Pith. sign in

REVIEW 4 major objections 5 minor 53 references

Robust Deepfake Detection for Electronic Know Your Customer Systems Using Registered Images

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper claims that a deepfake detector for eKYC systems can catch both face-swapping and face-reenactment attacks by comparing face-recognition identity vectors across video frames and against a registered photo, while staying accurate…

desk verdict Useful encoder-scaling results and a clean ablation, but the eKYC deployment claim is untested because the auxiliary image is never an ID photo. read the letter →

arxiv 2507.22601 v1 pith:4ULSB37U submitted 2025-07-30 cs.CV

classification cs.CV
keywords deepfakedetectioneKYCfaceswappingreenactmentidentityvectorsregisteredimagetemporalinconsistencydegradationrobustness
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 claims that a deepfake detector built for electronic Know Your Customer (eKYC) systems can flag both face-swapping and face-reenactment attacks by checking identity in two ways at once: how much a person's face-recognition identity vector jumps between consecutive video frames, and how far those frame vectors sit from a registered reference photo of the same claimed person. The authors show that the second check, which they call Auxiliary Difference Comparison, is what gives the method an edge on reenactment attacks, while the first check handles swapping, and that the two together beat the temporal-identity baseline on both in-dataset and cross-dataset tests. They also report that swapping the face encoder for one trained with AdaFace loss on the WebFace12M dataset raises in-dataset AUC from roughly 81% to 95% and sharply reduces accuracy loss under blur, noise, and compression. If the claim holds, eKYC systems already possess the extra signal they need—the enrolled identity photo—and the main remaining design choice is which face recognition backbone to use.

What carries the argument

The carrying object is the identity vector: a fixed-dimension embedding of a face produced by a face recognition network. The detector feeds a bidirectional GRU with two concatenated difference sequences—temporal differences between consecutive frames and frame-by-frame differences against a registered reference image—and the GRU's final hidden state is classified by two fully connected layers. The second carrying element is the feature extractor itself: a ResNet100 trained with AdaFace loss on WebFace12M, whose identity vectors remain stable when input frames are blurred, noisy, compressed, or altered in saturation and contrast.

What would settle it

Give the trained detector a real-user video paired with a genuine ID document photo taken under office lighting, with print texture and a different camera, and measure the video-level AUC on real users: if the ADC stream pushes genuine users' scores closer to the fake distribution and AUC collapses, the registered-image contribution to eKYC robustness is not as claimed.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that identity-vector dynamics, when measured as differences rather than absolute embeddings, are a sufficient and unusually degradation-tolerant basis for eKYC deepfake detection. The method computes two parallel difference sequences from a face recognition extractor: temporal differences between successive frames (TDC) and per-frame differences against a registered genuine image (ADC), concatenates them, and classifies the sequence with a bidirectional GRU plus a two-layer head. In KoDF in-dataset evaluation the full model reaches 95.70% video-level AUC across all fake types, ahead of both the TDC-only TI2Net baseline and the frame-level SBI baseline, and the paper reports a 21.5-point advantage over TI2Net in cross-dataset evaluation. The paper additionally demonstrates, by comparing three face encoders, that the accuracy and corruption robustness of the detector track the scale and quality of the face recognition training data, with the AdaFace/WebFace12M encoder giving both the best detection and the smallest AUC declines under six types of image degradation.

Load-bearing premise

The registered image really is the same person as the video and is similar enough in pose, lighting, and capture conditions that a genuine user's identity-vector differences stay close to zero; the paper only tests this with frames from other videos of the same identity, not with actual ID document photographs.

Editorial extensions

If this is right

  • eKYC systems can use the enrolled identity photo as an auxiliary signal without adding a new capture step, and the ADC branch gives the detector a specific advantage on face-reenactment attacks.
  • Because detection accuracy and robustness tracked the face encoder's training scale in the paper's comparisons, the method will likely continue to improve as face recognition models are trained on larger datasets.
  • The small AUC declines under six degradation types imply identity-vector-based detection is suitable for real eKYC video, where blur, noise, compression, and block distortion are common.
  • Cross-dataset results on Celeb-DF v2, DFD, and DFDCp indicate the detector generalizes to manipulation methods and identities not seen in training, and its DFDCp result shows particular strength on post-processed, low-quality videos.

