pith. sign in

arxiv: 2606.19184 · v1 · pith:FJ7RSHQCnew · submitted 2026-06-17 · 💻 cs.CV · cs.LG

When AUC Misleads: Polarization-Aware Evaluation of Deepfake Detectors under Domain Shift

Dat Nguyen , Cosmin Radoi , Romain Hermary , Marcella Astrid , Nesryne Mejri , Enjie Ghorbel , Djamila Aouada This is my paper

Pith reviewed 2026-06-26 21:06 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords deepfake detectiondomain shiftAUC evaluationWasserstein distanceprediction polarizationgeneralizationevaluation metric
0
0 comments X

The pith

Cross-AUC evaluates deepfake detectors by averaging per-domain performance with polarization measured via Wasserstein distance between score distributions.

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

Standard AUC scores computed separately on each dataset overlook how detectors behave when inputs arrive from multiple sources at once. The paper proposes Cross-AUC to correct this by combining those per-domain AUC values with a term that captures how strongly the detector separates real from fake scores. Polarization is quantified as the Wasserstein distance between the two class-conditional score distributions. The resulting number is claimed to reflect generalization under domain shift more closely than isolated AUCs while also indicating whether a performance drop stems from weaker separation or from loss of polarization. Readers would care if this single number better predicts behavior on the mixed data that appears in practice.

Core claim

Cross-dataset AUC (Cross-AUC) averages per-domain AUCs with a measure of prediction polarization quantified by the Wasserstein Distance between class score distributions. This metric assesses generalization under domain shifts more realistically than separate AUCs and explains performance drops by showing whether low scores arise from reduced polarization.

What carries the argument

Cross-AUC, which combines per-domain AUCs with Wasserstein distance between class-conditional score distributions to quantify prediction polarization.

If this is right

  • A single number now ranks detectors on their joint ability to keep high AUC and high polarization across domains.
  • Performance drops can be attributed specifically to loss of polarization rather than to AUC alone.
  • Evaluation protocols can move from reporting many separate AUC tables to one Cross-AUC value per detector.
  • Detectors that preserve polarized score distributions across shifts receive higher robustness credit.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same polarization term could be added to evaluation in other binary detection tasks that face domain shift.
  • Training objectives might be extended to maximize both accuracy and the Wasserstein separation between classes.
  • Re-ranking published detectors with Cross-AUC could change which methods are viewed as most generalizable.

Load-bearing premise

That weighting or combining per-domain AUCs with Wasserstein distance between class-conditional score distributions yields a more realistic and interpretable measure of robustness to domain shift than reporting separate AUCs.

What would settle it

A detector that ranks higher than competitors on Cross-AUC yet performs worse than them when deployed on a real mixture of the same domains.

Figures

Figures reproduced from arXiv: 2606.19184 by Cosmin Radoi, Dat Nguyen, Djamila Aouada, Enjie Ghorbel, Marcella Astrid, Nesryne Mejri, Romain Hermary.

Figure 1
Figure 1. Figure 1: Example of optimal thresholds for predictions on different datasets, A (top) and B (middle). While a perfect AUC = 1 is reached on both datasets separately, an obvious drop in AUC will appear when combining them to simulate real-world settings (bottom), since the scores are no longer well ordered. The different nature of score distributions leads to a clear disparity between optimal thresholds (purple rang… view at source ↗
Figure 3
Figure 3. Figure 3: Correlation between estimated po￾larity (averaged across datasets) and the target combined AUC. Pearson coefficient ρ = 0.96. DFDC [10], and DFW [43]. This divergence can result from differences in deep￾fake generation methods, compression levels, facial attributes, resolution, or lighting conditions across datasets. This can lead to inconsistencies in how the model scores samples across domains, as we can… view at source ↗
Figure 4
Figure 4. Figure 4: Example of the obtained optimal thresholds on two datasets, A (top) and B (middle). Although both detectors achieve perfect AUC = 1 when evaluated on each dataset separately, mixing A&B (bottom) can cause a sharp AUC drop (Detector X), while a more stable detector preserves ordering and maintains AUC = 1 (Detector Y). It can be observed that a larger gap exists between the two classes in the score space, r… view at source ↗
Figure 5
Figure 5. Figure 5: Class-conditional score distributions of the evaluated deepfake detectors across seven datasets [10, 11, 15, 24, 31, 37, 43]. Cross-AUC versus Average AUC [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Methods results (AUC) for each dataset. Average AUC, combined AUC, and the proposed Cross-AUC are also re￾ported. SBI SLADD LAA-Net CADDM RECCE ForensicsAdapter Xception 40 50 60 70 80 90 100 AUC (Average) AUC (Combined) Cross-AUC [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
read the original abstract

