Pith. sign in

REVIEW 4 major objections 4 minor 35 references

Sensitive Image Classification by Vision Transformers

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

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

Pith's one-line read Vision transformers surpass CNNs on sensitive-image classification benchmarks.

desk verdict Useful empirical comparison of ViTs for pornography classification, but the reported gains over prior methods are not trustworthy because the clean test images come from ImageNet, the same pool as the fine-tuning checkpoints. read the letter →

arxiv 2412.16446 v1 pith:5WCB7JHF submitted 2024-12-21 cs.CV cs.AI

Hanxian He , Campbell Wilson , Thanh Thi Nguyen , Janis Dalins This is my paper · ORCID
classification cs.CVcs.AI
keywords visiontransformerpornographyclassificationCSAMdetectionself-attentionHiLoattentionimagecontentmoderationbenchmarkdataset
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks whether vision transformer backbones can classify pornographic and porn-indicative images better than the convolutional networks that have dominated this task. It builds two new benchmark datasets, a two-class and a three-class version, with the third class covering images that are suggestive but not explicit, and evaluates ViT, DeiT, Swin, and LITv2 models against pre-trained ResNet18 and published CNN methods. The central claim is that fine-tuned transformer models exceed 95% test accuracy on the two-class dataset, surpassing the 92.72% reported for the AttM-CNN method, and that LITv2 models, which separate high- and low-frequency attention, generalize best on the harder three-class task. If true, this gives content-moderation and law-enforcement systems a concrete accuracy gain by swapping ResNet backbones for transformer encoders.

What carries the argument

The load-bearing mechanism is the multi-head self-attention of vision transformers, which lets the model relate image patches across the whole image instead of through local receptive fields. LITv2's HiLo attention sharpens this by splitting attention heads into high-frequency heads for local fine detail and low-frequency heads for global structure; the paper argues this balance is why LITv2 models perform best on the three-class data. The pre-trained checkpoints from ImageNet-1K supply the starting features, and the fine-tuning protocol on each dataset is what produces the reported gains.

What would settle it

Run a near-duplicate search between the P2 clean test images (and the ACI test set) and the ImageNet-1K training set used for pre-training. If a substantial fraction of test images have near-duplicates in the pre-training set, the fine-tuned transformer accuracy is inflated and the comparison against ResNet18 and AttM-CNN is not fair; without such overlap, the claim stands.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that self-attention-based vision transformers are better suited to pornography and porn-indicative image classification than the pre-trained ResNet18 baseline and established CNN-based detectors. Fine-tuned ViT-base, DeiT-S, and Swinv2-S each reach above 95% accuracy on the two-class P2 test set, outperforming AttM-CNN's 92.72%; on the three-class P2 set, which adds an ambiguous 'porn-indicative' category, the same models maintain roughly 89-90% accuracy while ResNet18 drops to 78.75%. The paper also finds that LITv2 models trained from scratch approach the accuracy of ImageNet-pretrained transformers on the two main classes, and that a LITv2-medium with a 0.9 attention-head ratio reaches 84.86% on the three-class validation set, the best among the from-scratch models.

Load-bearing premise

The paper assumes that the clean test images taken from ImageNet do not overlap with the ImageNet-1K images used to pre-train the transformer models, so the reported accuracy reflects genuine generalization rather than memory.

Editorial extensions

If this is right

  • Content-moderation systems can expect a practical accuracy gain by replacing ResNet18 backbones with fine-tuned ViT, DeiT, Swin, or LITv2 encoders.
  • The three-class formulation with a porn-indicative category is workable for transformer models, which keep roughly 89-90% accuracy on the two main classes when the ambiguous class is added.
  • LITv2-style attention that mixes local high-frequency and global low-frequency information gives better three-class performance than pure ViT or Swin-style attention.
  • Larger transformer variants with more parameters improve accuracy when compute is available.
  • Fine-tuning from ImageNet-1K checkpoints is substantially better than training from scratch for this task.

