Pith. sign in

REVIEW 3 major objections 5 minor 49 references

Integrating Skeleton Based Representations for Robust Yoga Pose Classification Using Deep Learning Models

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read The paper's central claim is that skeletal renderings of the body carry more pose-discriminating signal than full RGB images, and on the new Yoga-16 benchmark VGG16 with MediaPipe skeleton input reaches 96.09% accuracy.

desk verdict A useful curated dataset and a plausible skeleton-beats-RGB direction, but the 96.09% headline is inflated by test-set hyperparameter selection and a self-contradictory training protocol. read the letter →

arxiv 2512.00572 v2 pith:OW2YZHNV submitted 2025-11-29 cs.CV cs.AI

Mohammed Mohiuddin , Syed Mohammod Minhaz Hossain , Sumaiya Khanam , Prionkar Barua , Aparup Barua , MD Tamim Hossain This is my paper · ORCID
classification cs.CVcs.AI
keywords HumanPoseEstimationYogaClassificationDeepLearningMediaPipeYOLOv8Skeleton-BasedRepresentationsYoga-16datasetTransfer
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

This paper sets out to show that representing a yoga pose as a skeleton—body joints and the lines between them, stripped of background and clothing—makes pose classification more accurate than feeding a CNN the original photograph. To test this, the authors built Yoga-16, a curated and balanced dataset of 16 poses with overlapping visual features, and ran three architectures (VGG16, ResNet50, Xception) on three input types: raw images, MediaPipe Pose skeletons, and YOLOv8-Pose skeletons. Skeleton inputs won across the board; the best model, VGG16 on MediaPipe skeletons, reached 96.09% accuracy, versus 86.33% for the same network on raw images. If this holds, automated yoga coaching could rely on lightweight, real-time skeleton extraction rather than heavy image analysis, and the benchmark gives the field a common test bed for comparing approaches.

What carries the argument

The skeleton-image representation is the load-bearing mechanism: a pose-estimation model (MediaPipe Pose or YOLOv8-Pose) detects body keypoints and renders them as a clean skeletal overlay, which is then fed to a CNN trained from scratch. This transformation strips away background, clothing, and lighting so the classifier is forced to learn from joint geometry alone. VGG16's sequential stack of small 3×3 filters with 2×2 max pooling turned out to be the best exploiter of these skeletal images, and hyperparameter tuning (filter size, pooling, batch size, optimizer, dense width) confirmed those settings.

What would settle it

Run the identical VGG16 pipeline on uncurated or multi-person images where keypoint extraction is imperfect: if accuracy on skeleton inputs drops below raw-image accuracy, the claim that skeletons generally outperform photos is falsified. A second test: include a pose class that is distinguishable mainly by clothing or props (e.g., a pose holding a block); skeleton inputs would likely fail where raw images succeed.

Watch

Extended reading notes

Core claim

The paper's central discovery is that input representation, not architecture, is the dominant factor in yoga pose classification: converting each image into a skeleton visualization before classification consistently improves accuracy across all three CNN backbones. The best configuration, VGG16 with MediaPipe Pose skeleton images, achieves 96.09% accuracy, 96.27% precision, 96.09% recall, and 96.10% F1-score on Yoga-16—about ten points above the same VGG16 on raw photographs. The authors report that MediaPipe's 33-keypoint skeletons, rendered without background, let the network attend to limb alignments and joint angles, and that this advantage persists under 5-fold cross-validation (93.55

Load-bearing premise

The central claim rests on assuming that a clean skeleton image carries enough information to distinguish all 16 poses, an assumption the dataset curation was built to satisfy by keeping only single-subject, well-cropped, high-quality images.

Editorial extensions

If this is right

  • On Yoga-16, switching VGG16 from raw photos to MediaPipe skeletons raises accuracy from 86.33% to 96.09%, showing representation choice can matter more than model choice.
  • The skeleton advantage holds across all three architectures: every network scores higher on either skeleton input than on direct images, so the result is not specific to VGG16.
  • The 93.75% accuracy on a YouTube-derived test set suggests the pipeline transfers to less controlled imagery, supporting real-world deployment.
  • Because MediaPipe skeleton extraction is lightweight and real-time, the result implies on-device yoga coaching is feasible without sacrificing accuracy.
  • The 5-fold cross-validation mean of 93.55 ± 0.94% indicates the reported accuracy is stable across data splits, not a lucky split artifact.

