https://arxiv.org/abs/2107.02341v3

    2022 arXiv

  33. [41]

    Fine-Grained Visual Classification with High-Temperature Refinement and Background Suppression,

    P.-Y . Chou, Y .-Y . Kao, and C.-H. Lin, “Fine-Grained Visual Classification with High-Temperature Refinement and Background Suppression,” arXiv.org, April 2023. [Online]. Available: https: //arxiv.org/abs/2303.06442v2

    2023 arXiv

  34. [42]

    Reject rate analysis in digital radiography: an australian emergency imaging department case study,

    S. Atkinson, M. Neep, and D. Starkey, “Reject rate analysis in digital radiography: an australian emergency imaging department case study,” Journal of Medical Radiation Sciences , vol. 67, no. 1, pp. 72–79, 2019

    2019

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.