Recent advances in generative AI, such as diffusion models and face-swapping tools, have enabled the creation of highly realistic deepfakes, leading to real-world harms including financial fraud and non-consensual explicit content. In response, deepfake detection has become an active research area, with recent methods increasingly focusing on improving generalization to unseen manipulations. This is typically evaluated using the Area Under the ROC Curve (AUC) measured separately across multiple datasets. However, such an evaluation fails to reflect real-world scenarios where detectors face a mixture of data sources and varying artifact types. To address this limitation, we introduce a novel metric, Cross-dataset AUC (Cross-AUC) that averages per-domain AUCs with a measure of prediction polarization for taking into account the robustness to domain shift. The polarization extent is quantified by the Wasserstein Distance between class score distributions. Cross-AUC not only assesses the generalization capabilities of deepfake detectors under domain shifts more realistically, but it is also interpretable as it better explains the reason behind a drop in performance. Experiments performed on seven benchmark datasets demonstrate its practical relevance.

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 / 1 minor

Summary. The paper argues that per-dataset AUC evaluation for deepfake detectors does not capture real-world mixed-domain scenarios, and proposes Cross-AUC as a new metric that averages per-domain AUCs together with a polarization term given by the Wasserstein distance between class-conditional score distributions; the authors claim this yields a more realistic assessment of generalization under domain shift and better explains performance drops, with supporting experiments on seven benchmark datasets.

Significance. If the central claim holds, Cross-AUC could supply a practically useful, interpretable alternative to separate AUC reporting for evaluating robustness to domain shift in deepfake detection, potentially improving model selection and diagnosis of failure modes.

major comments (2)
  1. [Metric definition] The definition of Cross-AUC (presumably in the section introducing the metric) combines per-domain AUCs with Wasserstein distance on score distributions, but supplies no derivation or comparative analysis showing why this particular aggregation (rather than, e.g., a simple mean, a min, or a variance-weighted combination) aligns with real-world mixed-domain performance.
  2. [Experiments] The experimental section claims that results on seven datasets demonstrate practical relevance, yet contains no direct comparison of Cross-AUC against detector performance on a held-out mixed-domain test set whose domain mixture matches the claimed real-world scenario; without this, it remains possible that Cross-AUC merely re-expresses information already present in the separate AUCs.
minor comments (1)
  1. [Metric definition] Notation for the polarization term and the exact weighting in the Cross-AUC formula should be stated explicitly with an equation number.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to incorporate the suggested improvements.

read point-by-point responses

  1. Referee: [Metric definition] The definition of Cross-AUC (presumably in the section introducing the metric) combines per-domain AUCs with Wasserstein distance on score distributions, but supplies no derivation or comparative analysis showing why this particular aggregation (rather than, e.g., a simple mean, a min, or a variance-weighted combination) aligns with real-world mixed-domain performance.

    Authors: We agree that a formal derivation and comparative analysis are missing and would strengthen the paper. The aggregation (average AUC plus Wasserstein polarization term) is motivated by the goal of rewarding both high per-domain accuracy and score separation under shift, as overlapping distributions indicate poor robustness in mixed settings. In revision we will add a subsection deriving the metric from the expected performance under a domain mixture model and include explicit comparisons to alternatives such as min-AUC and variance-weighted means. revision: yes

  2. Referee: [Experiments] The experimental section claims that results on seven datasets demonstrate practical relevance, yet contains no direct comparison of Cross-AUC against detector performance on a held-out mixed-domain test set whose domain mixture matches the claimed real-world scenario; without this, it remains possible that Cross-AUC merely re-expresses information already present in the separate AUCs.