Reading between the lines

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

  • A likely boundary condition: the skeleton advantage depends on the curation choices in Yoga-16 (single subject, cropped, high-quality, no clutter). On messier multi-person or heavily occluded images, keypoint extraction errors could shrink or reverse the gap, so real systems may need a person-detection front end before skeletonization.
  • The consistent gap between MediaPipe and YOLOv8-Pose skeletons (e.g., VGG16: 96.09% vs 91.41%) suggests keypoint stability matters more than keypoint count; re-running the same protocol with other keypoint estimators would test whether the ranking is estimator-agnostic.
  • Misclassifications between poses with overlapping skeletal features (Dolphin Plank vs. Fish, Dolphin Plank vs. Wide Angle Seated Forward Bend) point to a concrete next experiment: adding pose-specific data augmentation or an auxiliary loss on joint angles could directly target the failure modes the authors identified.
  • Because training was from scratch on only 896 training images, the accuracy ceiling may be much higher with more data; a scaled-up version of Yoga-16 could reveal whether skeleton inputs retain their edge when appearance cues become available in larger variety.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The manuscript introduces a curated dataset, Yoga-16, containing 16 yoga pose classes with 1,280 images (80 per class), and reports a benchmark comparing VGG16, ResNet50, and Xception under three input modalities: direct RGB images, MediaPipe Pose skeleton images, and YOLOv8-Pose skeleton images. The paper's central empirical claim is that skeleton-based representations outperform raw images for yoga pose classification, with the best result being VGG16 on MediaPipe skeleton inputs at 96.09% accuracy. The authors also report 5-fold cross-validation (93.55±0.94%), a custom YouTube-derived test set (93.75% for the best model), Grad-CAM visualizations, and a qualitative comparison with prior work.

Significance. If the comparison is trustworthy, the paper provides a useful empirical benchmark and a reusable dataset for yoga pose classification. The public release of the dataset and code is a concrete strength that supports reproducibility. The contribution is primarily experimental and incremental rather than methodological: the architectures and skeleton extractors are established tools, and the main novelty lies in the systematic three-modality, three-architecture comparison. The dataset is small and deliberately curated, so the headline accuracy should be read as a proof-of-concept result on clean, single-subject, well-cropped images rather than as a claim about unconstrained real-world yoga videos.

major comments (3)
  1. [§3.6, §3.6.1–3.6.3, Tables 6–8] The training protocol is stated inconsistently. Section 3.6 says the CNN models are trained from scratch because pretrained ImageNet weights are not well aligned with skeleton inputs. However, §3.6.1 describes fine-tuning VGG16 by unfreezing the last dense layers and partially the deeper convolutional blocks; Table 6 marks all VGG16 layers trainable, while Tables 7 and 8 mark the ResNet50 and Xception backbone stages/flow layers as not trainable. These are different training protocols, and if actually used, the accuracy differences in Table 12 confound input modality with initialization and layer-freezing choices. The authors should specify the exact protocol per experiment and, ideally, rerun the nine experiments with a single consistent training protocol before drawing the central conclusion that skeleton inputs are superior to raw images.
  2. [§4.5, Table 14] The hyperparameter tuning experiments are reported using 'Overall Accuracy' without stating which data split is used for model selection. The best configuration (Experiment 1) reports exactly 96.09%, the same number as the headline result in Table 12, which raises the concern that the test split was used to select hyperparameters. No validation-set accuracies are given for the eight tuning runs. Because Table 12 is the basis for the paper's central claim, this is load-bearing. The authors should clearly report the split used in Table 14 and provide the corresponding validation results; the final test accuracy should be computed only after the hyperparameters are fixed. The 5-fold cross-validation in Table 15 was applied after the model was already selected, so it does not remove this selection bias.
  3. [§4.6, §3.2, Table 16] The external generalization evidence does not directly support the central modality comparison. The custom YouTube test set is evaluated only for VGG16 with MediaPipe skeleton inputs; no direct-image or YOLOv8-skeleton models are tested on it, so it cannot establish whether the claimed skeleton advantage persists under distribution shift. In addition, Yoga-16 curation explicitly excludes multi-subject, low-quality, non-human, and zoomed-out images (Section 3.2, Fig. 2), and the main test set is a random split from the same curated pool. The 96.09% vs. 86.33% gap therefore reflects a clean, curated setting. To support the broad conclusion that skeleton-based representations are more robust, the authors should compare raw-image and skeleton variants on the same external set, or explicitly scope the claim to the curated setting.
minor comments (5)
  1. [§4.3] The first sentence after the VGG16 description says the confusion matrix, loss/accuracy curves, and ROC curve are for Xception with YOLOv8-Pose input, but the surrounding text is discussing VGG16. The figure references (S19–S21) appear to be misassigned.
  2. [§4.2 and §4.3 headings] The section headings use the phrase 'Pretrained Deep Learning Models', which is inconsistent with Section 3.6's statement that models are trained from scratch. Please align the terminology.
  3. [Table 13] The per-class 'Accuracy' column is unusual because, for a single-label test set with one class per image, per-class accuracy coincides with recall. Clarify the definition or rename it to avoid confusion.
  4. [Table 17] The comparison with prior work is qualitative ('R/NR'). State whether categories such as 'low resolution' and 'class imbalance' are measured objectively or are author judgments.
  5. [References [34], [36], [37]] The self-citations for the dataset, custom test set, and code are appropriate for data availability, but the custom test set's construction and selection criteria should be described in the main text rather than left entirely to the referenced repository.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 96.09% headline is a measured benchmark result, self-citations are data releases, and the main risks are evaluation-protocol issues rather than reductions.

