Pith. sign in

REVIEW 3 major objections 5 minor 32 references

TSLFormer: A Lightweight Transformer Model for Turkish Sign Language Recognition Using Skeletal Landmarks

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Hand-joint landmarks alone hit 90.67% on Turkish signs

desk verdict A modest but honest engineering result on lightweight Turkish sign language recognition, with the main caveat that the test split may not be signer-independent. read the letter →

arxiv 2505.07890 v4 pith:5EMQNNEV submitted 2025-05-11 cs.CL eess.IV

Kutay Ertürk , Furkan Alt{i}n{i}ş{i}k , İrem Sar{i}alt{i}n , Ömer Nezih Gerek This is my paper · ORCID
classification cs.CLeess.IV
keywords TurkishSignLanguagerecognitiontransformerskeletallandmarksMediapipeAUTSLdatasetself-attentionlightweightmodelgestureclassification
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

TSLFormer is a lightweight transformer that recognizes isolated Turkish Sign Language words using only 3D skeletal joint landmarks extracted by Mediapipe, rather than raw RGB or depth video. Trained and tested on the AUTSL dataset's 226-word, 43-signer collection of over 36,000 videos, the model reports 90.67% test accuracy with a 4-fold cross-validated average of 92.85%. The paper argues this demonstrates that joint-based features alone are sufficient for word-level sign recognition, cutting per-frame input from hundreds of thousands of pixels to 144 coordinates. That matters because it makes real-time, mobile, and embedded assistive communication systems for hearing-impaired users computationally feasible.

What carries the argument

The central mechanism is the transformer encoder's multi-head self-attention over the frame sequence, which lets every frame attend to every other frame and thereby captures the temporal co-occurrence of joint positions as a sign unfolds. Each frame is a 144-dimensional vector of XYZ coordinates from Mediapipe's Pose and Hands models; an embedding layer projects it to 512 dimensions, positional encoding adds order information, two encoder layers with layer normalization and dropout process the sequence, mean pooling summarizes it, and a linear layer outputs one of 226 word classes. The pipeline treats the joint-coordinate sequence as a sentence-like linguistic unit, with an EOS token marking the end of the gesture.

What would settle it

Hold out all videos of a specific subset of AUTSL signers during training and test only on those signers; if TSLFormer's accuracy falls well below 90.67%—for instance toward the roughly 62% signer-independent level reported for AUTSL—the claim that joint-only features make practical sign recognition feasible for unseen users would be contradicted.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that a two-layer transformer encoder with 512 hidden units and four attention heads, fed 16 evenly sampled frames of 48 upper-body and hand landmarks (144 XYZ features per frame) with zero padding and an EOS token, classifies 226 Turkish sign words with 90.67% accuracy and 90.67% F1. The central claim is that this accuracy shows sign language recognition can be carried out successfully using only joint-based features, making the efficiency/accuracy trade-off attractive for real-time and low-resource settings. The paper positions TSLFormer as a simpler alternative to multi-modal systems that exceed 98% accuracy by combining RGB, depth, skeleton, and face cues, while needing far less computation.

Load-bearing premise

The reported 90.67% assumes the 80/20 split tests generalization to signers the model has never seen, but the paper does not state whether signers appear in both training and test sets.

Editorial extensions

If this is right

  • Per-frame input drops from about 786,432 RGB pixel values to 144 landmark coordinates, making inference cheap enough for real-time webcam-based recognition.
  • The near-diagonal confusion matrix shows most misclassifications occur between visually similar signs, indicating the joint representation retains discriminative motion cues.
  • Because the model needs only Mediapipe landmarks, a working demo can run on a laptop webcam with motion-triggered recording and automatic segmentation.
  • The reported efficiency/accuracy balance positions TSLFormer as a candidate for embedded and mobile assistive devices, trading a few points of accuracy against much lower computational demand.
  • The paper's stated future work includes sentence-level recognition and evaluation across signer groups and datasets, which would test how well the approach transfers to new signers.