Reading between the lines

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

  • A testable extension would swap the auxiliary image source from a same-identity video frame to an actual ID document scan or photo; if the capture-condition gap is large, ADC may need a calibration step or a learned tolerance before it helps in deployed eKYC.
  • The success of ADC suggests a broader design principle: any system that has a trusted reference image of the claimed identity can use identity-vector distance to that reference as an attack signal, which could apply to liveness checks, speaker verification, or document-holder matching.
  • Because the paper shows a scaling trend from ArcFace/MS1MV2 to AdaFace/MS1MV2 to AdaFace/WebFace12M, a natural next experiment is to measure whether even larger face recognition backbones continue to raise both detection AUC and degradation robustness, and whether gains saturate.
  • The random-sampling result hints that a simpler, order-agnostic aggregation of identity differences could match the GRU's accuracy at lower latency, which would be worth testing for real-time eKYC liveness.
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 / 5 minor

Summary. The paper proposes a video-level deepfake detector for eKYC that combines Temporal Difference Comparison (TDC) of identity vectors across frames with Auxiliary Difference Comparison (ADC) against a registered image, feeding both into an RNN with a classification head; the feature extractor is a pretrained face recognition model, and the authors evaluate the effect of extractor scale. Experiments are performed on KoDF for training and in-dataset testing, with cross-dataset evaluation on Celeb-DF v2, DFD, and DFDCp, plus robustness tests under six image degradations. The central claimed contributions are comprehensive detection of face swapping and face reenactment, improved accuracy from the registered image, and robustness to unseen degradation, with source code publicly released.

Significance. If the claims hold, the paper makes a practical contribution to eKYC deepfake defense by showing that an enrollment image can be leveraged for identity-consistency detection, and by demonstrating a positive correlation between face-recognition feature quality and deepfake-detection robustness. The systematic robustness evaluation across six corruptions, the feature-extractor ablation, and the release of source code are genuine strengths. However, the eKYC-specific conclusion depends on an auxiliary-image assumption that the experiments do not actually test, and the headline quantitative claims contain an inconsistency that must be resolved.

major comments (4)
  1. [Section IV-A1 and Eq. (1)] The eKYC premise stated in Section IV-A1 is that the input video is captured in real-time while the auxiliary image is a photo printed on an identity document, and that these 'often differ visually.' Yet in every experiment the auxiliary image is sampled from a different video of the same identity within the same dataset, so the ADC branch is never exposed to the document-photo-to-video domain gap that motivates the method. Robustness experiments corrupt only the input video, not the auxiliary image. To support the eKYC claim, please report experiments where the auxiliary image comes from a genuinely different capture domain (e.g., document scans, lower resolution, different lighting/pose), or at least analyze the distribution of ADC identity-vector distances for genuine users under such mismatch; without this, the central deployment claim is not validated.
  2. [Section I and Table III] The introduction states that the method outperforms TI2Net by 1.5 and 21.5 percentage points in in-dataset and cross-dataset evaluation, respectively. Table III, however, shows per-dataset gaps of 11.72 points on Celeb-DF v2, 11.66 points on DFD, and 19.21 points on DFDCp, and the average gap is about 14.2 points. The 21.5-point figure is not reproducible from the reported tables; please correct the statement or specify the exact computation used.
  3. [Section IV-A3 and Section IV-D] The face feature extractor is pretrained on WebFace12M, a large web-scraped dataset, and the paper does not address the possibility that WebFace12M contains identities that overlap with KoDF, Celeb-DF v2, DFD, or DFDCp evaluation subjects. Because the method's representation is explicitly identity-based, unseen-identity generalization is central to the cross-dataset claim. Please report whether overlap checks were performed, or temper the generalization claim accordingly.
  4. [Section IV-A3 and Table II] The headline in-dataset comparison is based on a single random split of KoDF and one random video/feature sampling, with no error bars or repeated runs. The reported 'all' AUC of 95.70 is only 0.3 points above SBI (95.40) and 1.5 points above TI2Net (94.20), so without variance estimates it is not clear that the proposed method's advantage is statistically meaningful. Please provide multiple seeds with mean and standard deviation, or otherwise report per-run values.
minor comments (5)
  1. [Table VII] The column header 'DFC' should be 'DFD' to match the dataset name used elsewhere in the paper.
  2. [Section IV-A3] The text says 'We randomly sample 403 real videos and 850 fake videos from the KoDF dataset,' but KoDF contains tens of thousands of videos; please clarify whether this sampling is done before or after the train/validation/test split and whether the same videos are reused for validation and test.
  3. [Figures 3-6] The robustness figures are difficult to read at the plotted scale; consider providing a supplementary table with the numeric AUC values for each method, corruption type, and severity level.
  4. [References] References [10], [11], and [13] appear to refer to the same eKYC deepfake papers with overlapping titles; please consolidate or disambiguate them.
  5. [Section V] The statement that 'ADC alone surpasses competing methods on four of six datasets' is not directly supported by Table V, which only compares embedding types and does not include competing methods; please clarify which table and which comparison is intended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is an empirical combination of TDC, ADC, and a pretrained face encoder, with cross-dataset and robustness results evaluated against external benchmarks.

