Pith. sign in

REVIEW 3 major objections 5 minor 65 references

Bringing Balance to Hand Shape Classification: Mitigating Data Imbalance Through Generative Models

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Pre-training on balanced GAN-generated hand images lifts RWTH hand-shape accuracy to 85.3%, a reported 5% gain over the prior best result.

desk verdict The GAN-pretraining recipe for imbalanced handshape data is credible and useful, but the SOTA claim is overreach: 85.3 vs 84.3 is a one-point gap, not the '5%' advertised, and the evaluation protocol is underspecified. read the letter →

arxiv 2507.17008 v1 pith:BZZ6FZGS submitted 2025-07-22 cs.CV cs.AI

Gaston Gustavo Rios , Pedro Dal Bianco , Franco Ronchetti , Facundo Quiroga , Oscar Stanchi , Santiago Ponte Ahón , Waldo Hasperué This is my paper
classification cs.CVcs.AI
keywords handshaperecognitionsignlanguagedataimbalancegenerativeadversarialnetworkssyntheticpretrainingRWTHdatasetHaGRIDEfficientNet
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

Most sign-language handshape datasets are small, heavily skewed toward a few frequent shapes, and expensive to expand by collecting real samples. The paper proposes to fix this with synthetic data: train a class- or pose-conditioned GAN on the available images, generate a balanced set of 1000 images per handshape, pre-train an EfficientNet-v2 classifier on that synthetic set, and fine-tune it on the real data. On the RWTH German Sign Language handshape dataset this reaches 85.3% accuracy, which the paper reports as a 5% improvement over the prior state of the art, and per-class accuracy for some minority shapes rises from zero to 100%. The same recipe with a pose-conditioned generator trained on a different large hand-gesture dataset transfers to RWTH without retraining the generator, reaching 85.15%. If correct, this gives low-resource sign languages a route to usable classifiers without collecting thousands of new labeled images.

What carries the argument

The load-bearing mechanism is a two-stage training strategy: pre-train on a balanced synthetic dataset, then fine-tune on the real data. The synthetic dataset is produced by one of two GAN conditioning schemes: ReACGAN, which conditions each generated image on its class label through an auxiliary classifier and a data-to-data cross-entropy loss on a projected hypersphere, or SPADE, which conditions on 21 hand keypoints rendered as per-channel heat maps or bone lines through spatially-adaptive normalization. Pre-training on the balanced 1000-images-per-class corpus, rather than mixing it with real data or using it as a loss regularizer, is the variant that consistently gives the best results.

What would settle it

Run the new model and the previous 84.3%-accuracy model on the same held-out RWTH test images after verifying that none of the one million weakly labeled frames used for generator training appears in that test set; if the test protocols differ or overlap, the reported 5% superiority is not established.

Watch

Extended reading notes

Core claim

The paper's central claim is that the gain comes from synthetic pre-training, not from extra real data, additional data augmentation, or class-rebalancing losses. Using ReACGAN, a label-conditioned GAN with a data-to-data cross-entropy discriminator loss, to generate a balanced 1000-images-per-class RWTH pre-training set, an EfficientNet-v2 M classifier pre-trained on that set and then fine-tuned on the real training images scores 85.3% on the held-out test set, against 80.62% for the same classifier trained from ImageNet initialization and 84.3% for the strongest previously published model. Per-class results show minority classes that the real-data baseline never predicted correctly reaching up to 100% accuracy, and the pre-trained models converge in roughly half the epochs. A pose-conditioned SPADE generator trained on HaGRID and applied to RWTH poses gives 85.15%, supporting the claim that pose-based generation makes the approach reusable across sign languages.

Load-bearing premise

The load-bearing premise is that the held-out test set used here is the same test protocol used by the earlier published results, and that the one million weakly labeled RWTH frames used to train the generator do not overlap the test images.