    Authors: We acknowledge the absence of this direct validation. While the existing results on seven datasets illustrate that the polarization term explains performance variation beyond per-domain AUCs alone, a controlled comparison on constructed mixed-domain test sets is needed to confirm alignment with real-world behavior. We will add such experiments by forming held-out mixed test sets from the benchmarks and report the correlation between Cross-AUC and observed mixed-domain AUC. revision: yes

Circularity Check

0 steps flagged

Cross-AUC introduced as explicit definition from standard components; no reduction to inputs

full rationale

The paper defines Cross-AUC directly as the average of per-domain AUCs combined with a polarization term given by the Wasserstein distance between class-conditional score distributions. No derivation chain is presented that reduces this quantity to itself by construction, no parameters are fitted on a subset and then relabeled as predictions, and no self-citation or uniqueness theorem is invoked to justify the aggregation choice. The metric is therefore a new definitional proposal whose content is independent of its own outputs; experiments on seven datasets are offered as external validation rather than internal tautology. This matches the default expectation of no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The proposal rests on the standard use of AUC for binary classification and the Wasserstein distance as a measure of distribution separation, with the domain assumption that their combination captures robustness under shift.

axioms (1)
  • domain assumption Wasserstein distance between class score distributions quantifies prediction polarization relevant to domain-shift robustness.
    Invoked to adjust the averaged AUC in the definition of Cross-AUC.

pith-pipeline@v0.9.1-grok · 5750 in / 1279 out tokens · 29391 ms · 2026-06-26T21:06:55.644541+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

43 extracted references · 15 canonical work pages · 7 internal anchors

  1. [1]

    In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR)

    Bai, W., Liu, Y., Zhang, Z., Li, B., Hu, W.: Aunet: Learning relations between action units for face forgery detection. In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR). pp. 24709–24719 (June 2023)

    2023

  2. [2]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Cao, J., Ma, C., Yao, T., Chen, S., Ding, S., Yang, X.: End-to-end reconstruction- classification learning for face forgery detection. In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4103–4112 (2022).https: //doi.org/10.1109/CVPR52688.2022.00408

    work page doi:10.1109/cvpr52688.2022.00408 2022

  3. [3]

    chief financial officer

    Chen, H., Magramo, K.: Finance worker pays out $25 million after video call with deepfake "chief financial officer".https://edition.cnn.com/2024/02/04/asia/ deepfake-cfo-scam-hong-kong-intl-hnk/index.html(2024), [Online; accessed 4-February-2024]

    2024

  4. [4]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Chen, L., Zhang, Y., Song, Y., Liu, L., Wang, J.: Self-supervised learning of adver- sarial example: Towards good generalizations for deepfake detection. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 18710–18719 (June 2022)

    2022

  5. [5]

    Advances in Neural Information Processing Systems35, 24597–24610 (2022)

    Chen, L., Zhang, Y., Song, Y., Wang, J., Liu, L.: Ost: Improving generalization of deepfake detection via one-shot test-time training. Advances in Neural Information Processing Systems35, 24597–24610 (2022)

    2022

  6. [6]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Chollet, F.: Xception: Deep learning with depthwise separable convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1251–1258 (2017)

    2017

  7. [7]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Cui, X., Li, Y., Luo, A., Zhou, J., Dong, J.: Forensics adapter: Adapting clip for generalizable face forgery detection. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 19207–19217 (2025)

    2025

  8. [8]

    Deepfakes: Faceswapdevs.https://github.com/deepfakes/faceswap(2019)

    2019

  9. [9]

    RetinaFace: Single-stage Dense Face Localisation in the Wild

    Deng, J., Guo, J., Zhou, Y., Yu, J., Kotsia, I., Zafeiriou, S.: Retinaface: Single-stage dense face localisation in the wild. CoRRabs/1905.00641(2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  10. [10]

    The DeepFake Detection Challenge (DFDC) Dataset

    Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R., Wang, M., Canton-Ferrer, C.: The deepfake detection challenge dataset. CoRRabs/2006.07397(2020), https://arxiv.org/abs/2006.07397

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  11. [11]

    CoRRabs/1910.08854(2019), http://arxiv.org/abs/1910.08854 16 D

    Dolhansky, B., Howes, R., Pflaum, B., Baram, N., Canton-Ferrer, C.: The deep- fake detection challenge (DFDC) preview dataset. CoRRabs/1910.08854(2019), http://arxiv.org/abs/1910.08854 16 D. NGUYEN et al

    work page arXiv 1910

  12. [12]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

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

    2023

  13. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Dong, X., Bao, J., Chen, D., Zhang, T., Zhang, W., Yu, N., Chen, D., Wen, F., Guo,B.:Protectingcelebritiesfromdeepfakewithidentityconsistencytransformer. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 9468–9478 (June 2022)