full rationale

This paper is an empirical method paper; there is no derivation chain that claims to predict a quantity from first principles. TDC is explicitly re-used from TI2Net [32] with attribution, and the TDC-only ablation fairly reproduces TI2Net; ADC is a new difference feature against a registered image. The robustness and cross-dataset results are measured against external benchmarks (KoDF train, CDF/DFD/DFDCp test) and compared with published baselines, so no fitted parameter is renamed as a prediction. The only self-citations are SBI [41] and BlendFace [42] used as baseline or related work; these citations are not load-bearing for the central claim. The manuscript's own limitation in Section IV-A1 — that the input video and an enrolled ID photo 'often differ visually' while all experiments sample auxiliary images from same-domain videos — is a deployment-validity concern, not a circularity: the method's output is not defined as the test set's label, and the auxiliary image is not a fitted quantity. Score 0.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The method rests on several unproven operational premises: that the registered photo is genuine and paired with the video's claimed identity, that identity vectors are stable across benign capture conditions, and that same-domain auxiliary videos approximate ID-document photos. These are domain assumptions rather than formal axioms. The only hand-chosen numerical inputs are loss weights, margins, sequence lengths, and training-sample sizes.

free parameters (8)
  • lambda_1 (triplet loss weight) = not reported
    Hyperparameter in Eq. 7 controlling triplet loss contribution; chosen by hand.
  • lambda_2 (anchor-positive loss weight) = not reported
    Hyperparameter in Eq. 7 controlling anchor-positive regularizer; chosen by hand.
  • triplet margin alpha = not reported
    Margin in Eq. 4; positive constant chosen by hand.
  • RNN sequence length = 64
    Frames per video in main experiments; ablated in Section IV-G, and shorter lengths sometimes improve cross-dataset AUC.
  • training video sample sizes = 403 real, 850 fake
    Random subset of KoDF used for training; no justification for these counts.
  • RNN hidden dimension = 1024
    Bidirectional GRU hidden size; chosen without ablation.
  • dropout rates = 0.2 and 0.5
    Dropout before RNN and classification head.
  • initial learning rate = 0.0005
    Adam optimizer setting.
assumptions (5)
  • domain assumption Registered auxiliary image is authentic and corresponds to the identity claimed in the input video.
    Stated in Section III-B1; ADC compares video identity features to this image. If the registered image is compromised or a different person, the whole detection signal is invalid.
  • domain assumption Identity vectors from a face recognition model are stable across benign videos of the same person while being sensitive to deepfake artifacts.
    Central premise of both TDC and ADC; not proven in the paper. Real videos can have large identity-vector shifts due to pose, lighting, and camera quality.
  • ad hoc to paper Auxiliary images sampled from other videos of the same identity in KoDF emulate the eKYC registered-photo setting.
    Section IV-A1 explicitly says the input video and auxiliary image differ visually, but no experiment uses actual ID document photos or a real eKYC enrollment corpus.
  • standard math MTCNN alignment and 112x112 cropping preserve enough identity information for the pretrained face recognition features.
    Standard preprocessing, but the interaction between alignment failures and degraded video is not analyzed.
  • domain assumption Pretrained feature extractors (AdaFace/WebFace12M) do not have significant identity overlap with evaluation subjects in KoDF, Celeb-DF, DFD, or DFDCp.
    Not audited in the paper. WebFace12M is a large web-scraped dataset that may contain celebrities who appear in Celeb-DF/DFD videos, which could inflate identity-vector discrimination.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Deepfake Detection for Electronic Know Your Customer Systems Using Registered Images." pith.science (2026). https://pith.science/paper/4ULSB37U