Editorial extensions

If this is right

  • Pre-training on balanced GAN-generated data raises per-class accuracy from 1.89% to 37.74% when only five RWTH samples per class are available.
  • Synthetic pre-training converges in roughly half the epochs needed from ImageNet, so the same accuracy can be reached with less training time.
  • A pose-conditioned generator trained on HaGRID and applied to RWTH poses reaches 85.15% without retraining, so the generator can be reused for a new target dataset.
  • Filtering generated samples to the top 30% by class-conditional confidence does not improve accuracy, so the full balanced synthetic set is preferable to a smaller filtered one.
  • The largest gains appear exactly where real data are scarcest, so the method is aimed at small and unbalanced handshape collections rather than large already-balanced datasets.

Reading between the lines

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

  • Editorial inference: if the benefit comes mainly from balanced initialization rather than from the photorealism of individual images, cheaper procedural or template-based hand rendering might capture part of the gain; the paper tests only GANs.
  • Editorial inference: the tables show 80.62% for the paper's own real-data baseline and 84.3% for the strongest prior published number, so the abstract's '5% improvement' is not uniquely pinned to either comparison; a same-protocol head-to-head would settle which comparison is meant.
  • Editorial inference: the multi-source result suggests a one-time investment in a large hand-pose generator could bootstrap classifiers for many low-resource sign languages once poses are extracted, but the paper demonstrates this only on RWTH.
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 / 5 minor

Summary. The paper proposes using GAN-generated synthetic images to mitigate data scarcity and class imbalance in static handshape classification. It compares ReACGAN (label-conditioned) and SPADE (pose-conditioned) generators on the RWTH and HaGRID datasets, and evaluates four ways of combining generated and real data: pre-training then fine-tuning, regularized joint training, mixup, and filtered top-K generation. The central empirical claim is that pre-training EfficientNet-v2 on a balanced set of 39,000 ReACGAN-generated RWTH images and fine-tuning on real data yields 85.3% test accuracy, which the authors describe as a 5% improvement over the state of the art. They also report faster convergence, improved minority-class accuracy, and a multi-source variant where a HaGRID-trained SPADE generator is applied to RWTH poses.

Significance. If the reported result is robust, the paper makes a useful empirical contribution: balanced synthetic pre-training appears to help more than standard ImageNet pre-training in a small imbalanced dataset, and the effect is largest in reduced-data regimes (Tables 8 and 9). The strengths are the systematic within-paper comparison of training strategies, the per-class accuracy analysis in Figure 7, the inclusion of multi-source transfer, and public code. However, the headline SOTA claim is not yet supported because the evaluation protocol is underspecified and single-run.

major comments (3)
  1. [§4.2 and Table 6] The paper does not define the held-out test split (number of images per class, sampling seed, or split rule), and every accuracy in Tables 5–9 is a single run without standard deviations or repeated-seed statistics. Since the headline claim is that 85.3% surpasses the previous best of 84.3% (Inception-ResNet-v2, [40]) by a 1.0-point margin, the absence of a shared protocol and variance estimate does not establish the claimed SOTA superiority. Please provide the split definition, per-class test counts, and results over multiple seeds; if the prior methods used a different evaluation protocol, the comparison should be framed accordingly.
  2. [Abstract, §5.2.2, and Table 6] The abstract's '5% improvement over the current state-of-the-art' is not supported by Table 6. The gain over the previous best (84.3%) is 1.0 percentage point; 85.3% is 4.7 percentage points above the authors' own ImageNet-pretrained baseline (80.6%). Please report the exact comparison basis in the abstract and conclusions, and avoid implying a 5-point gain over prior work.
  3. [§4.1 and §5.2] The generator is trained on 'the 1 million weakly labeled images' from RWTH (Section 4.1), while the classifier experiments appear to use the 2,098 pose-filtered labeled samples as the real training set (Section 5.2.1 states that 39,000 generated images are 19 times the regular data size). Section 5.2 then says 'We used the same training samples that were used to train the generator to train the classifier,' which is ambiguous at best and contradictory at worst. The manuscript must state exactly which images were used for generator training, classifier real-data training, validation, and testing, and demonstrate that the 1 million weakly labeled frames do not overlap the held-out test set; otherwise the 85.3% result may reflect test-set information or a different training distribution than prior work.
minor comments (5)
  1. [§4.2] The sentence 'We separated some of the samples of each dataset to use as our held-out test set' is too vague; please specify the exact split ratio, the seed, and how class imbalance was handled in the split.
  2. [§3.1.4 and Table 5] The filtering method is described as selecting the top-K samples, but the table caption says the filtered datasets contain the top 30%; please reconcile the notation and report the actual K or percentage used.
  3. [§3.1.2 and Table 2] Equations (1)–(3) depend on hyperparameters α and β, but their values are never reported; please provide the schedule settings and, if available, a sensitivity analysis.
  4. [Figure 7] The caption mentions green and red on the x-axis while the bars are described as blue and red; please clarify the color coding and include per-class test-set counts so the '100% accuracy' claims can be interpreted.
  5. [§5.1 and throughout] There are typographical and rendering issues, including the heading 'Hanshape generation' and inconsistent spacing in 'R WTH'; a careful proofread would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the GAN-pretraining result is an empirical comparison, not a consequence of the paper's definitions or self-citation.

