pith. machine review for the scientific record. sign in

arxiv: 2604.04086 · v1 · submitted 2026-04-05 · 💻 cs.CV

Recognition: no theorem link

LAA-X: Unified Localized Artifact Attention for Quality-Agnostic and Generalizable Face Forgery Detection

Anis Kacem, Dat Nguyen, Djamila Aouada, Enjie Ghorbel, Marcella Astrid

Pith reviewed 2026-05-13 17:18 UTC · model grok-4.3

classification 💻 cs.CV
keywords face forgery detectiondeepfakeattention mechanismmulti-task learninggeneralizationartifact localizationblending synthesissynthetic data
0
0 comments X

The pith

LAA-X uses a multi-task framework to explicitly direct attention to localized artifact-prone regions in faces, enabling detection of high-quality forgeries and generalization to unseen manipulations when trained only on real and pseudo-fake

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

The paper presents LAA-X as a framework for face forgery detection that replaces implicit attention in binary classifiers with an explicit strategy. It combines multi-task learning, where auxiliary tasks steer the model toward vulnerable regions, with blending-based synthesis to create pseudo-fake training samples. This setup produces models that remain effective on high-quality forgeries and adapt to new manipulation types. The approach supports both CNN and transformer backbones. Readers would care because it offers a route to detectors that do not require exposure to actual forged examples to achieve broad performance.

Core claim

LAA-X shows that an explicit attention strategy grounded in multi-task learning and blending-based data synthesis allows a detector to focus on localized artifact-prone regions. Trained solely on real images and pseudo-fakes, the resulting models compete with state-of-the-art methods on multiple benchmarks for both quality-agnostic and generalizable face forgery detection.

What carries the argument

Explicit Localized Artifact Attention implemented as auxiliary tasks within a multi-task learning framework, reinforced by blending-based synthesis of pseudo-fake samples.

Load-bearing premise

The auxiliary tasks will reliably steer attention to the actual regions containing forgery artifacts, and the blending synthesis will generate pseudo-fakes that closely enough resemble real high-quality forgeries.

What would settle it

A test set of high-quality forgeries produced by manipulation methods outside the blending synthesis family, where the LAA-X models fall below current top detection accuracy.

Figures

Figures reproduced from arXiv: 2604.04086 by Anis Kacem, Dat Nguyen, Djamila Aouada, Enjie Ghorbel, Marcella Astrid.