@misc{pith2026250722601,
  author       = {Pith},
  title        = {Pith review of: Robust Deepfake Detection for Electronic Know Your Customer Systems Using Registered Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4ULSB37U}},
  note         = {Machine review of arXiv:2507.22601}
}
read the original abstract

In this paper, we present a deepfake detection algorithm specifically designed for electronic Know Your Customer (eKYC) systems. To ensure the reliability of eKYC systems against deepfake attacks, it is essential to develop a robust deepfake detector capable of identifying both face swapping and face reenactment, while also being robust to image degradation. We address these challenges through three key contributions: (1)~Our approach evaluates the video's authenticity by detecting temporal inconsistencies in identity vectors extracted by face recognition models, leading to comprehensive detection of both face swapping and face reenactment. (2)~In addition to processing video input, the algorithm utilizes a registered image (assumed to be genuine) to calculate identity discrepancies between the input video and the registered image, significantly improving detection accuracy. (3)~We find that employing a face feature extractor trained on a larger dataset enhances both detection performance and robustness against image degradation. Our experimental results show that our proposed method accurately detects both face swapping and face reenactment comprehensively and is robust against various forms of unseen image degradation. Our source code is publicly available https://github.com/TaikiMiyagawa/DeepfakeDetection4eKYC.

Figures

Figures reproduced from arXiv: 2507.22601 by the authors.

Figure 1
Figure 1. Impersonation attack using a deepfake against an eKYC system. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Proposed deepfake detector. It consists of (1) Auxiliary Difference Comparison (ADC), (2) Temporal Difference Comparison (TDC), and (3) a [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. In-dataset evaluation of robustness against six types of image [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Robustness evaluation against six types of image degradation. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 46 canonical work pages

  1. [11]

    Do, M.-K

    T.-L. Do, M.-K. Tran, H. H. Nguyen, and M.-T. Tran. Potential threat of face swapping to ekyc with face registration and augmented solution with deepfake detection. In FDSE, pages 293–307, 2021

  2. [13]

    Do, M.-K

    T.-L. Do, M.-K. Tran, H. H. Nguyen, and M.-T. Tran. Potential attacks of deepfake on ekyc systems and remedy for ekyc with deepfake detection using two-stream network of facial appearance and motion features. SN Computer Science , 3(6):464, 2022

  3. [1]

    Afchar, V

    D. Afchar, V . Nozick, J. Yamagishi, and I. Echizen. MesoNet: a compact facial video forgery detection network. In WIFS, pages 1–7, 2018

  4. [2]

    Amerini, L

    I. Amerini, L. Galteri, R. Caldelli, and A. Del Bimbo. Deepfake video detection through optical flow based CNN. In ICCV workshops, 2019

  5. [3]

    Balntas, E

    V . Balntas, E. Riba, D. Ponsa, and K. Mikolajczyk. Learning local feature descriptors with triplets and shallow convolutional neural networks. In BMVC, volume 1, page 3, 2016

  6. [4]

    Bitouk, N

    D. Bitouk, N. Kumar, S. Dhillon, P. Belhumeur, and S. K. Nayar. Face swapping: automatically replacing faces in photographs. In ACM SIGGRAPH, pages 1–8. 2008

  7. [5]

    Blanz and T

    V . Blanz and T. Vetter. A morphable model for the synthesis of 3d faces. In ACM SIGGRAPH, pages 187–194. 1999

  8. [6]

    K. Cho, B. Van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio. Learning phrase representations using rnn encoder-decoder for statistical machine translation. arXiv:1406.1078, 2014

Show all 53 references
  1. [7]

    J. S. Chung, A. Nagrani, and A. Zisserman. V oxceleb2: Deep speaker recognition. arXiv:1806.05622, 2018

  2. [8]

    Cozzolino, A

    D. Cozzolino, A. R ¨ossler, J. Thies, M. Nießner, and L. Verdoliva. ID-reveal: Identity-aware deepfake video detection. In ICCV, pages 15108–15117, 2021

  3. [9]

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou. ArcFace: Additive angular margin loss for deep face recognition. In CVPR, pages 4690–4699, 2019

  4. [14]