full rationale

The paper's central claim is empirical: pre-training an EfficientNet-v2 classifier on balanced GAN-generated RWTH data and fine-tuning on real data yields 85.3% accuracy (Tables 5 and 6). No equation in the paper defines the reported accuracy in terms of the fitted generator or a self-cited result; the improvement is measured against an ImageNet-pretrained baseline and against published numbers from other groups. The only mechanism resembling self-reference is the filtering step in Section 3.1.4, where a classifier trained on real data scores generated samples and the top-K are kept, but this is sample selection rather than a derivation that assumes its conclusion, and the headline 85.34% result in Table 5 is achieved without filtering (filtered pretraining is 84.38%), so it is not load-bearing. Self-citations [12], [13], [17], and [39] are background references or prior baselines and do not supply the main result. The ambiguous held-out split and single-run accuracies described in Sections 4.2 and 5.2 are evaluation-transparency concerns, not circularity: they do not make the result true by construction. Accordingly, no circular step is identified.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central results rest mainly on empirical training choices rather than derivation. The key hand-tuned quantities are the sigma schedule parameters, the ReACGAN conditional loss weight, the generated dataset size, the filter threshold, and the pose rendering covariance. The main domain assumptions are pose transfer across datasets, the benign use of 1M weakly labeled frames, and the validity of classifier-based filtering. No new physical or conceptual entities are introduced.

free parameters (5)
  • sigma schedule parameters alpha and beta for regularization/mixup = not reported (described as fine-tuned)
    Equations 2 and 3 define sigma increasing and decreasing with alpha and beta, but the paper never states the chosen values. These parameters control how much generated data affects training in the regularization and mixup methods.
  • ReACGAN conditional loss weight lambda_cond = 1.0 (0.5 also tested)
    Table 3 reports results for lambda_cond = 0.5 and 1.0; the authors chose a high conditional loss to improve label fidelity, which directly affects generated image quality and, downstream, classifier accuracy.
  • Balanced generated dataset size per class = 1000 images per class (39,000 for RWTH)
    Section 5.2.1 states a balanced dataset of 1000 images per class was used and that more than 1000 gave no major improvement. This is a hand-chosen threshold that defines the pretraining set size.
  • Filter top-K threshold = top 30% of class-conditional scores
    Table 5 caption says filtered datasets contain the top 30% of samples ranked by class-conditional probability. No sensitivity analysis is provided for this threshold.
  • Pose rendering covariance for keypoints = not reported ('small covariance')
    Section 4.2 describes a multivariate normal distribution with a small covariance matrix at each keypoint; the exact covariance value is not given, and it controls the visual conditioning signal for SPADE.
assumptions (4)
  • domain assumption Pose information transfers across hand datasets: a generator trained on HaGRID poses can generate recognizable RWTH handshapes when conditioned on RWTH poses.
    This is the load-bearing premise for the multi-source experiments in Section 3 and Table 7. If the pose-to-handshape mapping does not transfer, the 85.15% multi-source result would not follow.
  • domain assumption The 1 million weakly labeled RWTH frames can be used to train the generator without contaminating the classifier comparison.
    Section 4.1 states the 1M weak labels were used to train the generator. The paper assumes this is fair relative to prior SOTA methods in Table 6 and that no test-set leakage occurs, but neither is demonstrated.
  • domain assumption A classifier trained on real data can rank generated images by class-conditional probability, and the top-K selection does not bias the final classifier.
    Section 3.1.4 uses a real-data classifier to score generated images and keep the top 30%. The paper's own results show filtering gave 'no noticeable improvement', so the benefit of this assumption is not established.
  • standard math Standard GAN training practices (hinge loss, spectral normalization, data-to-data cross-entropy) work reliably for handshape image generation.
    Section 5.1 describes these choices as the backbone for all GAN models; they are standard in the literature and are not central to the paper's claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bringing Balance to Hand Shape Classification: Mitigating Data Imbalance Through Generative Models." pith.science (2026). https://pith.science/paper/BZZ6FZGS