Reading between the lines

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

  • A direct test of the overlap concern would be to compare model accuracy on clean P2 test images that are confirmed unique versus those with near-duplicates in ImageNet-1K; the gap would quantify how much of the transformer advantage is memorization.
  • The same three-class formulation could be extended to video frames or to finer CSAM severity gradations, where the porn-indicative category is even more loaded.
  • The attention-map visualizations the authors plan could reveal whether transformer errors on porn-indicative images stem from missing local cues or from global context, which would guide whether high-frequency head ratios should be tuned per dataset.
  • Because the P2 datasets are small and imbalanced in the indicative class, the reported three-class numbers likely understate real-world performance on rare categories; balancing or synthetic augmentation would be a natural next probe.
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

4 major / 4 minor

Summary. The paper studies vision transformer (ViT) classifiers for sensitive-image detection, motivated by CSAM moderation. The authors construct a three-class dataset (P2) with clean, porn-indicative, and pornographic images, and also use the ACI benchmark. They compare fine-tuned ViT-base, DeiT-S, and Swinv2-S models against a fine-tuned ResNet18, as well as against published results of AttM-CNN and the Bumble private-detector model. On the two-class P2 test set, the fine-tuned ViT models report accuracies above 95%, exceeding ResNet18 (89.50%) and the 92.72% figure quoted for AttM-CNN. On the three-class P2 dataset, LITv2 models trained from scratch are reported to outperform Swin and DeiT. The paper concludes that ViT networks surpass pre-trained ResNet models and established detection methods for this task.

Significance. If the central comparison were sound, the paper would provide a useful practical evaluation of transformer backbones for a socially important content-moderation task, and the proposed three-class P2 dataset could be a resource for future work. The internal comparison among transformer variants (LITv2 vs Swin vs DeiT, and the effect of high/low-frequency attention ratios) is informative and appears internally consistent. However, the headline claims against established methods rest on three load-bearing weaknesses: possible train/test overlap with ImageNet-1K pretraining, a Bumble threshold tuned on the test set, and an AttM-CNN accuracy taken from a different dataset. These issues currently leave the main superiority claim unsupported, although they are in principle addressable with a revised experimental protocol.

major comments (4)
  1. [Section IV-A, Table VI, Table X] The clean images in the P2 dataset are sampled from ImageNet (Section IV-A), while the fine-tuned ViT, DeiT, and Swinv2 models use checkpoints pre-trained on ImageNet-1K (Table VI). The paper provides no deduplication or exclusion of P2 test images that may overlap with the pretraining distribution. As the authors partially acknowledge in the Conclusion only for ACI, this overlap can inflate test accuracy through memorization. Because the reported superiority over ResNet18 in Table X is the central claim, the authors must either demonstrate that the P2 clean test images are disjoint from ImageNet-1K, remove overlapping images, or retrain the comparison without pretraining data that contains the test distribution.
  2. [Section IV-C] The Bumble comparison is compromised by test-set threshold selection. The paper states that 'we selected 20% as the boundary threshold between clean and pornography categories after a series of threshold comparisons,' and then reports accuracies of 89.27% and 89.87% on P2 and ACI. Tuning the threshold on the test set and then reporting accuracy on the same test set makes the comparison with Bumble partly circular and overstates the authors' model's advantage. The threshold should be fixed on a validation set or the comparison should be reported as a precision-recall curve over thresholds.
  3. [Section IV-C] The comparison with AttM-CNN is not apples-to-apples. The paper claims that ViT models achieved accuracy greater than 95% on the P2 two-class dataset, 'surpassing the performance of AttM-CNN [6], which achieved an accuracy of 92.72%.' However, the 92.72% figure is reported by Gangwar et al. on their own dataset, not on the P2 or ACI test sets used in this paper. A meaningful comparison requires evaluating AttM-CNN (or a reimplementation) on the same test sets as the proposed models.
  4. [Tables VII-X] All reported accuracies are single numbers without error bars, repeated runs, or statistical significance tests. The differences among the fine-tuned transformer models are small (for example, ViT-base 0.9596, DeiT-S 0.9522, Swinv2-S 0.9520 in Table X), and without variance estimates it is unclear whether these differences or even the gap over ResNet18 are reliable. Given the potential ImageNet overlap identified above, the authors should report multiple seeds with means and standard deviations, or otherwise establish that the observed gaps are not noise.
minor comments (4)
  1. [Table X] The heading says 'RE-TRAINING FROM CHECKPOINTS PRE-TRAINED ON IMAGENET-1K,' but the described procedure is fine-tuning from checkpoints, not retraining from scratch; please use 'fine-tuning' for clarity.
  2. [Section III-C and Table VII] The LITv2 attention-head ratio is described only via the HiLo figure and the statement that α is the ratio of low-frequency heads. Please state explicitly whether the reported values 0.1, 0.4, and 0.9 correspond to α or to the high-frequency fraction in Tables VII and VIII.
  3. [Table III] The porn-indicative class has 512 training images but 640 validation images; this reversed size ratio is unusual and should be explained or corrected, as it may affect the reported validation accuracy for the three-class task.
  4. [Section IV-C] The phrase 'despite the occurrence of overfitting' is asserted without supporting training/validation curves or regularization details; please either provide evidence of overfitting or rephrase.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central ViT-versus-ResNet benchmark is self-contained; the Bumble threshold choice is a methodology concern, not a circular step.