    Dolhansky, R

    B. Dolhansky, R. Howes, B. Pflaum, N. Baram, and C. C. Fer- rer. The deepfake detection challenge (DFDC) preview dataset. arXiv:1910.08854, 2019

  5. [15]

    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. In CVPR, pages 9468–9478, 2022

  6. [16]

    Felouat, H

    H. Felouat, H. H. Nguyen, T.-N. Le, J. Yamagishi, and I. Echizen. ekyc-df: A large-scale deepfake dataset for developing and evaluating ekyc systems. IEEE Access, 2024

  7. [17]

    G ¨uera and E

    D. G ¨uera and E. J. Delp. Deepfake video detection using recurrent neural networks. In AVSS, pages 1–6, 2018

  8. [18]

    Haliassos, K

    A. Haliassos, K. V ougioukas, S. Petridis, and M. Pantic. Lips don’t lie: A generalisable and robust approach to face forgery detection. In CVPR, pages 5039–5049, 2021

  9. [19]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016

  10. [20]

    Huang, Z

    B. Huang, Z. Wang, J. Yang, J. Ai, Q. Zou, Q. Wang, and D. Ye. Implicit identity driven deepfake face swapping detection. In CVPR, pages 4490–4499, 2023

  11. [21]

    Jiang, R

    L. Jiang, R. Li, W. Wu, C. Qian, and C. C. Loy. DeeperForensics-1.0: A large-scale dataset for real-world face forgery detection. In CVPR, pages 2889–2898, 2020

  12. [22]

    N. D. Kalka, B. Maze, J. A. Duncan, K. O’Connor, S. Elliott, K. Hebert, J. Bryan, and A. K. Jain. Ijb-s: Iarpa janus surveillance video benchmark. In BTAS, pages 1–9, 2018

  13. [23]

    H. Kim, P. Garrido, A. Tewari, W. Xu, J. Thies, M. Niessner, P. P ´erez, C. Richardt, M. Zollh¨ofer, and C. Theobalt. Deep video portraits. ACM TOG, 37(4):1–14, 2018

  14. [24]

    M. Kim, A. K. Jain, and X. Liu. AdaFace: Quality adaptive margin for face recognition. In CVPR, pages 18750–18759, 2022

  15. [25]

    D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. arXiv:1412.6980, 2014

  16. [26]

    Korshunova, W

    I. Korshunova, W. Shi, J. Dambre, and L. Theis. Fast face-swap using convolutional neural networks. In ICCV, pages 3677–3685, 2017

  17. [27]

    P. Kwon, J. You, G. Nam, S. Park, and G. Chae. Kodf: A large- scale korean deepfake detection dataset. In ICCV, pages 10744–10753, 2021

  18. [28]

    L. Li, J. Bao, H. Yang, D. Chen, and F. Wen. FaceShifter: Towards high fidelity and occlusion aware face swapping. arXiv:1912.13457, 2019

  19. [29]

    L. Li, J. Bao, T. Zhang, H. Yang, D. Chen, F. Wen, and B. Guo. Face X-ray for more general face forgery detection. In CVPR, pages 5001–5010, 2020

  20. [30]

    Y . Li, C. Ma, Y . Yan, W. Zhu, and X. Yang. 3D-aware face swapping. In CVPR, pages 12705–12714, 2023

  21. [31]

    Y . Li, X. Yang, P. Sun, H. Qi, and S. Lyu. Celeb-DF: A large-scale challenging dataset for deepfake forensics. In CVPR, pages 3207– 3216, 2020

  22. [32]

    B. Liu, B. Liu, M. Ding, T. Zhu, and X. Yu. TI2Net: temporal identity inconsistency network for deepfake detection. In WACV, pages 4691– 4700, 2023

  23. [33]

    K. Liu, I. Perov, D. Gao, N. Chervoniy, W. Zhou, and W. Zhang. Deep- FaceLab: Integrated, flexible and extensible face-swapping framework. Pattern Recognition, 141:109628, 2023

  24. [34]

    H. H. Nguyen, J. Yamagishi, and I. Echizen. Use of a capsule network to detect fake images and videos. arXiv:1910.12467, 2019

  25. [35]

    Nirkin, Y

    Y . Nirkin, Y . Keller, and T. Hassner. FSGAN: Subject agnostic face swapping and reenactment. In ICCV, pages 7184–7193, 2019