Reading between the lines

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

  • A direct signer-independent split of AUTSL would reveal how much of the reported 90.67% depends on signer identity; if accuracy drops toward the roughly 62% signer-independent baseline reported for AUTSL, the real-world generalization claim would need tempering.
  • The same landmark-plus-transformer recipe could plausibly transfer to other sign languages with comparable vocabulary sizes, since the method is dataset-agnostic, but that transfer remains untested.
  • Adding lightweight hand-shape or finger-spread descriptors to the 144-dimensional input might recover some of the finger-articulation detail the paper identifies as lost, potentially closing the gap to multi-modal models without full RGB processing.
  • The paper's observation that the model is confidently wrong on some errors suggests that a calibration or uncertainty-estimation layer would make the demo more trustworthy for assistive use.
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 paper introduces TSLFormer, a lightweight transformer-based model for isolated Turkish Sign Language (TİD) word recognition, using only 3D skeletal landmarks extracted with Mediapipe (hand and upper-body joints) instead of raw RGB or depth video. On the AUTSL dataset, the authors report a 4-fold cross-validation accuracy of 92.85% and a final held-out test accuracy of 90.67%, and compare this with prior AUTSL results, arguing that the landmark-only input offers a favorable accuracy/efficiency trade-off for real-time and mobile assistive systems. The manuscript also describes a real-time webcam demonstration application.

Significance. If the 90.67% accuracy genuinely reflects signer-independent generalization, the paper would provide a useful lightweight baseline for isolated sign recognition, and its explicit comparison of landmark-only input against multi-modal RGB/skeleton/face systems is a valuable design point. The paper is transparent about its preprocessing pipeline, reports cross-validation results, includes a confusion matrix, and acknowledges the limitations of skeletal landmarks (e.g., difficulty with fine finger articulation). However, the central empirical claim is currently not supported because the evaluation protocol—-specifically, whether the test split is signer-disjoint—-is not specified, and the comparison in Table III appears to mix evaluation protocols. The reproducibility-oriented choices (public dataset, standard architecture, documented hyperparameters) are strengths that would allow the evaluation issue to be corrected in a revision.

major comments (3)
  1. [III.C (two-stage validation)] The description of the 80/20 train/test split states only that both subsets have a similar class distribution; it never states that the split is signer-independent. Since AUTSL contains multiple videos per signer (43 signers, over 36,000 videos), a random class-stratified split will almost certainly place videos from the same signer in both training and test sets, allowing the model to exploit signer-specific motion and appearance patterns. The paper itself notes in Section III.A that the AUTSL baseline drops from about 95.95% to about 62% when tested on users not in the training set, which demonstrates how large this effect can be. Therefore, the central claim in Section VI that 90.67% test accuracy shows successful sign recognition using joint-based features is not yet supported. The authors should re-run the evaluation with a signer-disjoint split (e.g., following the ChaLearn/AUTSL protocol referenced in [26]) and report the signer-independent accuracy.
  2. [Table III (comparison with existing work)] The comparison in Table III mixes incompatible evaluation protocols. The cited methods SAM-SLR [23], S3D [24], TD-SL [25], USTC-SLR [31], Jalba [30], and VLE-Transformer [27] were evaluated under the ChaLearn 2021 signer-independent challenge protocol, whereas the proposed TSLFormer result is, based on the description in Section III.C, obtained on a non-signer-disjoint random split. Under these conditions, the table does not support the statement that TSLFormer achieves 'competitive performance' or that it is 'a valid and effective alternative' to those methods. The authors should either evaluate TSLFormer under the same signer-independent protocol or clearly label the protocols and restrict the comparison to methods evaluated under the same conditions.
  3. [IV.A (final test results)] The final test performance is reported as a single point estimate (90.67% accuracy, recall, F1) with no variance, no per-fold test results, no confidence intervals, and no class-wise statistics. Because the test set is a single random split, the reported numbers could vary substantially with the split, and the equality of accuracy, recall, and F1 suggests that the averaging scheme is not explained. The authors should report the standard deviation across multiple seeds or splits, and clarify how recall and F1 are aggregated (micro vs. macro) given the 226-class setting.
minor comments (5)
  1. [Abstract and I (dataset statistics)] The abstract states '227 different words' while the main text (Sections I and III.A) consistently says '226 different words'; this inconsistency should be corrected.
  2. [I and V (temporal sampling)] The introduction says '16 evenly sampled frames' and Section V says '16 anchor frames,' but Section III.C says frames were sampled at '15 frames per second.' These descriptions are inconsistent regarding the number vs. rate of sampled frames; the authors should clarify the actual temporal sampling procedure.
  3. [Throughout (terminology)] The paper alternates between 'TSL' and 'TİD' for Turkish Sign Language, and between 'Turkish Sign Language' and 'Turkish Sign Language (TİD)'; a single consistent abbreviation should be used.
  4. [IV.B (demo application)] The real-time demo is described qualitatively without quantitative latency or accuracy measurements on live input; adding even basic metrics would strengthen the claim of suitability for real-time use.
  5. [III.C (model description)] The statement 'Since there are no loops or recurrent structures in RNNs or LSTMs' is imprecise: RNNs and LSTMs do have recurrent loops. The sentence appears to mean that transformers have no recurrent structure, and should be reworded.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the reported accuracy is an empirical holdout result, and the main residual concern (signer identity in the split) is a generalization-risk issue, not a circularity.