full rationale

This paper is an empirical benchmark, not a derivation. The central claim (skeleton inputs outperform raw images, with VGG16+MediaPipe at 96.09%) is a measured test-set accuracy difference reported in Table 12; no equation defines the reported accuracy in terms of the model inputs or the labels. The same-author citations [34], [36], and [37] are dataset/code releases and a YouTube-derived custom test set. They do not smuggle in an external theorem, and the custom test set is not used for training, so the generalization check is a self-evaluation rather than a circular reduction. The more serious concerns are evaluation-protocol issues, not circularity: Section 4.5/Table 14 does not state which split is used for hyperparameter selection, and if the 'Overall Accuracy' column is computed on the test split, the reported 96.09% would be a selection artifact; likewise, Section 3.6 says models are trained from scratch while Section 3.6.1 and Tables 6–8 describe fine-tuning with frozen layers. These would undermine the internal validity of the comparison but are not cases where the output equals the input by construction. The paper's own limitation statement ('the relatively limited size of the Yoga-16 dataset raises concerns regarding generalization') is a weakness, not evidence of circularity. No load-bearing step reduces to a definition, a fitted parameter disguised as a prediction, or a self-citation chain.

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

No new scientific entities are invented; the only postulates are data-quality assumptions and the curation decisions. The most important ledger entries are the 80-per-class balancing and the exclusions that make the clean test environment.

free parameters (2)
  • Dataset curation choices = 80 images per class; exclusions of multi-subject/low-quality/non-human/zoomed-out images
    The dataset construction is the key hand-made ingredient. The curation rules and the 80-per-class balance determine the reported accuracy and are not independently validated. All main results are measured on a random split of this curated set.
  • Hyperparameter configuration of the best model = 3x3 filters, 2x2 pooling, batch size 32, Adam, dense 1024
    Table 14 shows the best configuration was selected from a small grid. The final 96.09% is reported on the held-out split after this selection; no correction for multiple comparisons is described. This is a standard, mild 'selection after tuning' parameter.
assumptions (2)
  • domain assumption Skeleton renderings preserve the pose-discriminative information needed for the 16-class task.
    The whole claim that skeletons are superior depends on this. It is assumed in Section 3.2/3.6 and supported only by the in-dataset accuracy numbers and the small custom test. It is not proved and is exactly what the benchmark purports to test.
  • domain assumption The images in the test split are representative of the images a real system would encounter without further curation.
    The test set is a random split from the same curated pool (Section 3.2, Table 5). Any deployment setting with multiple subjects, occlusion, or unusual crops is underrepresented. The custom test set is meant to mitigate this but is small and under-described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Integrating Skeleton Based Representations for Robust Yoga Pose Classification Using Deep Learning Models." pith.science (2026). https://pith.science/paper/OW2YZHNV

@misc{pith2026251200572,
  author       = {Pith},
  title        = {Pith review of: Integrating Skeleton Based Representations for Robust Yoga Pose Classification Using Deep Learning Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OW2YZHNV}},
  note         = {Machine review of arXiv:2512.00572}
}
read the original abstract