Figure 1
Figure 1. Figure 1: Comparison of LAA-Net(•), LAA-Former(•), and LAA-Swin(•) with respect to existing methods, namely, Multi-attentional(•) [13], SBI(•) [14], Xception(•) [3], RECCE(•) [15], CADDM(•) [16], FAViT(•) [17], Forensic￾sAdapter(•) [18] using (a) the AUC performance with respect to different ranges of Mask-SSIM, and (b) its associated boxplots. *The results were obtained using the official source codes pretrained on… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed LAA-X framework. LAA￾X is a multi-task learning framework that incorporates an explicit attention mechanism to vulnerable regions through the integration of generic auxiliary tasks. This strategy enables LAA-X to adequately attend to fine-grained artifact-prone areas. Particularly, these additional tasks can be removed at inference, reducing computational cost at test time. binary … view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed LAA-Net approach. It is formed by two main components, namely, (1) an explicit attention mechanism based on a multi-task learning framework composed of three branches, i.e., the binary classification branch, the heatmap branch, and the self-consistency branch. The heatmap and self-consistency ground-truth data are generated based on the detected vulnerable points, and (2) an Enhanc… view at source ↗
Figure 4
Figure 4. Figure 4: Extraction of the vulnerable points. ⊕ ⊕ ⊕ ⊙ ⊙ ⊕ E⁽⁴⁾ E⁽³⁾ E⁽² F ⁾ '⁽⁴⁾ F'⁽³⁾ ⊙ F'⁽²⁾ ⊕ ⊙ (a) (a) (a) 1 – Sigmoid(.) Convolution 5×5 Block Convolution 3×3 Block Max Pooling Convolution 1×1 Block Transpose Convolution Block Element-wise Multiplication Result of Multi-Layer Fusion Concatenation (a) Multi-layer Fusion F⁽⁰⁾ F⁽¹⁾F⁽²⁾ F⁽³⁾ F⁽⁴⁾ F⁽⁵⁾ F⁽⁶⁾ ⊕ ⊙ (a) F'⁽⁵⁾ E (5) [PITH_FULL_IMAGE:figures/full_fig_p00… view at source ↗
Figure 6
Figure 6. Figure 6: Experiments to analyze the capability of transformer-based networks in deepfake detection. (a) Generalization performance comparison of baseline classifiers (ViT [29]+SBI [14](•), Swin [32]+SBI [14](•)) with two specialized transformer-based (FAViT [17](•), ForensicsAdapter [18](•)) and four CNN-based methods (LAA-Net [23]+SBI [14](•), CADDM [16](•), EfficientNet [25]+SBI [14](•), Xception [26]+SBI [14](•)… view at source ↗
Figure 7
Figure 7. Figure 7: Examples are randomly selected to illustrate four types of deepfakes in the common FF++ [3] dataset. It includes Deepfakes (DF) [80], FaceSwap (FS) [81], Face2Face (F2F) [82], and NeuralTextures (NT) [83]. b) Performance of ViTs with respect to the patch size and the type of deepfakes: We further investigate whether there exists a correlation between the patch size and the performance of ViT in deepfake de… view at source ↗
Figure 8
Figure 8. Figure 8: The proposed Transformer-based LAA-X method. (I) The overall LAA-Former framework, (II) the L2-Att module, and (III) the ground-truth generation of vulnerable patches. Consequently, the implicit SA mechanism might overlook or miss important patches, as those containing forgeries can appear too analogous to those without. As also highlighted in previous work [18], the blending boundary forgery only occupy a… view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of saliency maps of different com￾ponents in LAA-Net w/o E-FPN, w/o H, and w/o C refer to ablating E-FPN, heatmap branch, and self-consistency branch, respectively. TABLE VI: Ablation study of LAA-Net’s components including the Consistency branch (C), Heatmap branch (H), and E-FPN. C H E-FPN Test set AUC (%) CDF2 DFD DFDCP DFW Avg. × × × 74.54 92.24 70.81 59.81 74.35 × ✓ × 80.89 94.53 77.93 … view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of saliency maps on different types of manipulation from FF++ [3]. LAA-Net and LAA-Former is compared to SBI [14], Xception [3], and MAT [13]. and Contrast, do not impact the performance. However, the proposed method is extremely sensitive to structural pertur￾bations such as Gaussian Noise. One possible reason is due to the introduction of noise that elevates the difficulty of detecting the … view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of saliency maps using E-FPN and FPN with different integration of multi-scale layers. We can see that E-FPN can focus better on artifacts as compared to FPN. The setup details are provided in Table V. E. Additional Discussions on Transformer-based LAA-X: LAA-Former Effect of Patch Size. As shown in Figure 6b, patch size has a clear effect on ViT’s learning capability under different train￾i… view at source ↗
Figure 12
Figure 12. Figure 12: In order to generate the consistency map prediction [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The generation process of ground truth heatmaps by producing using an [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Detection of vulnerable points w/o and w/ Gaussian [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
read the original abstract

In this paper, we propose Localized Artifact Attention X (LAA-X), a novel deepfake detection framework that is both robust to high-quality forgeries and capable of generalizing to unseen manipulations. Existing approaches typically rely on binary classifiers coupled with implicit attention mechanisms, which often fail to generalize beyond known manipulations. In contrast, LAA-X introduces an explicit attention strategy based on a multi-task learning framework combined with blending-based data synthesis. Auxiliary tasks are designed to guide the model toward localized, artifact-prone (i.e., vulnerable) regions. The proposed framework is compatible with both CNN and transformer backbones, resulting in two different versions, namely, LAA-Net and LAA-Former, respectively. Despite being trained only on real and pseudo-fake samples, LAA-X competes with state-of-the-art methods across multiple benchmarks. Code and pre-trained weights for LAA-Net\footnote{https://github.com/10Ring/LAA-Net} and LAA-Former\footnote{https://github.com/10Ring/LAA-Former} are publicly available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes LAA-X, a unified deepfake detection framework that employs explicit localized artifact attention through a multi-task learning setup combined with blending-based synthesis to generate pseudo-fake training samples. It introduces two backbone variants (LAA-Net for CNNs and LAA-Former for transformers) and claims that training solely on real and pseudo-fake images yields competitive performance with state-of-the-art methods on multiple benchmarks while improving robustness to high-quality forgeries and generalization to unseen manipulations.