full rationale

TSLFormer is an empirical machine-learning study rather than a derivation chain. Mediapipe landmarks are produced by an external library, a transformer classifier is trained with cross-entropy on an 80% training split, and the reported 90.67% accuracy is measured on a separate 20% test set, with additional 4-fold cross-validation on the training data. No equation or construction step makes the predicted quantity equal to a fitted input: the class distribution is stratified, but the accuracy itself is not defined by any training-set statistic, and no parameter fitted on the test set is later relabeled as a prediction. The paper's comparisons in Table III are external baselines, not evidence that TSLFormer's result is forced by self-citation or by definition. The paper's own caveats, such as the landmark-resolution trade-off in Section III.B and the miscalibration observation in Section IV.B, are empirical limitations rather than definitional reductions. The most serious concern is that the 80/20 split is described only as having 'a similar distribution of classes' and signer identity is never stated as a split criterion, so the 90.67% figure may partly reflect familiar signer identity instead of signer-independent generalization; the paper even notes that the AUTSL baseline drops to around 62% for users not in the training set. That concern, however, is an evaluation-protocol and external-validity risk, not circularity: the accuracy is not equivalent by construction to the training labels or to a self-citation. No circular step meets the required evidence bar of quoting a specific reduction, so the honest finding is no significant circularity (score 0).

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The paper does not introduce new physical entities or theoretical constructs. Its contribution is an empirical model configuration, so the ledger primarily contains hand-chosen hyperparameters and domain assumptions about the dataset and evaluation protocol. The most consequential assumption is the representativeness of the train/test split, which is not justified.

free parameters (6)
  • number of sampled frames per video = inconsistent: 16 in introduction, 15 per second in method
    The paper reports different frame sampling approaches, and the chosen number affects input sequence length and model performance. This is a hand-chosen design parameter without systematic comparison.
  • embedding dimension = 512
    The input features are projected to 512 dimensions, a standard transformer hyperparameter selected by hand.
  • number of encoder layers = 2
    The architecture uses two encoder layers, chosen for a lightweight model without reported ablation.
  • number of attention heads = 4
    Four attention heads are used, another hand-chosen architecture detail.
  • dropout rate = 0.2
    Dropout of 0.2 is set for regularization, selected without reported tuning.
  • learning rate = 1e-4 with scheduler
    Adam optimizer with this learning rate and a scheduler is used, a typical training choice.
assumptions (3)
  • domain assumption Mediapipe landmark coordinates are a sufficiently rich representation of the 226 TSL words to support the reported accuracy.
    The model input contains no RGB or depth information; if the landmark set misses finger-level distinctions or fine hand motions, the classification accuracy would degrade. The paper itself acknowledges that minor finger articulation and wrist rotation may be lost, so the sufficiency of the landmark representation is a load-bearing premise.
  • domain assumption The 80/20 train/test split reflects deployment conditions, including generalization to new signers.
    The paper does not state that the test split is signer-independent. If it is not, the reported 90.67% accuracy overestimates real-world performance on unseen signers, which is the central deployment scenario claimed by the authors.
  • domain assumption AUTSL dataset labels and video contents are correct and complete.
    The results depend entirely on the correctness of the AUTSL annotations and the integrity of the over 36,000 video samples. Any systematic label noise or video corruption would directly affect the measured accuracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TSLFormer: A Lightweight Transformer Model for Turkish Sign Language Recognition Using Skeletal Landmarks." pith.science (2026). https://pith.science/paper/5EMQNNEV