Yoga is a popular form of exercise worldwide due to its spiritual and physical health benefits, but incorrect postures can lead to injuries. Automated yoga pose classification has therefore gained importance to reduce reliance on expert practitioners. While human pose keypoint extraction models have shown high potential in action recognition, systematic benchmarking for yoga pose recognition remains limited, as prior works often focus solely on raw images or a single pose extraction model. In this study, we introduce a curated dataset, 'Yoga-16', which addresses limitations of existing datasets, and systematically evaluate three deep learning architectures (VGG16, ResNet50, and Xception), using three input modalities (direct images, MediaPipe Pose skeleton images, and YOLOv8 Pose skeleton images). Our experiments demonstrate that skeleton-based representations outperform raw image inputs, with the highest accuracy of 96.09% achieved by VGG16 with MediaPipe Pose skeleton input. Additionally, we provide interpretability analysis using Grad-CAM, offering insights into model decision-making for yoga pose classification with cross-validation analysis.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 6 canonical work pages

  1. [1]

    https://doi.org/10.1016/j.dsm.2024.11.003

    Saha U, Minhaz Hossain SM and Sarker IH 2024 Predicting depression level based on human activities and feelings: A fuzzy logic-based analysis.Data Science and Management. https://doi.org/10.1016/j.dsm.2024.11.003

    work page doi:10.1016/j.dsm.2024.11.003 2024

  2. [2]

    Ann Data Sci: 1–32

    Akhter M, Hossain SMM, Nigar RS, Paul S, Kamal KMA, Sen A, Sarker IH (2024) Covid-19 fake news detection using deep learning model. Ann Data Sci: 1–32

    2024

  3. [3]

    In: Proc 2nd Int Conf Electr Comput Telecommun Eng (ICECTE), IEEE, pp 1–4

    Alam KT, Hossain SMM, Arefin MS (2016) Developing a framework for analyz- ing social networks to identify human behaviours. In: Proc 2nd Int Conf Electr Comput Telecommun Eng (ICECTE), IEEE, pp 1–4

    2016

  4. [4]

    In: Int Conf Intell Comput Optim, Springer, pp 561–571

    Das D, Sen A, Hossain SMM, Deb K (2022) Trash image classification using transfer learning based deep neural network. In: Int Conf Intell Comput Optim, Springer, pp 561–571

    2022

  5. [5]

    Ann Data Sci 11(5):1821–1864

    Das T, Mobassirin S, Hossain SMM, Das A, Sen A, Kamal KMA, Deb K (2024) Patient questionnaires based Parkinson’s disease classification using artificial neural network. Ann Data Sci 11(5):1821–1864

    2024

  6. [6]

    In: Applied Intelligence for Industry 4.0, Chapman and Hall/CRC, pp 176–187

    Euna NJ, Hossain SMM, Anwar MM, Sarker IH (2023) A chapter in Applied Intelligence for Industry 4.0. In: Applied Intelligence for Industry 4.0, Chapman and Hall/CRC, pp 176–187

    2023

  7. [7]

    In: Adv Data Min Appl: Proc 16th Int Conf ADMA 2020, Foshan, China, Springer, pp 299–314

    Hossain SMM, Tanjil MMM, Ali MAB, Islam MZ, Islam MS, Mobassirin S, Sarker IH, Islam SR (2020) Rice leaf diseases recognition using convolutional neural networks. In: Adv Data Min Appl: Proc 16th Int Conf ADMA 2020, Foshan, China, Springer, pp 299–314

    2020

  8. [8]

    In: Intell Comput Optim: Proc 3rd Int Conf ICO 2020, Springer, pp 530–545

    Hossain SMM, Deb K (2021) Plant leaf disease recognition using histogram based gradient boosting classifier. In: Intell Comput Optim: Proc 3rd Int Conf ICO 2020, Springer, pp 530–545

    2021

Show all 49 references
  1. [9]

    Symmetry 13(3):511 30

    Hossain SMM, Deb K, Dhar PK, Koshiba T (2021) Plant leaf disease recognition using depth-wise separable convolution-based models. Symmetry 13(3):511 30

    2021

  2. [10]