Significance. If the empirical results hold, the work would be significant for addressing data scarcity in forgery detection by showing that blending-based pseudo-fakes plus auxiliary localization tasks can produce generalizable attention maps. The public release of code and pre-trained weights for both variants is a clear strength for reproducibility.

major comments (2)
  1. [Abstract] Abstract: the claim of competitive benchmark performance and generalization to unseen manipulations is asserted without any quantitative metrics, ablation studies, error bars, or description of the test protocol for unseen manipulations, leaving the central empirical claim unsupported in the provided text.
  2. [Method] Method (multi-task framework): the assumption that auxiliary tasks will reliably surface the same vulnerable regions as real high-quality forgeries is load-bearing, yet blending synthesis (alpha or Poisson) primarily introduces low-level boundary discontinuities rather than the frequency or identity-inconsistency artifacts of methods such as FaceSwap; no evidence is shown that the learned attention transfers when the training distribution is replaced by real forged samples.
minor comments (2)
  1. Clarify the exact formulation of the multi-task loss (including balancing weights) and whether they are tuned on a validation set or held fixed.
  2. Specify the precise auxiliary heads (e.g., pixel-wise localization, region classification) and their individual loss functions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and have revised the manuscript to strengthen the presentation of results and supporting evidence.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of competitive benchmark performance and generalization to unseen manipulations is asserted without any quantitative metrics, ablation studies, error bars, or description of the test protocol for unseen manipulations, leaving the central empirical claim unsupported in the provided text.

    Authors: We agree that the abstract should be more self-contained. In the revised version we have added specific quantitative results (AUC on FF++, Celeb-DF, and cross-manipulation settings) together with a concise description of the unseen-manipulation evaluation protocol. Full ablations, standard deviations across runs, and detailed protocols remain in the experimental section. revision: yes

  2. Referee: [Method] Method (multi-task framework): the assumption that auxiliary tasks will reliably surface the same vulnerable regions as real high-quality forgeries is load-bearing, yet blending synthesis (alpha or Poisson) primarily introduces low-level boundary discontinuities rather than the frequency or identity-inconsistency artifacts of methods such as FaceSwap; no evidence is shown that the learned attention transfers when the training distribution is replaced by real forged samples.

    Authors: We acknowledge that blending synthesis primarily produces boundary artifacts. Nevertheless, the auxiliary localization tasks are intended to direct attention toward a broader set of vulnerable regions. The manuscript already includes qualitative attention-map comparisons (Figure 5) on both pseudo-fakes and real FaceSwap samples that show substantial spatial overlap. Competitive performance on real high-quality forgeries (Tables 2–3) provides indirect support for transfer. To address the referee’s request for direct evidence, we have added a quantitative attention-similarity analysis between models trained on pseudo-fakes versus real forgeries in the revised supplementary material. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper proposes LAA-X as an empirical deep learning framework using multi-task auxiliary losses and blending-based pseudo-fake synthesis to train attention on localized artifacts. No equations, first-principles derivations, or parameter-fitting steps are presented that reduce any claimed performance or generalization to quantities defined from the same data or self-citations. Training on real and pseudo-fake samples follows standard practice in forgery detection; reported benchmark results are external empirical measurements, not tautological outputs of the inputs. No self-citation load-bearing steps or ansatz smuggling appear in the manuscript.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that auxiliary tasks can localize forgery artifacts and that blending synthesis sufficiently mimics real forgeries to enable generalization; no new physical entities are introduced.

free parameters (1)
  • multi-task loss balancing weights
    Weights that combine the main detection loss with auxiliary task losses; these are typically tuned during training on validation data.
axioms (1)
  • domain assumption Blending-based data synthesis produces pseudo-fake samples whose artifacts are representative of those in real high-quality forgeries
    Invoked to justify training exclusively on real and synthesized samples while claiming generalization to unseen manipulations.

pith-pipeline@v0.9.0 · 5500 in / 1362 out tokens · 50018 ms · 2026-05-13T17:18:31.484103+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

104 extracted references · 104 canonical work pages · 3 internal anchors

  1. [1]

    Deepfake presidents used in Russia-Ukraine war,

    J. Wakefield, “Deepfake presidents used in Russia-Ukraine war,” https: //www.bbc.com/news/technology-60780142, 2022, [Online; accessed 7- March-2023]

    work page 2022

  2. [2]

    How misinformation helped spark an attempted coup in Gabon,

    S. Cahlan, “How misinformation helped spark an attempted coup in Gabon,” https://wapo.st/3KZARDF, 2020, [Online; accessed 7-March- 2023]

    work page 2020

  3. [3]