@misc{pith2026250507890,
  author       = {Pith},
  title        = {Pith review of: TSLFormer: A Lightweight Transformer Model for Turkish Sign Language Recognition Using Skeletal Landmarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5EMQNNEV}},
  note         = {Machine review of arXiv:2505.07890}
}
read the original abstract

This study presents TSLFormer, a light and robust word-level Turkish Sign Language (TSL) recognition model that treats sign gestures as ordered, string-like language. Instead of using raw RGB or depth videos, our method only works with 3D joint positions - articulation points - extracted using Google's Mediapipe library, which focuses on the hand and torso skeletal locations. This creates efficient input dimensionality reduction while preserving important semantic gesture information. Our approach revisits sign language recognition as sequence-to-sequence translation, inspired by the linguistic nature of sign languages and the success of transformers in natural language processing. Since TSLFormer uses the self-attention mechanism, it effectively captures temporal co-occurrence within gesture sequences and highlights meaningful motion patterns as words unfold. Evaluated on the AUTSL dataset with over 36,000 samples and 227 different words, TSLFormer achieves competitive performance with minimal computational cost. These results show that joint-based input is sufficient for enabling real-time, mobile, and assistive communication systems for hearing-impaired individuals.

Figures

Figures reproduced from arXiv: 2505.07890 by the authors.

Figure 1
Figure 1. Some Sampled Frames from the AUTSL Dataset [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overall scheme of our proposed TSLFormer-based sign language recognition system. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Diagram of Mediapipe Feature Extraction C. Sign Language Transformer Transformers are distinct from the traditional convolu￾tional neural networks (CNNs) and recurrent neural networks (RNNs) since they allow every item in a sequence to attend to every other item directly through self-attention mechanisms [32], [33]. While CNNs are efficient at learning local spatial patterns and RNNs are efficient at modeling short-… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Confusion Matrix for 226 Classes and TSLFormer [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 4
Figure 4. Figure 4: Diagram of TSLFormer After this, we evaluated the model on a separate 20% test set, which was never seen during training. The final test results were also strong: TABLE II: Final Test Results Metric Score (%) Accuracy 90.67% Recall 90.67% F1 Score 90.67% The results sh…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 12 canonical work pages

  1. [26]

    Ozge Mercanoglu Sincan and Julio C. S. Jacques Junior and Sergio Escalera and Hacer Yalim Keles, ”ChaLearn LAP Large Scale Signer Independent Isolated Sign Language Recognition Challenge: Design, Results and Future Research,” arXiv preprint arXiv:2105.05066, 2021. [Online]. Available: https://arxiv.org/abs/2105.05066

    work page Pith review arXiv 2021

  2. [23]

    [Online]

    Songyao Jiang and Bin Sun and Lichen Wang and Yue Bai and Kunpeng Li and Yun Fu, ”Skeleton Aware Multi-modal Sign Language Recog- nition,” arXiv preprint arXiv:2103.08833, 2021. [Online]. Available: https://arxiv.org/abs/2103.08833

    arXiv 2021

  3. [25]

    2555-2559

    Sarhan, Noha and Wilms, Christian and Closius, Vanessa and Brefeld, Ulf and Frintrop, Simone, ”Hands in Focus: Sign Lan- guage Recognition Via Top-Down Attention,” in2023 IEEE Interna- tional Conference on Image Processing (ICIP), 2023 pp. 2555-2559. doi:10.1109/ICIP49359.2023.10222729

    arXiv 2023

  4. [31]

    [Online]

    Zhou, Hao, ”USTC-SLR Submission to ChaLearn 2021 LAP Challenge,” 2021. [Online]. Available: https://github.com/ustc-slr/ ChaLearn-2021-ISLR-Challenge Accessed: March 2025

    work page 2021

  5. [30]

    [Online]

    Alba, Juan and Docio, Laura and Banga, Eva Rueda and Vazquez, Marta and Torres, Sergio and Perez, Antonio, ”Fact Sheet: CVPR 2021 ChaLearn Looking at People Large Scale Signer Independent Isolated SLR Challenge,” 2021. [Online]. Available: https://tinyurl.com/ jalba2021fact Accessed: March 2025

    work page 2021

  6. [1]