full rationale

This paper is an empirical benchmark comparison rather than a derivation from first principles. The central claim, that ImageNet-1K-pretrained ViT-base, DeiT-S, and Swinv2-S fine-tuned on the P2 dataset exceed 95% test accuracy and outperform the fine-tuned ResNet18 (Table X), is supported by held-out test accuracies; no parameter fitted to those test labels is used to generate the claimed ViT results. The only fitted quantity is the 20% decision threshold applied to the external Bumble model, selected after a series of threshold comparisons in Section IV-C. That affects the reported accuracy of the external comparator but does not enter the paper's own model outputs, so the ViT-versus-ResNet comparison is not circular; rather, it is a benchmark-fairness concern. The Conclusion admits uncertainty about ImageNet-1K and ACI overlap, which is a data-leakage or correctness risk rather than circularity. Self-citation to reference [13] is a baseline-method reference and is not load-bearing. Therefore, no step in the paper's derivation chain is equivalent by construction to its inputs.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central results rest on the correctness of manually assigned labels, the reliability of scraped Reddit labels, and the assumption that ImageNet pre-training does not overlap with test images. The Bumble threshold is the only explicit free parameter fitted to data.

free parameters (1)
  • Bumble classification threshold = 20%
    Selected after 'a series of threshold comparisons' on the test data (Section IV-C); the threshold choice directly affects the reported Bumble accuracy of 89.27% and 89.87%, making the ViT-vs-Bumble comparison partly fitted.
assumptions (3)
  • domain assumption The porn-indicative category labels assigned by manual selection by the authors are reliable ground truth.
    Section IV-A describes manual selection of images into pornography, porn-indicative, and clean without reporting inter-annotator agreement or a labeling protocol; the central 3-class results depend on these labels.
  • domain assumption ImageNet-1K pre-trained checkpoints have no overlap with the ACI or P2 test images.
    The paper itself flags in the Conclusion that overlap between ImageNet-1K and ACI could affect accuracy; if such overlap exists, the fine-tuned models' test accuracy is inflated.
  • domain assumption The Reddit NSFW dataset labels used to source pornographic and porn-indicative images are accurate and complete.
    Section IV-A sources positive samples from the Reddit NSFW dataset; no verification of label accuracy for the scraped images is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sensitive Image Classification by Vision Transformers." pith.science (2026). https://pith.science/paper/5WCB7JHF