    FaceForensics++: Learning to detect manipulated facial images,

    A. R ¨ossler, D. Cozzolino, L. Verdoliva, C. Riess, J. Thies, and M. Nießner, “FaceForensics++: Learning to detect manipulated facial images,” inInternational Conference on Computer Vision (ICCV), 2019

    work page 2019

  4. [5]

    Available: http://arxiv.org/abs/1809.00888

    [Online]. Available: http://arxiv.org/abs/1809.00888

    work page arXiv

  5. [6]

    Fakespotter: A simple baseline for spotting ai-synthesized fake faces,

    R. Wang, L. Ma, F. Juefei-Xu, X. Xie, J. Wang, and Y . Liu, “Fakespotter: A simple baseline for spotting ai-synthesized fake faces,”CoRR, vol. abs/1909.06122, 2019. [Online]. Available: http: //arxiv.org/abs/1909.06122

    work page arXiv 1909

  6. [7]

    Thinking in frequency: Face forgery detection by mining frequency-aware clues,

    Y . Qian, G. Yin, L. Sheng, Z. Chen, and J. Shao, “Thinking in frequency: Face forgery detection by mining frequency-aware clues,” inEuropean conference on computer vision. Springer, 2020, pp. 86–103

    work page 2020

  7. [8]

    Capsule-forensics: Using capsule networks to detect forged images and videos,

    H. H. Nguyen, J. Yamagishi, and I. Echizen, “Capsule-forensics: Using capsule networks to detect forged images and videos,”CoRR, vol. abs/1810.11215, 2018. [Online]. Available: http://arxiv.org/abs/1810. 11215

    work page arXiv 2018

  8. [9]

    Deepfake detection using spatiotemporal trans- former,

    B. Kaddar, S. A. Fezza, Z. Akhtar, W. Hamidouche, A. Hadid, and J. Serra-Sagrist `a, “Deepfake detection using spatiotemporal trans- former,”ACM Transactions on Multimedia Computing, Communications and Applications, vol. 20, no. 11, pp. 1–21, 2024

    work page 2024

  9. [10]

    Combining efficientnet and vision transformers for video deepfake detection,

    D. A. Coccomini, N. Messina, C. Gennaro, and F. Falchi, “Combining efficientnet and vision transformers for video deepfake detection,” in Image Analysis and Processing – ICIAP 2022, S. Sclaroff, C. Distante, M. Leo, G. M. Farinella, and F. Tombari, Eds. Cham: Springer International Publishing, 2022, pp. 219–229

    work page 2022

  10. [11]

    Fdftnet: Facing off fake images using fake detection fine-tuning network,

    H. Jeon, Y . Bang, and S. S. Woo, “Fdftnet: Facing off fake images using fake detection fine-tuning network,” inIFIP international conference on ICT systems security and privacy protection. Springer, 2020, pp. 416– 430

    work page 2020

  11. [12]

    Deep convolutional pooling transformer for deepfake detection,

    T. Wang, H. Cheng, K. P. Chow, and L. Nie, “Deep convolutional pooling transformer for deepfake detection,”ACM Trans. Multimedia Comput. Commun. Appl., vol. 19, no. 6, 2023

    work page 2023

  12. [13]

    Deepfake video detection with spatiotemporal dropout transformer,

    D. Zhang, F. Lin, Y . Hua, P. Wang, D. Zeng, and S. Ge, “Deepfake video detection with spatiotemporal dropout transformer,” inProceedings of the 30th ACM international conference on multimedia, 2022, pp. 5833– 5841

    work page 2022

  13. [14]

    Multi- attentional deepfake detection,

    H. Zhao, W. Zhou, D. Chen, T. Wei, W. Zhang, and N. Yu, “Multi- attentional deepfake detection,” inProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, 2021, pp. 2185–2194. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15

    work page 2021

  14. [15]

    Detecting deepfakes with self-blended images,

    K. Shiohara and T. Yamasaki, “Detecting deepfakes with self-blended images,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 18 720–18 729

    work page 2022

  15. [16]

    End-to-end reconstruction-classification learning for face forgery detection,

    J. Cao, C. Ma, T. Yao, S. Chen, S. Ding, and X. Yang, “End-to-end reconstruction-classification learning for face forgery detection,” in2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 4103–4112

    work page 2022

  16. [17]

    Implicit identity leakage: The stumbling block to improving deepfake detection gener- alization,