    Unutmaz, Berkan and Karaca, Ali Can and G ¨ull¨u, M. Kemal, ”Turkish Sign Language Recognition Using Kinect Skeleton and Convolutional Neural Network,” in2019 27th Signal Processing and Communications Applications Conference (SIU), 2019 pp. 1-4. doi:10.1109/SIU.2019.8806380

    work page arXiv 2019

  7. [2]

    Sincan, Ozge Mercanoglu and Keles, Hacer Yalim, ”AUTSL: A Large Scale Multi-Modal Turkish Sign Language Dataset and Base- line Methods,”IEEE Access, vol. 8, pp. 181340–181355, 2020. doi:10.1109/access.2020.3028072

    arXiv 2020

  8. [3]

    [Online]

    Camillo Lugaresi and Jiuqiang Tang and Hadon Nash and Chris Mc- Clanahan and Esha Uboweja and Michael Hays and Fan Zhang and Chuo-Ling Chang and Ming Guang Yong and Juhyun Lee and Wan- Teh Chang and Wei Hua and Manfred Georg and Matthias Grundmann, ”MediaPipe: A Framework for Building Perception Pipelines,” arXiv preprint arXiv:1906.08172, 2019. [Online]....

    arXiv 1906

Show all 32 references
  1. [4]

    [Online]

    Anurag Arnab and Mostafa Dehghani and Georg Heigold and Chen Sun and Mario Lu ˇci´c and Cordelia Schmid, ”ViViT: A Video Vision Transformer,” arXiv preprint arXiv:2103.15691, 2021. [Online]. Avail- able: https://arxiv.org/abs/2103.15691

    2021 arXiv

  2. [5]

    and Juang, B., ”An introduction to hidden Markov models,”IEEE ASSP Magazine, vol

    Rabiner, L. and Juang, B., ”An introduction to hidden Markov models,”IEEE ASSP Magazine, vol. 3, no. 1, pp. 4-16, 1986. doi:10.1109/MASSP.1986.1165342

    1986

  3. [6]

    Zhang, Jihai and Zhou, Wengang and Xie, Chao and Pu, Junfu and Li, Houqiang, ”Chinese sign language recognition with adaptive HMM,” in 2016 IEEE International Conference on Multimedia and Expo (ICME), 2016 pp. 1-6. doi:10.1109/ICME.2016.7552950

    2016

  4. [7]

    Starner, Thad and Group, Massachusetts, ”Visual Recognition of Amer- ican Sign Language Using Hidden Markov Models,” 1995

    1995

  5. [8]

    and Bottou, L

    Lecun, Y . and Bottou, L. and Bengio, Y . and Haffner, P., ”Gradient- based learning applied to document recognition,”Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998. doi:10.1109/5.726791

    1998 doi

  6. [9]

    [Online]

    Karen Simonyan and Andrew Zisserman, ”Very Deep Convolu- tional Networks for Large-Scale Image Recognition,” arXiv preprint arXiv:1409.1556, 2015. [Online]. Available: https://arxiv.org/abs/1409. 1556

    2015 arXiv

  7. [10]

    [Online]

    Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun, ”Deep Residual Learning for Image Recognition,” arXiv preprint arXiv:1512.03385, 2015. [Online]. Available: https://arxiv.org/abs/1512. 03385

    2015 arXiv

  8. [11]

    doi:10.1007/978-3-319-16178-5 40

    Pigou, Lionel and Dieleman, Sander and Kindermans, Pieter-Jan and Schrauwen, Benjamin, ”Sign Language Recognition Using Convolu- tional Neural Networks,” 2015. doi:10.1007/978-3-319-16178-5 40

    2015 doi

  9. [12]

    [Online]

    Du Tran and Lubomir Bourdev and Rob Fergus and Lorenzo Torresani and Manohar Paluri, ”Learning Spatiotemporal Features with 3D Con- volutional Networks,” arXiv preprint arXiv:1412.0767, 2015. [Online]. Available: https://arxiv.org/abs/1412.0767

    2015 arXiv

  10. [13]

    43, 2012

    Rautaray, Siddharth and Agrawal, Anupam, ”Vision based Hand Gesture Recognition for Human Computer Interaction: A Survey,”Artificial Intelligence Review, vol. 43, 2012. doi:10.1007/s10462-012-9356-9

    2012 doi

  11. [14]