@misc{pith2026250717008,
  author       = {Pith},
  title        = {Pith review of: Bringing Balance to Hand Shape Classification: Mitigating Data Imbalance Through Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BZZ6FZGS}},
  note         = {Machine review of arXiv:2507.17008}
}
read the original abstract

Most sign language handshape datasets are severely limited and unbalanced, posing significant challenges to effective model training. In this paper, we explore the effectiveness of augmenting the training data of a handshape classifier by generating synthetic data. We use an EfficientNet classifier trained on the RWTH German sign language handshape dataset, which is small and heavily unbalanced, applying different strategies to combine generated and real images. We compare two Generative Adversarial Networks (GAN) architectures for data generation: ReACGAN, which uses label information to condition the data generation process through an auxiliary classifier, and SPADE, which utilizes spatially-adaptive normalization to condition the generation on pose information. ReACGAN allows for the generation of realistic images that align with specific handshape labels, while SPADE focuses on generating images with accurate spatial handshape configurations. Our proposed techniques improve the current state-of-the-art accuracy on the RWTH dataset by 5%, addressing the limitations of small and unbalanced datasets. Additionally, our method demonstrates the capability to generalize across different sign language datasets by leveraging pose-based generation trained on the extensive HaGRID dataset. We achieve comparable performance to single-source trained classifiers without the need for retraining the generator.

Figures

Figures reproduced from arXiv: 2507.17008 by the authors.

Figure 1
Figure 1. Diagram (a) shows the regular training approach of our classifier model. For [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Count of training images belonging to the 39 hand shapes of RWTH. Each hand [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Visualization of the original image, its joints, and bones. Each keypoint con [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Diagram depicting the ReACGAN model. The generator takes as input a latent [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Diagram depicting the SPADE model. The generator takes as input a latent [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Real and generated samples of RWTH and HaGRID. Generated images were [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Per-class accuracy difference in RWTH between the ReACGAN-pre-trained [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: Plots showing the accuracy of the classifier model on the training dataset in each [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

65 extracted references · 57 canonical work pages

  1. [40]

    Hand shape recognition using very deep convolutional neural networks, in: Proceedings of the 1st Interna- tional Conference on Control and Computer Vision, p

    Rakowski, A., Wandzik, L., 2018. Hand shape recognition using very deep convolutional neural networks, in: Proceedings of the 1st Interna- tional Conference on Control and Computer Vision, p. 812

    work page 2018

  2. [1]

    Do gans actually learn the distribution? an empirical study

    Arora, S., Zhang, Y., 2017. Do gans actually learn the distribution? an empirical study. CoRR abs/1706.08224

    arXiv 2017

  3. [2]

    Neural sign actors: A diffusion model for 3d sign language production from text, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Baltatzis, V., Potamias, R.A., Ververas, E., Sun, G., Deng, J., Zafeiriou, S., 2024. Neural sign actors: A diffusion model for 3d sign language production from text, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1985–1995

    work page 2024

  4. [3]

    Au- tosimulate: (quickly) learning synthetic data generation, in: Computer Vision – ECCV 2020, pp

    Behl, H.S., Baydin, A.G., Gal, R., Torr, P.H.S., Vineet, V., 2020. Au- tosimulate: (quickly) learning synthetic data generation, in: Computer Vision – ECCV 2020, pp. 255–271

    work page 2020

  5. [4]