    S. Dong, J. Wang, R. Ji, J. Liang, H. Fan, and Z. Ge, “Implicit identity leakage: The stumbling block to improving deepfake detection gener- alization,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 3994–4004

    work page 2023

  17. [18]

    Forgery-aware adaptive learning with vision transformer for generalized face forgery detection,

    A. Luo, R. Cai, C. Kong, Y . Ju, X. Kang, J. Huang, and A. C. K. Life, “Forgery-aware adaptive learning with vision transformer for generalized face forgery detection,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

    work page 2024

  18. [19]

    Forensics adapter: Adapting clip for generalizable face forgery detection,

    X. Cui, Y . Li, A. Luo, J. Zhou, and J. Dong, “Forensics adapter: Adapting clip for generalizable face forgery detection,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 19 207–19 217

    work page 2025

  19. [20]

    Celeb-df: A large-scale challenging dataset for deepfake forensics,

    Y . Li, X. Yang, P. Sun, H. Qi, and S. Lyu, “Celeb-df: A large-scale challenging dataset for deepfake forensics,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020

    work page 2020

  20. [21]

    Self-supervised learn- ing of adversarial example: Towards good generalizations for deepfake detection,

    L. Chen, Y . Zhang, Y . Song, L. Liu, and J. Wang, “Self-supervised learn- ing of adversarial example: Towards good generalizations for deepfake detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 18 710–18 719

    work page 2022

  21. [22]

    Face x-ray for more general face forgery detection,

    L. Li, J. Bao, T. Zhang, H. Yang, D. Chen, F. Wen, and B. Guo, “Face x-ray for more general face forgery detection,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020

    work page 2020

  22. [23]

    Learning self-consistency for deepfake detection,

    E. Zhao, X. Xu, M. Xu, H. Ding, Y . Xiong, and W. Xia, “Learning self-consistency for deepfake detection,” inICCV 2021,

    work page 2021

  23. [24]

    Available: https://www.amazon.science/publications/ learning-self-consistency-for-deepfake-detection

    [Online]. Available: https://www.amazon.science/publications/ learning-self-consistency-for-deepfake-detection

    work page

  24. [25]

    Laa-net: Localized artifact attention network for quality-agnostic and generalizable deepfake detection,

    D. Nguyen, N. Mejri, I. P. Singh, P. Kuleshova, M. Astrid, A. Kacem, E. Ghorbel, and D. Aouada, “Laa-net: Localized artifact attention network for quality-agnostic and generalizable deepfake detection,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 17 395–17 405

    work page 2024

  25. [26]

    Vulnerability-aware spatio-temporal learning for generalizable deepfake video detection,

    D. Nguyen, M. Astrid, A. Kacem, E. Ghorbel, and D. Aouada, “Vulnerability-aware spatio-temporal learning for generalizable deepfake video detection,” inProceedings of the IEEE/CVF International Con- ference on Computer Vision (ICCV), October 2025, pp. 10 786–10 796

    work page 2025

  26. [27]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,”CoRR, vol. abs/1905.11946, 2019. [Online]. Available: http://arxiv.org/abs/1905.11946

    work page arXiv 1905

  27. [28]

    Xception: Deep learning with depthwise separable convolu- tions,

    F. Chollet, “Xception: Deep learning with depthwise separable convolu- tions,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1251–1258

    work page 2017

  28. [29]

    Deep Residual Learning for Image Recognition

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

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  29. [30]

    Dynamic graph learning with content-guided spatial-frequency relation reasoning for deepfake detection,

    Y . Wang, K. Yu, C. Chen, X. Hu, and S. Peng, “Dynamic graph learning with content-guided spatial-frequency relation reasoning for deepfake detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 7278–7287

    work page 2023

  30. [31]

    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: Trans- formers for image recognition at scale,”CoRR, vol. abs/2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  31. [32]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,”CoRR, vol. abs/2012.12877, 2020

    work page arXiv 2012

  32. [33]

    Twins: Revisiting the design of spatial attention in vision transformers,

    X. Chu, Z. Tian, Y . Wang, B. Zhang, H. Ren, X. Wei, H. Xia, and C. Shen, “Twins: Revisiting the design of spatial attention in vision transformers,” inNeural Information Processing Systems,

    work page

  33. [34]

    Available: https://api.semanticscholar.org/CorpusID: 234364557

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 234364557

    work page

  34. [35]

    https://arxiv.org/abs/2103.14030

    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,”CoRR, vol. abs/2103.14030, 2021

    work page arXiv 2021

  35. [36]