    2022

  14. [14]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  15. [15]

    html(2019)

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

    2019

  16. [16]

    In: European conference on computer vision

    Fu, B., Cao, Z., Long, M., Wang, J.: Learning to detect open classes for universal domain adaptation. In: European conference on computer vision. pp. 567–583. Springer (2020)

    2020

  17. [17]

    Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples (2015)

    2015

  18. [18]

    IEEE Transactions on Knowledge and Data Engineering17(3), 299–310 (2005)

    Huang, J., Ling, C.: Using auc and accuracy in evaluating learning algorithms. IEEE Transactions on Knowledge and Data Engineering17(3), 299–310 (2005). https://doi.org/10.1109/TKDE.2005.50

    work page doi:10.1109/tkde.2005.50 2005

  19. [19]

    Kowalski, M.: Faceswap.https://github.com/MarekKowalski/FaceSwap(2018)

    2018

  20. [20]

    In: Globerson, A., Mackey, L., Bel- grave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Kumagai, A., Iwata, T., Takahashi, H., Nishiyama, T., Fujiwara, Y.: Auc max- imization under positive distribution shift. In: Globerson, A., Mackey, L., Bel- grave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (eds.) Advances in Neu- ral Information Processing Systems. vol. 37, pp. 36071–36096. Curran Associates, Inc. (2024),https://proceedings.neurips...

    2024

  21. [21]

    Adversarial Machine Learning at Scale

    Kurakin, A., Goodfellow, I.J., Bengio, S.: Adversarial machine learning at scale. ArXivabs/1611.01236(2016),https://api.semanticscholar.org/CorpusID: 9059612

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  22. [22]

    Proceedings ofthe2023ACMConferenceonFairness,Accountability,andTransparency(2023), https://api.semanticscholar.org/CorpusID:258960540

    Kwegyir-Aggrey,K.,Gerchick,M.,Mohan,M.,Horowitz,A.,Venkatasubramanian, S.: The misuse of auc: What high impact risk assessment gets wrong. Proceedings ofthe2023ACMConferenceonFairness,Accountability,andTransparency(2023), https://api.semanticscholar.org/CorpusID:258960540

    2023

  23. [23]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (June 2020)

    Li,L.,Bao,J.,Zhang,T.,Yang,H.,Chen,D.,Wen,F.,Guo,B.:Facex-rayformore general face forgery detection. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (June 2020)

    2020

  24. [24]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (June 2020)

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

    2020

  25. [25]

    2024 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR) pp

    Lin, L., He, X., Ju, Y., Wang, X., Ding, F., Hu, S.: Preserving fairness gener- alization in deepfake detection. 2024 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR) pp. 16815–16825 (2024),https://api. semanticscholar.org/CorpusID:268031845

    2024

  26. [26]

    arXiv preprint arXiv:2410.21964 (2024) Polarization-Aware Evaluation of Deepfake Detectors under Domain Shift 17

    Nguyen, D., Astrid, M., Ghorbel, E., Aouada, D.: Fakeformer: Efficient vulnerability-driven transformers for generalisable deepfake detection. arXiv preprint arXiv:2410.21964 (2024) Polarization-Aware Evaluation of Deepfake Detectors under Domain Shift 17

    work page arXiv 2024

  27. [27]

    Nguyen, D., Astrid, M., Kacem, A., Ghorbel, E., Aouada, D.: Vulnerability-aware spatio-temporallearningforgeneralizabledeepfakevideodetection.In:Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 10786–10796 (2025)

    2025

  28. [28]

    Nguyen, D., Mejri, N., Singh, I.P., Kuleshova, P., Astrid, M., Kacem, A., Ghorbel, E., Aouada, D.: Laa-net: Localized artifact attention network for quality-agnostic andgeneralizabledeepfakedetection.In:ProceedingsoftheIEEE/CVFConference on Computer Vision and Pattern Recognition (CVPR). pp. 17395–17405 (June 2024)

    2024

  29. [29]

    In: International Confer- ence on Machine Learning (2021),https://api.semanticscholar.org/CorpusID: 231591445

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning trans- ferable visual models from natural language supervision. In: International Confer- ence on Machine Learning (2021),https://api.semanticscholar.org/CorpusID: 231591445

    2021

  30. [30]