@misc{pith2026241216446,
  author       = {Pith},
  title        = {Pith review of: Sensitive Image Classification by Vision Transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5WCB7JHF}},
  note         = {Machine review of arXiv:2412.16446}
}
read the original abstract

When it comes to classifying child sexual abuse images, managing similar inter-class correlations and diverse intra-class correlations poses a significant challenge. Vision transformer models, unlike conventional deep convolutional network models, leverage a self-attention mechanism to capture global interactions among contextual local elements. This allows them to navigate through image patches effectively, avoiding incorrect correlations and reducing ambiguity in attention maps, thus proving their efficacy in computer vision tasks. Rather than directly analyzing child sexual abuse data, we constructed two datasets: one comprising clean and pornographic images and another with three classes, which additionally include images indicative of pornography, sourced from Reddit and Google Open Images data. In our experiments, we also employ an adult content image benchmark dataset. These datasets served as a basis for assessing the performance of vision transformer models in pornographic image classification. In our study, we conducted a comparative analysis between various popular vision transformer models and traditional pre-trained ResNet models. Furthermore, we compared them with established methods for sensitive image detection such as attention and metric learning based CNN and Bumble. The findings demonstrated that vision transformer networks surpassed the benchmark pre-trained models, showcasing their superior classification and detection capabilities in this task.

Figures

Figures reproduced from arXiv: 2412.16446 by the authors.

Figure 1
Figure 1. A structure of the ViT transformer module (redrawn from [30]). [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. A structure of the Swin transformer model (redrawn from [31]). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The Structure of the Hi-Lo attention model LITv2 (redrawn from [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Image classification accuracy with different models on the P2 3- [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Image classification per-class accuracy with different models on [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Image classification accuracy with different models on the P2 3- [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 26 canonical work pages

  1. [6]

    AttM- CNN: Attention and metric learning based CNN for pornography, age and child sexual abuse (CSA) detection in images,

    A. Gangwar, V . Gonz ´alez-Castro, E. Alegre, and E. Fidalgo, “AttM- CNN: Attention and metric learning based CNN for pornography, age and child sexual abuse (CSA) detection in images,” Neurocomputing, vol. 445, pp. 81–104, Jul 2021

    work page 2021

  2. [1]

    Findings from WeProtect global alliance/ technology coalition survey of technology companies

    W. G. Alliance, “Findings from WeProtect global alliance/ technology coalition survey of technology companies.” [Online]. Available: https://www.weprotect.org/survey-of-tech-companies/

    work page

  3. [2]

    The tale of Telegram governance: When the rule of thumb fails,

    F. Badiei, “The tale of Telegram governance: When the rule of thumb fails,” SMGI Social Media Governance Initiative , January 28 2021

    work page 2021

  4. [3]

    Detecting child sexual abuse material: A comprehensive survey,

    H.-E. Lee, T. Ermakova, V . Ververis, and B. Fabian, “Detecting child sexual abuse material: A comprehensive survey,” Forensic Science International: Digital Investigation , vol. 34, p. 301022, Sep. 2020

    work page 2020

  5. [4]

    Smart content recognition from images using a mixture of convolutional neural networks,

    T. Connie, M. Al-Shabi, and M. Goh, “Smart content recognition from images using a mixture of convolutional neural networks,” in IT Convergence and Security: Volume 1 . Springer, 2017, pp. 11–18

    work page 2017

  6. [5]

    20k nudity dataset,

    A. Noktedan, “20k nudity dataset,” 2020. [Online]. Available: https://figshare.com/articles/dataset/Adult content dataset/13456484

    work page arXiv 2020

  7. [7]

    Description of the neural network based on AB/DL pictures. Possible implications for forensic sexology,

    W. Oronowicz-Ja ´skowiak and P. Wasilewski, “Description of the neural network based on AB/DL pictures. Possible implications for forensic sexology,” Postepy Psychiatrii Neurologii, vol. 31, no. 4, pp. 161–166, Dec. 2022

    work page 2022

  8. [8]