    Neighborhood attention transformer,

    A. Hassani, S. Walton, J. Li, S. Li, and H. Shi, “Neighborhood attention transformer,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 6185–6194

    work page 2023

  36. [37]

    Quantifying attention flow in transformers,

    S. Abnar and W. Zuidema, “Quantifying attention flow in transformers,” inAnnual Meeting of the Association for Computational Linguistics,

    work page

  37. [38]

    Available: https://api.semanticscholar.org/CorpusID: 218487351

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 218487351

    work page

  38. [39]

    Ucf: Uncovering common features for generalizable deepfake detection,

    Z. Yan, Y . Zhang, Y . Fan, and B. Wu, “Ucf: Uncovering common features for generalizable deepfake detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Oc- tober 2023, pp. 22 412–22 423

    work page 2023

  39. [40]

    Ost: Improving generalization of deepfake detection via one-shot test-time training,

    L. Chen, Y . Zhang, Y . Song, J. Wang, and L. Liu, “Ost: Improving generalization of deepfake detection via one-shot test-time training,” inAdvances in Neural Information Processing Systems, S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, Eds., vol. 35. Curran Associates, Inc., 2022, pp. 24 597–24 610. [Online]. Available: https://proce...

    work page 2022

  40. [41]

    Standing on the shoulders of giants: Reprogramming visual-language model for general deepfake detection,

    K. Lin, Y . Lin, W. Li, T. Yao, and B. Li, “Standing on the shoulders of giants: Reprogramming visual-language model for general deepfake detection,” inProceedings of the AAAI Conference on Artificial Intelli- gence, vol. 39, no. 5, 2025, pp. 5262–5270

    work page 2025

  41. [42]

    Exploring unbiased deepfake detection via token-level shuffling and mixing,

    X. Fu, Z. Yan, T. Yao, S. Chen, and X. Li, “Exploring unbiased deepfake detection via token-level shuffling and mixing,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 3, 2025, pp. 3040–3048

    work page 2025

  42. [43]

    Dfdt: An end-to-end deepfake detection framework using vision transformer,

    A. Khormali and J.-S. Yuan, “Dfdt: An end-to-end deepfake detection framework using vision transformer,”Applied Sciences, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:247495859

    work page 2022

  43. [44]

    Pose guided person image generation,

    L. Ma, X. Jia, Q. Sun, B. Schiele, T. Tuytelaars, and L. Van Gool, “Pose guided person image generation,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  44. [45]

    Image quality assess- ment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assess- ment: from error visibility to structural similarity,”IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, 2004

    work page 2004

  45. [46]

    Contributing data to deepfake detection research,

    N. Dufour and A. Gully, “Contributing data to deepfake detection research,” https://ai.googleblog.com/2019/09/ contributing-data-to-deepfake-detection.html, 2019

    work page 2019

  46. [47]

    Wilddeepfake: A challenging real-world dataset for deepfake detection,

    B. Zi, M. Chang, J. Chen, X. Ma, and Y .-G. Jiang, “Wilddeepfake: A challenging real-world dataset for deepfake detection,”Proceedings of the 28th ACM International Conference on Multimedia, 2020

    work page 2020

  47. [48]

    The deepfake detection challenge (DFDC) preview dataset,

    B. Dolhansky, R. Howes, B. Pflaum, N. Baram, and C. Canton-Ferrer, “The deepfake detection challenge (DFDC) preview dataset,”CoRR, vol. abs/1910.08854, 2019. [Online]. Available: http://arxiv.org/abs/ 1910.08854

    work page arXiv 1910

  48. [49]

    The DeepFake Detection Challenge (DFDC) Dataset

    B. Dolhansky, J. Bitton, B. Pflaum, J. Lu, R. Howes, M. Wang, and C. Canton-Ferrer, “The deepfake detection challenge dataset,”CoRR, vol. abs/2006.07397, 2020. [Online]. Available: https://arxiv.org/abs/ 2006.07397

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  49. [50]

    Df40: Toward next-generation deepfake detec- tion,

    Z. Yan, T. Yao, S. Chen, Y . Zhao, X. Fu, J. Zhu, D. Luo, C. Wang, S. Ding, Y . Wuet al., “Df40: Toward next-generation deepfake detec- tion,”Advances in Neural Information Processing Systems, vol. 37, pp. 29 387–29 434, 2024

    work page 2024

  50. [51]

    Diffusionface: Towards a comprehensive dataset for diffusion-based face forgery analysis,