    In: Int Conf Hybrid Intell Syst, Springer, pp 106–116

    Hossain SMM, Sumon JA, Sen A, Alam MI, Kamal KMA, Alqahtani H, Sarker IH (2021) Spam filtering of mobile SMS using CNN–LSTM based deep learning model. In: Int Conf Hybrid Intell Syst, Springer, pp 106–116

    2021

  3. [11]

    In: Intell Comput Optim: Proc 4th Int Conf ICO 2021, Springer, pp 341–351

    Hossain SMM, Kamal KMA, Sen A, Deb K (2022) Tomato leaf disease recognition using depthwise separable convolution. In: Intell Comput Optim: Proc 4th Int Conf ICO 2021, Springer, pp 341–351

    2022

  4. [12]

    Hossain SMM, Kamal KMA, Sen A, Sarker IH (2023) In: Applied Intelligence for Industry 4.0, Chapman and Hall/CRC, pp 162–175

    2023

  5. [13]

    In: 2023 Int Workshop Intell Syst (IWIS), IEEE, pp 1–6

    Hossain MS, Deb K, Hossain SMM, Jo KH (2023) Daily living human activity recognition using deep neural networks. In: 2023 Int Workshop Intell Syst (IWIS), IEEE, pp 1–6

    2023

  6. [14]

    In: Int Conf Intell Comput Optim, Springer, pp 265–274

    Maimuna M, Hossain SMM, Deb K (2022) Masked face inpainting using gener- ative adversarial network based architecture. In: Int Conf Intell Comput Optim, Springer, pp 265–274

    2022

  7. [15]

    In: 15th Int Conf Human Syst Interact (HSI), IEEE, pp 1–6

    Sen A, Hossain SMM, Russo MA, Deb K, Jo KH (2022) Fine-grained soccer actions classification using deep neural network. In: 15th Int Conf Human Syst Interact (HSI), IEEE, pp 1–6

    2022

  8. [16]

    In: Int Workshop Front Comput Vis, Springer, pp 312–324

    Sen A, Hossain SMM, Uddin R, Deb K, Jo KH (2022) Sequence recognition of indoor tennis actions using transfer learning and long short-term memory. In: Int Workshop Front Comput Vis, Springer, pp 312–324

    2022

  9. [17]

    (2019) Mediapipe: A frame- work for perceiving and processing reality

    Lugaresi C, Tang J, Nash H, McClanahan C, et al. (2019) Mediapipe: A frame- work for perceiving and processing reality. In: Third Workshop on Computer Vision for Augmented and Virtual Reality (CV4AR VR)

    2019

  10. [18]

    AGPL-3.0 License

    Jocher G, Chaurasia A, Qiu J (2023) Ultralytics YOLOv8.URL https://github.com/ultralytics/ultralytics. AGPL-3.0 License

    2023

  11. [19]

    In: IEEE/CVF Conf Comput Vis Pattern Recognit Workshops (CVPR W), pp 4472–4479

    Verma M, Kumawat S, Nakashima Y, Raman S (2020) Yoga-82: a new dataset for fine-grained classification of human poses. In: IEEE/CVF Conf Comput Vis Pattern Recognit Workshops (CVPR W), pp 4472–4479

    2020

  12. [20]

    J Ambient Intell Human Comput 14:16551–16562

    Garg S, Saxena A, Gupta R (2023) Yoga pose classification: a CNN and Medi- aPipe inspired deep learning approach for real-world application. J Ambient Intell Human Comput 14:16551–16562. https://doi.org/10.1007/s12652-022-03910-0

    2023 doi

  13. [21]

    arXiv preprint arXiv:1409.1556

    Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556

    2014 arXiv

  14. [22]

    In: Proc IEEE Conf Comput Vis Pattern Recognit (CVPR), pp 770–778

    He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recogni- tion. In: Proc IEEE Conf Comput Vis Pattern Recognit (CVPR), pp 770–778. 31 https://doi.org/10.1109/CVPR.2016.90

    2016 doi

  15. [23]

    In: Proc IEEE Conf Comput Vis Pattern Recognit (CVPR), pp 1251–1258

    Chollet F (2017) Xception: deep learning with depthwise separable convolutions. In: Proc IEEE Conf Comput Vis Pattern Recognit (CVPR), pp 1251–1258. URL https://arxiv.org/abs/1610.02357

    2017 arXiv

  16. [24]