    LSPD: A large-scale pornographic dataset for detection and classification,

    D. D. Phan, T. T. Nguyen, Q. H. Nguyen, H. L. Tran, K. N. K. Nguyen, and D. L. Vu, “LSPD: A large-scale pornographic dataset for detection and classification,” International Journal of Intelligent Engineering and Systems, vol. 15, no. 1, 2022

    work page 2022

Show all 35 references
  1. [9]

    NotAI-tech. Nudenet. [Online]. Available: https://archive.org/details/ NudeNet classifier dataset v1

  2. [10]

    Detecting pornographic images by localizing skin rois,

    S. Karavarsamis, N. Ntarmos, K. Blekas, and I. Pitas, “Detecting pornographic images by localizing skin rois,” International Journal of Digital Crime and Forensics (IJDCF) , vol. 5, pp. 39–53, 01 2013

    2013

  3. [11]

    NuDetective: A Forensic Tool to Help Combat Child Pornography through Automatic Nudity Detection,

    M. de Castro Polastro and P. M. da Silva Eleuterio, “NuDetective: A Forensic Tool to Help Combat Child Pornography through Automatic Nudity Detection,” in Workshops on Database and Expert Systems Applications, 2010, pp. 349–353

    2010

  4. [12]

    Open nsfw model,

    J. Mahadeokar, S. Farfade, A. R. Kamat, and A. Kappeler, “Open nsfw model,” Jul. 2023, published date: Aug 2016. [Online]. Available: https://github.com/yahoo/open nsfw

    2023

  5. [13]

    Laying foundations for effective machine learning in law enforce- ment. Majura – A labelling schema for child exploitation materials,

    J. Dalins, Y . Tyshetskiy, C. Wilson, M. J. Carman, and D. Boudry, “Laying foundations for effective machine learning in law enforce- ment. Majura – A labelling schema for child exploitation materials,” Digital Investigation, vol. 26, pp. 40–54, 2018

    2018

  6. [14]

    Neural machine trans- lation by jointly learning to align and translate,

    D. Bahdanau, K. Cho, and Y . Bengio, “Neural machine trans- lation by jointly learning to align and translate,” arXiv preprint arXiv:1409.0473, 2014

    2014 arXiv

  7. [15]

    Survey on the attention based RNN model and its applications in computer vision,

    F. Wang and D. M. Tax, “Survey on the attention based RNN model and its applications in computer vision,” arXiv preprint arXiv:1601.06823, 2016

    2016 arXiv

  8. [16]

    The kinetics human action video dataset,

    W. Kay, J. Carreira, K. Simonyan, B. Zhang, C. Hillier, S. Vijaya- narasimhan, F. Viola, T. Green, T. Back, P. Natsevet al., “The kinetics human action video dataset,” arXiv preprint arXiv:1705.06950 , 2017

    2017 arXiv

  9. [17]

    The “something something

    R. Goyal, S. Ebrahimi Kahou, V . Michalski, J. Materzynska, S. West- phal, H. Kim, V . Haenel, I. Fruend, P. Yianilos, M. Mueller-Freitag et al. , “The “something something” video database for learning and evaluating visual common sense,” in Proceedings of the IEEE Internation...

    2017

  10. [18]

    Multimodal learning with trans- formers: A survey,

    P. Xu, X. Zhu, and D. A. Clifton, “Multimodal learning with trans- formers: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 12 113–12 132, 2023

    2023

  11. [19]

    Frozen CLIP models are efficient video learners,

    Z. Lin, S. Geng, R. Zhang, P. Gao, G. de Melo, X. Wang, J. Dai, Y . Qiao, and H. Li, “Frozen CLIP models are efficient video learners,” in European Conference on Computer Vision . Springer, 2022, pp. 388–404

    2022

  12. [20]

    VideoMAE: Masked autoencoders are data-efficient learners for self-supervised video pre- training,

    Z. Tong, Y . Song, J. Wang, and L. Wang, “VideoMAE: Masked autoencoders are data-efficient learners for self-supervised video pre- training,” Advances in Neural Information Processing Systems, vol. 35, pp. 10 078–10 093, 2022

    2022

  13. [21]

    Is space-time attention all you need for video understanding?