    Z. Chen, K. Sun, Z. Zhou, X. Lin, X. Sun, L. Cao, and R. Ji, “Diffusionface: Towards a comprehensive dataset for diffusion-based face forgery analysis,”arXiv preprint arXiv:2403.18471, 2024

    work page arXiv 2024

  51. [52]

    Quality-agnostic deepfake detection with intra-model collaborative learning,

    B. M. Le and S. S. Woo, “Quality-agnostic deepfake detection with intra-model collaborative learning,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 22 378–22 389

    work page 2023

  52. [53]

    Diffusionfake: Enhancing generalization in deepfake detection via guided stable diffu- sion,

    S. Chen, T. Yao, H. Liu, X. Sun, S. Ding, R. Jiet al., “Diffusionfake: Enhancing generalization in deepfake detection via guided stable diffu- sion,”Advances in Neural Information Processing Systems, vol. 37, pp. 101 474–101 497, 2024

    work page 2024

  53. [54]

    Local relation learning for face forgery detection,

    S. Chen, T. Yao, Y . Chen, S. Ding, J. Li, and R. Ji, “Local relation learning for face forgery detection,” inAAAI Conference on Artificial Intelligence, 2021

    work page 2021

  54. [55]

    Exploring disentangled content infor- mation for face forgery detection,

    J. Liang, H. Shi, and W. Deng, “Exploring disentangled content infor- mation for face forgery detection,” inEuropean conference on computer vision. Springer, 2022, pp. 128–145

    work page 2022

  55. [56]

    Freqdebias: Towards generalizable deepfake detection via consistency-driven frequency debi- asing,

    H. Kashiani, N. A. Talemi, and F. Afghah, “Freqdebias: Towards generalizable deepfake detection via consistency-driven frequency debi- asing,” inProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), June 2025, pp. 8775–8785

    work page 2025

  56. [57]

    Freqblender: Enhancing deepfake detection by blending frequency knowledge,

    H. Li, J. Zhou, Y . Li, B. Wu, B. Li, and J. Dong, “Freqblender: Enhancing deepfake detection by blending frequency knowledge,” in Advances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, Eds., vol. 37. Curran Associates, Inc., 2024, pp. 44 965–44 988. [Online]. Available: https:...

    work page 2024

  57. [58]

    Aunet: Learning relations between action units for face forgery detection,

    W. Bai, Y . Liu, Z. Zhang, B. Li, and W. Hu, “Aunet: Learning relations between action units for face forgery detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2023, pp. 24 709–24 719

    work page 2023

  58. [59]

    Xinlei Chen, Hao Fang, Tsung-yi Lin, Ramakrishna Vedantam, Saurabh Gupta, Piotr Dollar, and C Lawrence Zitnick

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,”CoRR, vol. abs/2005.12872, 2020

    work page arXiv 2005

  59. [60]

    ViTPose: Simple vision transformer baselines for human pose estimation,

    Y . Xu, J. Zhang, Q. Zhang, and D. Tao, “ViTPose: Simple vision transformer baselines for human pose estimation,” inAdvances in Neural Information Processing Systems, 2022

    work page 2022

  60. [61]

    Masked au- toencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. Girshick, “Masked au- toencoders are scalable vision learners,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 16 000–16 009

    work page 2022

  61. [62]

    Deepfake detection algorithm based on improved vision transformer,

    Y .-J. Heo, W.-H. Yeo, and B.-G. Kim, “Deepfake detection algorithm based on improved vision transformer,”Applied Intelligence, vol. 53, no. 7, pp. 7512–7527, 2023

    work page 2023

  62. [63]

    Deepfake video detection using convolutional vision transformer,

    D. Wodajo and S. Atnafu, “Deepfake video detection using convolutional vision transformer,”arXiv preprint arXiv:2102.11126, 2021

    work page arXiv 2021

  63. [64]

    Video transformer for deepfake detection with incremental learning,

    S. A. Khan and H. Dai, “Video transformer for deepfake detection with incremental learning,” inProceedings of the 29th ACM international conference on multimedia, 2021, pp. 1821–1828

    work page 2021

  64. [65]

    M2tr: Multi-modal multi-scale transformers for deepfake detection,

    J. Wang, Z. Wu, W. Ouyang, X. Han, J. Chen, Y .-G. Jiang, and S.-N. Li, “M2tr: Multi-modal multi-scale transformers for deepfake detection,” inProceedings of the 2022 international conference on multimedia retrieval, 2022, pp. 615–623

    work page 2022

  65. [66]