    Camgoz, Necati and Hadfield, Simon and Koller, Oscar and Ney, Hermann and Bowden, Richard, ”Neural Sign Language Translation,”

  12. [15]

    [Online]

    Ali Ghadami and Alireza Taheri and Ali Meghdari, ”A Transformer- Based Multi-Stream Approach for Isolated Iranian Sign Language Recognition,” arXiv preprint arXiv:2407.09544, 2024. [Online]. Avail- able: https://arxiv.org/abs/2407.09544

    2024 arXiv

  13. [16]

    3154-3160

    Hara, Kensho and Kataoka, Hirokatsu and Satoh, Yutaka, ”Learning Spatio-Temporal Features with 3D Residual Networks for Action Recognition,” in2017 IEEE International Conference on Computer Vision Workshops (ICCVW), 2017 pp. 3154-3160. doi:10.1109/ICCVW.2017.373

    2017 doi

  14. [17]

    1250-1253

    Buckley, Neil and Sherrett, Lewis and Lindo Secco, Emanuele, ”A CNN sign language recognition system with single & double- handed gestures,” in2021 IEEE 45th Annual Computers, Software, and Applications Conference (COMPSAC), 2021 pp. 1250-1253. doi:10.1109/COMPSAC51774.2021.00173

    2021

  15. [18]

    Gomez and Lukasz Kaiser and Illia Polosukhin, ”Attention Is All You Need,” arXiv preprint arXiv:1706.03762, 2023

    Ashish Vaswani and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N. Gomez and Lukasz Kaiser and Illia Polosukhin, ”Attention Is All You Need,” arXiv preprint arXiv:1706.03762, 2023. [Online]. Available: https://arxiv.org/abs/1706. 03762

    2023 arXiv

  16. [19]

    Hinton, ”Layer Normalization,” arXiv preprint arXiv:1607.06450, 2016

    Jimmy Lei Ba and Jamie Ryan Kiros and Geoffrey E. Hinton, ”Layer Normalization,” arXiv preprint arXiv:1607.06450, 2016. [Online]. Avail- able: https://arxiv.org/abs/1607.06450

    2016 arXiv

  17. [20]

    Nitish Srivastava and Geoffrey Hinton and Alex Krizhevsky and Ilya Sutskever and Ruslan Salakhutdinov, ”Dropout: A Simple Way to Pre- vent Neural Networks from Overfitting,”Journal of Machine Learning Research, vol. 15, no. 56, pp. 1929–1958, 2014. [Online]. Available: http://...

    1929

  18. [21]

    Kingma and Jimmy Ba, ”Adam: A Method for Stochastic Optimization,” arXiv preprint arXiv:1412.6980, 2017

    Diederik P. Kingma and Jimmy Ba, ”Adam: A Method for Stochastic Optimization,” arXiv preprint arXiv:1412.6980, 2017. [Online]. Avail- able: https://arxiv.org/abs/1412.6980

    2017 arXiv

  19. [22]

    Kohavi, Ron, ”A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection,” 2001

    2001

  20. [28]

    Mercanoglu Sincan, Ozge and Keles, Hacer Yalim, ”Using Motion History Images With 3D Convolutional Networks in Isolated Sign Language Recognition,”IEEE Access, vol. 10, pp. 18608–18618, 2022. doi:10.1109/access.2022.3151362

    2022

  21. [29]

    3436-3445

    De Coster, Mathieu and Van Herreweghe, Mieke and Dambre, Joni, ”Isolated Sign Recognition from RGB Video using Pose Flow and Self-Attention,” in2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2021 pp. 3436-3445. doi:10.1109/CVPRW53098.2021.00383

    2021

  22. [32]

    [Online]

    Keiron O’Shea and Ryan Nash, ”An Introduction to Convolutional Neural Networks,” arXiv preprint arXiv:1511.08458, 2015. [Online]. Available: https://arxiv.org/abs/1511.08458

    2015 arXiv

  23. [33]

    Rumelhart and Geoffrey E

    David E. Rumelhart and Geoffrey E. Hinton and Ronald J. Williams, ”Learning internal representations by error propagation,” 1986. [Online]. Available: https://api.semanticscholar.org/CorpusID:62245742

    1986

  24. [2018]

    doi:10.1109/CVPR.2018.00812

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.