    Improving image generation with better captions

    Betker, J., Goh, G., Jing, L., TimBrooks, ., Wang, J., Li, L., Lon- gOuyang, ., JuntangZhuang, ., JoyceLee, ., YufeiGuo, ., Wesam- Manassra, ., PrafullaDhariwal, ., CaseyChu, ., YunxinJiao, ., Ramesh, A., 2023. Improving image generation with better captions

    work page 2023

  6. [5]

    Sampling strate- gies for GAN synthetic data, in: ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp

    Bhattarai, B., Baek, S., Bodur, R., Kim, T.K., 2020. Sampling strate- gies for GAN synthetic data, in: ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2303–2307. 24

    work page 2020

  7. [6]

    Vald´ es Hern´ andez, M., Wardlaw, J.M., Rueck- ert, D., 2018

    Bowles, C., Chen, L., Guerrero, R., Bentley, P., Gunn, R.N., Hammers, A., Dickie, D.A., del C. Vald´ es Hern´ andez, M., Wardlaw, J.M., Rueck- ert, D., 2018. GAN augmentation: Augmenting training data using generative adversarial networks. CoRR abs/1810.10863

    arXiv 2018

  8. [7]

    Bragg, D., Caselli, N., Hochgesang, J.A., Huenerfauth, M., Katz- Hernandez, L., Koller, O., Kushalnagar, R., Vogler, C., Ladner, R.E.,

    work page

Show all 65 references
  1. [8]

    Bragg, D., Koller, O., Bellard, M., Berke, L., Boudreault, P., Braf- fort, A., Caselli, N., Huenerfauth, M., Kacorri, H., Verhoef, T., et al.,

  2. [9]

    Subunets: End-to-end hand shape and continuous sign language recognition, in: 2017 IEEE International Conference on Computer Vision (ICCV), pp

    Camgoz, N.C., Hadfield, S., Koller, O., Bowden, R., 2017. Subunets: End-to-end hand shape and continuous sign language recognition, in: 2017 IEEE International Conference on Computer Vision (ICCV), pp. 3075–3084

    2017

  3. [10]

    A survey on generative diffusion model

    Cao, H., Tan, C., Gao, Z., Xu, Y., Chen, G., Heng, P.A., Li, S.Z., 2023. A survey on generative diffusion model. arXiv:2209.02646

    2023 arXiv

  4. [11]

    OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields

    Cao, Z., Hidalgo, G., Simon, T., Wei, S.E., Sheikh, Y., 2021. OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields . IEEE Transactions on Pattern Analysis & Machine Intelligence 43, 172–186

    2021

  5. [12]

    Recognizing handshapes using small datasets, in: XXV Congreso Argentino de Ciencias de la Computaci´ on (CACIC 2019, Universidad Nacional de Ro Cuarto)

    Cornejo Fandos, U.J., Rios, G.G., Ronchetti, F., Quiroga, F., Hasperu´ e, W., Lanzarini, L.C., 2019. Recognizing handshapes using small datasets, in: XXV Congreso Argentino de Ciencias de la Computaci´ on (CACIC 2019, Universidad Nacional de Ro Cuarto)

    2019

  6. [13]

    Lsa-t: The first continuous argen- tinian sign language dataset forsign language translation, in: Advances in Artificial Intelligence – IBERAMIA 2022, pp

    Dal Bianco, P., R ´ ıos, G., Ronchetti, F., Quiroga, F., Stanchi, O., Hasperu´ e, W., Rosete, A., 2022. Lsa-t: The first continuous argen- tinian sign language dataset forsign language translation, in: Advances in Artificial Intelligence – IBERAMIA 2022, pp. 293–304. 25

    2022

  7. [14]

    Diffusion models beat gans on image synthesis

    Dhariwal, P., Nichol, A., 2021. Diffusion models beat gans on image synthesis. Advances in neural information processing systems 34, 8780– 8794

    2021

  8. [15]

    Sign language fingerspelling recogni- tion using synthetic data, in: Irish Conference on Artificial Intelligence and Cognitive Science

    Fowley, F., Ventresque, A., 2021. Sign language fingerspelling recogni- tion using synthetic data, in: Irish Conference on Artificial Intelligence and Cognitive Science