    Protecting celebrities from deepfake with identity consistency transformer,

    X. Dong, J. Bao, D. Chen, T. Zhang, W. Zhang, N. Yu, D. Chen, F. Wen, and B. Guo, “Protecting celebrities from deepfake with identity consistency transformer,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 9468–9478

    work page 2022

  66. [67]

    Efficient deepfake detection using shallow vision transformer,

    S. Usmani, S. Kumar, and D. Sadhya, “Efficient deepfake detection using shallow vision transformer,”Multimedia Tools and Applications, vol. 83, no. 4, pp. 12 339–12 362, 2024

    work page 2024

  67. [68]

    Tall: Thumbnail layout for deepfake video detection,

    Y . Xu, J. Liang, G. Jia, Z. Yang, Y . Zhang, and R. He, “Tall: Thumbnail layout for deepfake video detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 22 658–22 668

    work page 2023

  68. [69]

    Self-supervised transformer for deepfake detection,

    H. Zhao, W. Zhou, D. Chen, W. Zhang, and N. Yu, “Self-supervised transformer for deepfake detection,”arXiv preprint arXiv:2203.01265, 2022

    work page arXiv 2022

  69. [70]

    Istvt: interpretable spatial-temporal video transformer for deepfake detection,

    C. Zhao, C. Wang, G. Hu, H. Chen, C. Liu, and J. Tang, “Istvt: interpretable spatial-temporal video transformer for deepfake detection,” IEEE Transactions on Information Forensics and Security, vol. 18, pp. 1335–1348, 2023

    work page 2023

  70. [71]

    Generalizing deepfake video detection with plug-and-play: Video-level blending and spatiotemporal adapter tuning,

    Z. Yan, Y . Zhao, S. Chen, M. Guo, X. Fu, T. Yao, S. Ding, Y . Wu, and L. Yuan, “Generalizing deepfake video detection with plug-and-play: Video-level blending and spatiotemporal adapter tuning,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 12 615–12 625

    work page 2025

  71. [72]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PmLR, 2021, pp. 8748–8763

    work page 2021

  72. [73]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

    work page 2022

  73. [74]

    Ms-celeb-1m: A dataset and benchmark for large-scale face recognition,

    Y . Guo, L. Zhang, Y . Hu, X. He, and J. Gao, “Ms-celeb-1m: A dataset and benchmark for large-scale face recognition,” in European Conference on Computer Vision, 2016. [Online]. Available: https://api.semanticscholar.org/CorpusID:2908606

    work page 2016

  74. [75]

    Delving into the local: Dynamic inconsistency learning for deepfake video detection,

    Z. Gu, Y . Chen, T. Yao, S. Ding, J. Li, and L. Ma, “Delving into the local: Dynamic inconsistency learning for deepfake video detection,” in Proceedings of the AAAI conference on artificial intelligence, vol. 36, no. 1, 2022, pp. 744–752

    work page 2022

  75. [76]

    Exposing deepfake videos by detecting face warping artifacts,

    Y . Li and S. Lyu, “Exposing deepfake videos by detecting face warping artifacts,” inIEEE Conference on Computer Vision and Pattern Recog- nition Workshops (CVPRW), 2019

    work page 2019

  76. [77]

    Cornernet: Detecting objects as paired keypoints,

    H. Law and J. Deng, “Cornernet: Detecting objects as paired keypoints,” International Journal of Computer Vision, vol. 128, pp. 642–656, 2018

    work page 2018

  77. [78]

    https://doi.org/10.48550/arXiv.1708.02002

    T. Lin, P. Goyal, R. B. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,”CoRR, vol. abs/1708.02002, 2017. [Online]. Available: http://arxiv.org/abs/1708.02002

    work page arXiv 2017

  78. [79]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117–2125

    work page 2017

  79. [80]

    FCOS: fully convolutional one-stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “FCOS: fully convolutional one-stage object detection,”CoRR, vol. abs/1904.01355, 2019. [Online]. Available: http://arxiv.org/abs/1904.01355

    work page arXiv 1904

  80. [81]

    Feature pyramid network for multi-class land segmentation,

    S. S. Seferbekov, V . I. Iglovikov, A. V . Buslaev, and A. A. Shvets, “Feature pyramid network for multi-class land segmentation,”CoRR, vol. abs/1806.03510, 2018. [Online]. Available: http://arxiv.org/abs/ 1806.03510

    work page arXiv 2018

Showing first 80 references.