    In: 3rd Int Conf Electr Comput Commun Mechatron Eng (ICECCME), pp 1–6

    Imran S, Sadman Z, Islam A, Karim DZ (2023) Enhanced yoga pos- ture detection using deep learning and ensemble modeling. In: 3rd Int Conf Electr Comput Commun Mechatron Eng (ICECCME), pp 1–6. https://doi.org/10.1109/ICECCME57830.2023.10252764

    2023

  17. [25]

    Appl Sci 10(4):1287

    Byeon YH, Lee JY, Kim DH, Kwak KC (2020) Posture recognition using ensemble deep models under various home environments. Appl Sci 10(4):1287. https://doi.org/10.3390/app10041287

    2020 doi

  18. [26]

    IEEE Sens J 21(7):9515–9522

    Liaqat S, Dashtipour K, Arshad K, Assaleh K, Ramzan N (2021) A hybrid posture detection framework: integrating machine learning and deep neural networks. IEEE Sens J 21(7):9515–9522. https://doi.org/10.1109/JSEN.2021.3055898

    2021

  19. [27]

    In: 11th Int Conf Reliab Infocom Technol Optim (ICRITO), pp 1–6

    Wadhwa S, Garg A, Munjal G (2024) Yoga posture analysis using deep learn- ing. In: 11th Int Conf Reliab Infocom Technol Optim (ICRITO), pp 1–6. https://doi.org/10.1109/ICRITO61523.2024.10522380

    2024

  20. [28]

    In: 9th IEEE Int Conf Commun Syst Netw Technol (CSNT), pp 40–43

    Agrawal Y, Shah Y, Sharma A (2020) Implementation of machine learning tech- nique for identification of yoga poses. In: 9th IEEE Int Conf Commun Syst Netw Technol (CSNT), pp 40–43. https://doi.org/10.1109/CSNT48778.2020.9115758

    2020

  21. [29]

    A, Sajan S, K.A

    Anilkumar A, K.T. A, Sajan S, K.A. S (2021) Pose estimated yoga monitoring system. In: Proc Int Conf IoT Based Control Networks & Intelligent Systems (ICICNIS 2021). https://doi.org/10.2139/ssrn.3882498

    2021 doi

  22. [30]

    J Supercomput 78:5269–5284

    Long C, Jo E, Nam Y (2022) Development of a yoga posture coaching system using an interactive display based on transfer learning. J Supercomput 78:5269–5284. https://doi.org/10.1007/s11227-021-04076-w

    2022 doi

  23. [31]

    J Supercomput 78:2873–2908

    Bilal M, Maqsood M, Yasmin S, Mohsin S, Ali T (2022) A trans- fer learning-based efficient spatiotemporal human action recognition frame- work for long and overlapping action classes. J Supercomput 78:2873–2908. https://doi.org/10.1007/s11227-021-03957-4

    2022 doi

  24. [32]

    In: IOP Conf Ser Mater Sci Eng 1110:012002

    Jose J, Shailesh S (2021) Yoga asana identification: a deep learning approach. In: IOP Conf Ser Mater Sci Eng 1110:012002. https://doi.org/10.1088/1757- 899X/1110/1/012002. Presented at: Annual International Conference on Data Science, Machine Learning and Blockchain Technolog...

    2021 doi

  25. [33]

    URL https://www.kaggle.com/datasets/niharika41298/yoga-poses-dataset

    Pandit N (2021) Yoga poses dataset. URL https://www.kaggle.com/datasets/niharika41298/yoga-poses-dataset. Accessed 02 Sept 2024

    2021

  26. [34]

    URL https://www.kaggle.com/datasets/mohiuddin2531/yoga- 16/data?select=yoga16-dataset

    Mohiuddin M (2024) Yoga-16 dataset. URL https://www.kaggle.com/datasets/mohiuddin2531/yoga- 16/data?select=yoga16-dataset. Accessed 26 Dec 2024

    2024

  27. [35]

    CoRR abs/2006.10204

    Bazarevsky V, Grishchenko I, Raveendran K, Zhu T, Zhang F, Grundmann M (2020) BlazePose: On-device real-time body pose tracking. CoRR abs/2006.10204. URL https://arxiv.org/abs/2006.10204

    2020 arXiv

  28. [36]