    Reuters: South korea to criminalize watching or possessing sexually explicit deep- fakes.https://edition.cnn.com/2024/09/26/asia/south- korea- deepfake- bill-passed-intl-hnk/index.html(2024), [Online; accessed 26-September-2024]

    2024

  31. [31]

    In: International Con- ference on Computer Vision (ICCV) (2019)

    Rössler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., Nießner, M.: Face- Forensics++: Learning to detect manipulated facial images. In: International Con- ference on Computer Vision (ICCV) (2019)

    2019

  32. [32]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion

    Shiohara, K., Yamasaki, T.: Detecting deepfakes with self-blended images. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. pp. 18720–18729 (2022)

    2022

  33. [33]

    EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

    Tan, M., Le, Q.V.: Efficientnet: Rethinking model scaling for convolutional neural networks. CoRRabs/1905.11946(2019),http://arxiv.org/abs/1905.11946

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  34. [34]

    Deferred Neural Rendering: Image Synthesis using Neural Textures

    Thies, J., Zollhöfer, M., Nießner, M.: Deferred neural rendering: Image synthesis using neural textures. CoRRabs/1904.12356(2019),http://arxiv.org/abs/ 1904.12356

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  35. [35]

    CoRRabs/2007.14808 (2020),https://arxiv.org/abs/2007.14808

    Thies, J., Zollhöfer, M., Stamminger, M., Theobalt, C., Nießner, M.: Face2face: Real-time face capture and reenactment of RGB videos. CoRRabs/2007.14808 (2020),https://arxiv.org/abs/2007.14808

    work page arXiv 2007

  36. [36]

    Partial AUC Maximization via Nonlinear Scoring Functions

    Ueda, N., Fujino, A.: Partial auc maximization via nonlinear scoring functions. arXiv preprint arXiv:1806.04838 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  37. [37]

    Advances in Neural Information Processing Systems37, 29387–29434 (2024)

    Yan, Z., Yao, T., Chen, S., Zhao, Y., Fu, X., Zhu, J., Luo, D., Wang, C., Ding, S., Wu, Y., et al.: Df40: Toward next-generation deepfake detection. Advances in Neural Information Processing Systems37, 29387–29434 (2024)

    2024

  38. [38]

    arXiv preprint arXiv:2408.17065 (2024)

    Yan, Z., Zhao, Y., Chen, S., Guo, M., Fu, X., Yao, T., Ding, S., Yuan, L.: Gen- eralizing deepfake video detection with plug-and-play: Video-level blending and spatiotemporal adapter tuning. arXiv preprint arXiv:2408.17065 (2024)

    work page arXiv 2024

  39. [39]

    Advances in Neural Information Pro- cessing Systems36, 31428–31449 (2023)

    Zhang, K., Mo, L., Chen, W., Sun, H., Su, Y.: Magicbrush: A manually annotated dataset for instruction-guided image editing. Advances in Neural Information Pro- cessing Systems36, 31428–31449 (2023)

    2023

  40. [40]

    ArXivabs/1912.11188(2019),https://api.semanticscholar.org/CorpusID: 209460718

    Zhang, X., Wang, Q., Zhang, J., Zhong, Z.: Adversarial autoaugment. ArXivabs/1912.11188(2019),https://api.semanticscholar.org/CorpusID: 209460718

    work page arXiv 1912

  41. [41]

    In: ICCV 2021 (2021),https://www.amazon.science/ publications/learning-self-consistency-for-deepfake-detection

    Zhao, E., Xu, X., Xu, M., Ding, H., Xiong, Y., Xia, W.: Learning self-consistency for deepfake detection. In: ICCV 2021 (2021),https://www.amazon.science/ publications/learning-self-consistency-for-deepfake-detection

    2021

  42. [42]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhao, H., Zhou, W., Chen, D., Wei, T., Zhang, W., Yu, N.: Multi-attentional deepfake detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 2185–2194 (2021) 18 D. NGUYEN et al

    2021

  43. [43]

    Combined

    Zi, B., Chang, M., Chen, J., Ma, X., Jiang, Y.G.: Wilddeepfake: A challenging real- world dataset for deepfake detection. Proceedings of the 28th ACM International Conference on Multimedia (2020) Polarization-Aware Evaluation of Deepfake Detectors under Domain Shift 19 6 Appendix Our technical appendix is organized as follows. Section 6.1 presents addi- t...

    work page arXiv 2020