    2021

  9. [16]

    Frid-Adar, M., Klang, E., Amitai, M., Goldberger, J., Greenspan, H.,

  10. [17]

    Redes GANs como t´ ecnica de data augmentation para el reconocimiento de lengua de senas

    Gaggiotti, W., 2021. Redes GANs como t´ ecnica de data augmentation para el reconocimiento de lengua de senas. Ph.D. thesis. Universidad Nacional de La Plata

    2021

  11. [18]

    Unifying semi-supervised and robust learning by mixup

    Hataya, R., Nakayama, H., 2019. Unifying semi-supervised and robust learning by mixup

    2019

  12. [19]

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., Hochreiter, S.,

  13. [20]

    Visual synthetic data generation for sign language recognition

    Ibrahim, A., Kashef, R., 2012. Visual synthetic data generation for sign language recognition

    2012

  14. [21]

    Synthetic data generation technique in signer-independent sign language recognition

    Jiang, F., Gao, W., Yao, H., Zhao, D., Chen, X., 2009. Synthetic data generation technique in signer-independent sign language recognition. Pattern Recognition Letters 30, 513–524

    2009

  15. [22]

    Rebooting ACGAN: aux- iliary classifier gans with stable training

    Kang, M., Shim, W., Cho, M., Park, J., 2021. Rebooting ACGAN: aux- iliary classifier gans with stable training. Advances in neural information processing systems 34, 23505–23518

    2021

  16. [23]

    Hagrid – hand gesture recognition image dataset, in: Proceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV), pp

    Kapitanov, A., Kvanchiani, K., Nagaev, A., Kraynov, R., Makhliarchuk, A., 2024. Hagrid – hand gesture recognition image dataset, in: Proceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV), pp. 4572–4581. 26

    2024

  17. [24]

    Improving American Sign Language recognition with synthetic data, in: Proceedings of Machine Translation Summit XVII: Research Track, pp

    Kim, J., O’Neill-Brown, P., 2019. Improving American Sign Language recognition with synthetic data, in: Proceedings of Machine Translation Summit XVII: Research Track, pp. 151–161

    2019

  18. [25]

    Quantitative survey of the state of the art in sign language recognition

    Koller, O., 2020. Quantitative survey of the state of the art in sign language recognition. CoRR abs/2008.09918

    2020 arXiv

  19. [26]

    Koller, O., Ney, H., Bowden, R., 2016. Deep hand: How to train a cnn on 1 million hand images when your data is continuous and weakly labelled, in: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3793–3802

    2016

  20. [27]

    Robust learning from untrusted sources, in: International conference on machine learning, pp

    Konstantinov, N., Lampert, C., 2019. Robust learning from untrusted sources, in: International conference on machine learning, pp. 3488– 3498

    2019

  21. [28]

    Training deep face recognition systems with synthetic data

    Kortylewski, A., Schneider, A., Gerig, T., Egger, B., Morel-Forster, A., Vetter, T., 2018. Training deep face recognition systems with synthetic data. CoRR abs/1802.05891

    2018 arXiv

  22. [29]

    Imagenet classifica- tion with deep convolutional neural networks, in: Advances in Neural Information Processing Systems

    Krizhevsky, A., Sutskever, I., Hinton, G.E., 2012. Imagenet classifica- tion with deep convolutional neural networks, in: Advances in Neural Information Processing Systems

    2012

  23. [30]

    Word-level deep sign lan- guage recognition from video: A new large-scale dataset and methods comparison, in: The IEEE Winter Conference on Applications of Com- puter Vision, pp

    Li, D., Rodriguez, C., Yu, X., Li, H., 2020. Word-level deep sign lan- guage recognition from video: A new large-scale dataset and methods comparison, in: The IEEE Winter Conference on Applications of Com- puter Vision, pp. 1459–1469

    2020

  24. [31]

    Moreno-Barea, F.J., Strazzera, F., Jerez, J.M., Urda, D., Franco, L.,

  25. [32]

    Generating synthetic data with variational autoencoder to address class imbalance of graph attention network prediction model for construction management

    Mostofi, F., Behzat Tokdemir, O., Toan, V., 2024. Generating synthetic data with variational autoencoder to address class imbalance of graph attention network prediction model for construction management. Ad- vanced Engineering Informatics 62, 102606

    2024

  26. [33]

    A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial net- works for medical image synthesis

    Mller-Franzes, G., Niehues, J.M., Khader, F., Arasteh, S.T., Haar- burger, C., Kuhl, C., Wang, T., Han, T., Nolte, T., Nebelung, S., 27 Kather, J.N., Truhn, D., 2023. A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial net- wor...