    G. Bertasius, H. Wang, and L. Torresani, “Is space-time attention all you need for video understanding?” arXiv preprint arXiv:2102.05095, 2021

    2021 arXiv

  14. [22]

    PolyViT: Co-training vision transform- ers on images, videos and audio,

    V . Likhosherstov, A. Arnab, K. Choromanski, M. Lucic, Y . Tay, A. Weller, and M. Dehghani, “PolyViT: Co-training vision transform- ers on images, videos and audio,” arXiv preprint arXiv:2111.12993 , 2021

    2021 arXiv

  15. [23]

    Omnimae: Single model masked pretraining on images and videos,

    R. Girdhar, A. El-Nouby, M. Singh, K. V . Alwala, A. Joulin, and I. Misra, “Omnimae: Single model masked pretraining on images and videos,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 10 406–10 417

    2023

  16. [24]

    Omnivore: A single model for many visual modalities,

    R. Girdhar, M. Singh, N. Ravi, L. van der Maaten, A. Joulin, and I. Misra, “Omnivore: A single model for many visual modalities,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 16 102–16 112

    2022

  17. [25]

    M&M mix: A multimodal multiview transformer ensemble,

    X. Xiong, A. Arnab, A. Nagrani, and C. Schmid, “M&M mix: A multimodal multiview transformer ensemble,” arXiv preprint arXiv:2206.09852, 2022

    2022 arXiv

  18. [26]

    MultiMAE: Multi-modal multi-task masked autoencoders,

    R. Bachmann, D. Mizrahi, A. Atanov, and A. Zamir, “MultiMAE: Multi-modal multi-task masked autoencoders,” in European Confer- ence on Computer Vision . Springer, 2022, pp. 348–367

    2022

  19. [27]

    Video swin transformer,

    Z. Liu, J. Ning, Y . Cao, Y . Wei, Z. Zhang, S. Lin, and H. Hu, “Video swin transformer,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 3202–3211

    2022

  20. [28]

    BEVT: BERT pretraining of video transform- ers,

    R. Wang, D. Chen, Z. Wu, Y . Chen, X. Dai, M. Liu, Y .-G. Jiang, L. Zhou, and L. Yuan, “BEVT: BERT pretraining of video transform- ers,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 14 733–14 743

    2022

  21. [29]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of the 31st International Conference on Neural Informa- tion Processing Systems , 2017, p. 6000–6010

    2017

  22. [30]

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

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” in International Conference on Learning...

    2021

  23. [31]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2021, pp. 10 012–10 022

    2021

  24. [32]

    Fast vision transformers with HiLo at- tention,

    Z. Pan, J. Cai, and B. Zhuang, “Fast vision transformers with HiLo at- tention,” Advances in Neural Information Processing Systems , vol. 35, pp. 14 541–14 554, 2022

    2022

  25. [33]

    State-of-the-art in nudity classification: A comparative analysis,

    F. C. Akyon and A. Temizel, “State-of-the-art in nudity classification: A comparative analysis,” in IEEE International Conference on Acous- tics, Speech, and Signal Processing Workshops (ICASSPW) , 2023, pp. 1–5

    2023

  26. [34]

    The Bumble’s private detector model

    The Bumble Tech, “The Bumble’s private detector model.” [Online]. Available: https://github.com/bumble-tech/private-detector

  27. [35]

    EfficientNet: Rethinking model scaling for con- volutional neural networks,

    M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for con- volutional neural networks,” in International Conference on Machine Learning. PMLR, 2019, pp. 6105–6114

    2019

Pith tools

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