    URL https://www.kaggle.com/datasets/mohiuddin2531/yoga-16/data?select=yt test

    Mohiuddin M (2024) Custom test set for Yoga-16. URL https://www.kaggle.com/datasets/mohiuddin2531/yoga-16/data?select=yt test. Accessed 26 Dec 2024

    2024

  29. [37]

    URL https://github.com/mohiuddin2531/yoga-16

    Mohiuddin M (2024) Python script for implemented deep learning models. URL https://github.com/mohiuddin2531/yoga-16. Accessed 26 Dec 2024

    2024

  30. [38]

    A., Srivastava, A., Badal, T., Mishra, V

    Thoutam, V. A., Srivastava, A., Badal, T., Mishra, V. K., Sinha, G. R., Sakalle, A., & Raj, M. (2022) Yoga pose estimation and feedback generation using deep learning. Computational Intelligence and Neuroscience, 2022(1), 4311350

    2022

  31. [39]

    Q., Gao, Y., Ji, Z., Lv, J., Sajjatul Islam, M

    Mohammed, A. Q., Gao, Y., Ji, Z., Lv, J., Sajjatul Islam, M. D., & Sang, Y. (2021, November) Automatic 3D skeleton-based dynamic hand gesture recognition using multi-layer convolutional LSTM. In Proceedings of the 7th International Conference on Robotics and Artificial Intelli...

    2021

  32. [40]

    Fuad, K. N. R., Rozario, U., Shoaib, H. A., Khatun, M. M., Islam, M. R., Mridha, M. F., & Shin, J. (2024, September) Yoga Posture Image Classification Using Big Transfer (BiT). In 2024 IEEE International Conference on Computing, Applications and Systems (COMPAS) (pp. 1–6). IEEE

    2024

  33. [41]

    A., Lv, J., Islam, M

    Mohammed, A. A., Lv, J., Islam, M. S., & Sang, Y. (2023) Multi-model ensemble gesture recognition network for high-accuracy dynamic hand gesture recognition. Journal of Ambient Intelligence and Humanized Computing, 14(6), 6829–6842

    2023

  34. [42]

    M., & Mohamed, S

    Ali, M. M., & Mohamed, S. I. (2025) A pose estimation for motion tracking of infants cerebral palsy. Multimedia Tools and Applications, 84(10), 8261–8286

    2025

  35. [43]

    Muksimova, S., Valikhujaev, Y., Umirzakova, S., Baltayev, J., & Cho, Y. I. (2025) GazeCapsNet: A Lightweight Gaze Estimation Framework. Sensors, 25(4), 1224

    2025

  36. [44]

    (2022) Pedestrian Intention Prediction for Autonomous Vehicles: A Comprehensive Survey

    Sharma, A., Singh, P., & Kumar, V. (2022) Pedestrian Intention Prediction for Autonomous Vehicles: A Comprehensive Survey. Neurocomputing, 500, 1–27. 33

    2022

  37. [45]

    Rasouli, A., Kotseruba, I., & Tsotsos, J. K. (2020) Autonomous Vehicles That Interact With Pedestrians. IEEE Transactions on Intelligent Transportation Systems, 21(12), 5128–5144. sssssss

    2020

  38. [46]

    (2022) Predicting pedestrian crossing inten- tion with feature fusion and spatio-temporal attention

    Zhang, X., Li, Y., & Wang, H. (2022) Predicting pedestrian crossing inten- tion with feature fusion and spatio-temporal attention. IEEE Transactions on Intelligent Vehicles, 7(4), 1245–1257

    2022

  39. [47]

    (2023) Visual–Motion–Interaction-Guided Pedestrian Intention Prediction Framework

    Li, J., Chen, Y., & Zhao, Q. (2023) Visual–Motion–Interaction-Guided Pedestrian Intention Prediction Framework. IEEE Sensors Journal, 23(5), 4567–4578

    2023

  40. [48]

    (2025) Predicting pedestrian intentions with Multimodal IntentFormer

    Wang, K., & Liu, S. (2025) Predicting pedestrian intentions with Multimodal IntentFormer. Pattern Recognition, 142, 109577

    2025

  41. [49]

    (2022) Intelligent Pedestrian Intention Prediction Frame- work

    Chen, T., & Zhou, L. (2022) Intelligent Pedestrian Intention Prediction Frame- work. IEEE Sensors Letters, 6(2), 1–5. 34

    2022

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.