    2023

  27. [34]

    Reliable fidelity and diversity metrics for generative models, in: Proceedings of the 37th International Conference on Machine Learning

    Naeem, M.F., Oh, S.J., Uh, Y., Choi, Y., Yoo, J., 2020. Reliable fidelity and diversity metrics for generative models, in: Proceedings of the 37th International Conference on Machine Learning

    2020

  28. [35]

    Theoretical insights into memorization in gans, in: Neural Information Processing Systems Workshop, p

    Nagarajan, V., Raffel, C., Goodfellow, I.J., 2018. Theoretical insights into memorization in gans, in: Neural Information Processing Systems Workshop, p. 3

    2018

  29. [36]

    728– 734

    Forward noise adjustment scheme for data augmentation, in: 2018 IEEE Symposium Series on Computational Intelligence (SSCI), pp. 728– 734

    2018

  30. [37]

    Conditional Image Synthesis With Auxiliary Classifier GANs

    Odena, A., Olah, C., Shlens, J., 2016. Conditional Image Synthesis With Auxiliary Classifier GANs. arXiv e-prints , arXiv:1610.09585

    2016 arXiv

  31. [38]

    Semantic image synthesis with spatially-adaptive normalization, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Park, T., Liu, M.Y., Wang, T.C., Zhu, J.Y., 2019. Semantic image synthesis with spatially-adaptive normalization, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 2337–2346

    2019

  32. [39]

    Quiroga, F., Antonio, R., Ronchetti, F., Lanzarini, L.C., Rosete, A.,

  33. [41]

    A survey on sign language machine translation

    Nez-Marcos, A., de Viaspre, O.P., Labaka, G., 2023. A survey on sign language machine translation. Expert Systems with Applications 213, 118993

    2023

  34. [42]

    A broad review on class imbalance learning techniques

    Rezvani, S., Wang, X., 2023. A broad review on class imbalance learning techniques. Applied Soft Computing 143, 110415. 28

    2023

  35. [43]

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.,

  36. [44]

    Improved techniques for training gans, in: Proceedings of the 30th International Conference on Neural Information Processing Systems, p

    Salimans, T., Goodfellow, I., Zaremba, W., Cheung, V., Radford, A., Chen, X., 2016. Improved techniques for training gans, in: Proceedings of the 30th International Conference on Neural Information Processing Systems, p. 22342242

    2016

  37. [45]

    A study of convolutional architectures for handshape recognition applied to sign language, in: XXIII Congreso Argentino de Ciencias de la Computaci´ on (La Plata, 2017)

    2017

  38. [46]

    Learning from simulated and unsupervised images through ad- versarial training, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Shrivastava, A., Pfister, T., Tuzel, O., Susskind, J., Wang, W., Webb, R., 2017. Learning from simulated and unsupervised images through ad- versarial training, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2107–2116

    2017

  39. [47]

    A survey of deep active learning

    Ren, P., Xiao, Y., Chang, X., Huang, P.Y., Li, Z., Gupta, B.B., Chen, X., Wang, X., 2021. A survey of deep active learning. ACM computing surveys (CSUR) 54, 1–40

    2021

  40. [48]

    Tremblay, J., Prakash, A., Acuna, D., Brophy, M., Jampani, V., Anil, C., To, T., Cameracci, E., Boochoon, S., Birchfield, S., 2018. Training deep networks with synthetic data: Bridging the reality gap by domain randomization, in: Proceedings of the IEEE conference on computer ...