  26. [36]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. volume 32, 2019

  27. [37]

    Prajwal, R

    K. Prajwal, R. Mukhopadhyay, V . P. Namboodiri, and C. Jawahar. A lip sync expert is all you need for speech to lip generation in the wild. In ACM MM, pages 484–492, 2020

  28. [38]

    Ramachandran, A

    S. Ramachandran, A. V . Nadimpalli, and A. Rattani. An experimental evaluation on deepfake detection using deep face recognition. In ICCST, pages 1–6, 2021

  29. [39]

    Rossler, D

    A. Rossler, D. Cozzolino, L. Verdoliva, C. Riess, J. Thies, and M. Nießner. FaceForensics++: Learning to detect manipulated facial images. In ICCV, pages 1–11, 2019

  30. [40]

    Sabir, J

    E. Sabir, J. Cheng, A. Jaiswal, W. AbdAlmageed, I. Masi, and P. Natarajan. Recurrent convolutional strategies for face manipulation detection in videos. Interfaces (GUI), 3(1):80–87, 2019

  31. [41]

    Shiohara and T

    K. Shiohara and T. Yamasaki. Detecting deepfakes with self-blended images. In CVPR, pages 18720–18729, 2022

  32. [42]

    Shiohara, X

    K. Shiohara, X. Yang, and T. Taketomi. Blendface: Re-designing identity encoders for face-swapping. In ICCV, pages 7634–7644, 2023

  33. [43]

    Siarohin, S

    A. Siarohin, S. Lathuili `ere, S. Tulyakov, E. Ricci, and N. Sebe. First Order Motion Model for Image Animation. In NeurIPS, 2019

  34. [44]

    Z. Sun, Y . Han, Z. Hua, N. Ruan, and W. Jia. Improving the efficiency and robustness of deepfakes detection through precise geometric features. In CVPR, pages 3609–3618, 2021

  35. [45]

    Tan and Q

    M. Tan and Q. Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In ICML, pages 6105–6114, 2019

  36. [46]

    Thies, M

    J. Thies, M. Zollh ¨ofer, and M. Nießner. Deferred neural rendering: Image synthesis using neural textures. ACM TOG, 38(4):1–12, 2019

  37. [47]

    Thies, M

    J. Thies, M. Zollhofer, M. Stamminger, C. Theobalt, and M. Nießner. Face2Face: Real-time face capture and reenactment of rgb videos. In CVPR, pages 2387–2395, 2016

  38. [48]

    Z. Wang, J. Bao, W. Zhou, W. Wang, and H. Li. Altfreezing for more general video face forgery detection. In CVPR, pages 4129–4138, 2023

  39. [49]

    Z. Yan, Y . Luo, S. Lyu, Q. Liu, and B. Wu. Transcending forgery specificity with latent space augmentation for generalizable deepfake detection. In CVPR, pages 8984–8994, 2024

  40. [50]

    R. Yi, Z. Ye, J. Zhang, H. Bao, and Y .-J. Liu. Audio-driven talking face video generation with learning-based personalized head pose. arXiv:2002.10137, 2020

  41. [51]

    Zhang, Z

    K. Zhang, Z. Zhang, Z. Li, and Y . Qiao. Joint face detection and alignment using multitask cascaded convolutional networks. IEEE signal processing letters , 23(10):1499–1503, 2016

  42. [52]

    Zhang, X

    W. Zhang, X. Cun, X. Wang, Y . Zhang, X. Shen, Y . Guo, Y . Shan, and F. Wang. Sadtalker: Learning realistic 3D motion coefficients for stylized audio-driven single image talking face animation. In CVPR, pages 8652–8661, 2023

  43. [53]

    Zhao and H

    J. Zhao and H. Zhang. Thin-plate spline motion model for image animation. In CVPR, pages 3657–3666, 2022

  44. [54]

    T. Zhao, X. Xu, M. Xu, H. Ding, Y . Xiong, and W. Xia. Learning self-consistency for deepfake detection. In ICCV, pages 15023–15033, 2021

  45. [55]

    Z. Zhu, G. Huang, J. Deng, Y . Ye, J. Huang, X. Chen, J. Zhu, T. Yang, J. Lu, D. Du, et al. WebFace260M: A benchmark unveiling the power of million-scale deep face recognition. In CVPR, pages 10492–10502, 2021

Pith tools

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