    2018

  41. [49]

    Improving diffusion models as an alternative to gans

    Vahdat, A., Kreis, K., 2022. Improving diffusion models as an alternative to gans. NVIDIA Developer Blog

    2022

  42. [50]

    arXiv:2112.10752

    High-resolution image synthesis with latent diffusion models. arXiv:2112.10752

    arXiv

  43. [51]

    Signavatars: A large- scale 3d sign language holistic motion dataset andbenchmark, in: Pro- ceedings of the European Conference on Computer Vision (ECCV), pp

    Yu, Z., Huang, S., Cheng, Y., Birdal, T., 2024. Signavatars: A large- scale 3d sign language holistic motion dataset andbenchmark, in: Pro- ceedings of the European Conference on Computer Vision (ECCV), pp. 1–19

    2024

  44. [52]

    A survey on generative adversarial networks for imbalance problems in computer vision tasks

    Sampath, V., Maurtua, I., Aguilar Mart ´ ın, J.J., Gutierrez, A., 2021. A survey on generative adversarial networks for imbalance problems in computer vision tasks. Journal of Big Data 8, 27

    2021

  45. [53]

    How does mixup help with robustness and generalization? CoRR abs/2010.04819

    Zhang, L., Deng, Z., Kawaguchi, K., Ghorbani, A., Zou, J.Y., 2020. How does mixup help with robustness and generalization? CoRR abs/2010.04819

    2020 arXiv

  46. [54]

    Efficientnetv2: Smaller models and faster train- ing, in: International conference on machine learning, pp

    Tan, M., Le, Q., 2021. Efficientnetv2: Smaller models and faster train- ing, in: International conference on machine learning, pp. 10096–10106

    2021

  47. [55]

    Random erasing data augmentation

    Zhong, Z., Zheng, L., Kang, G., Li, S., Yang, Y., 2020. Random erasing data augmentation. Proceedings of the AAAI Conference on Artificial Intelligence 34, 13001–13008

    2020

  48. [56]

    Domain general- ization: A survey

    Zhou, K., Liu, Z., Qiao, Y., Xiang, T., Loy, C.C., 2022. Domain general- ization: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 4396–4415

    2022

  49. [57]

    Generative adver- sarial network with transformer generator for boosting ecg classification

    Xia, Y., Xu, Y., Chen, P., Zhang, J., Zhang, Y., 2023. Generative adver- sarial network with transformer generator for boosting ecg classification. Biomedical Signal Processing and Control 80, 104276

    2023

  50. [58]

    Learning to estimate 3d hand pose from single RGB images, in: Proceedings of the IEEE international conference on computer vision, pp

    Zimmermann, C., Brox, T., 2017. Learning to estimate 3d hand pose from single RGB images, in: Proceedings of the IEEE international conference on computer vision, pp. 4903–4911. 30

    2017

  51. [59]

    mixup: Beyond empirical risk minimization

    Zhang, H., Ciss´ e, M., Dauphin, Y.N., Lopez-Paz, D., 2017. mixup: Beyond empirical risk minimization. CoRR abs/1710.09412. 29

    2017 arXiv

  52. [61]

    An overview of multi-task learning

    Zhang, Y., Yang, Q., 2017. An overview of multi-task learning. National Science Review 5, 30–43

    2017

  53. [64]

    A gan-based hybrid sampling method for imbalanced customer classification

    Zhu, B., Pan, X., vanden Broucke, S., Xiao, J., 2022. A gan-based hybrid sampling method for imbalanced customer classification. Information Sciences 609, 1397–1411

    2022

  54. [2017]

    Gans trained by a two time-scale update rule converge to a lo- cal nash equilibrium, in: Advances in Neural Information Processing Systems

  55. [2018]

    Synthetic data augmentation using GAN for improved liver lesion classification, in: 2018 IEEE 15th international symposium on biomed- ical imaging (ISBI 2018), pp. 289–293

    2018

  56. [2019]

    Sign language recognition, generation, and translation: An inter- disciplinary perspective, in: Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility, pp. 16–31

  57. [2021]

    ACM Trans

    The fate landscape of sign language ai datasets: An interdisci- plinary perspective. ACM Trans. Access. Comput. 14

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.