Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 60 of the 60 references it cites were found in a registry or an index; 19 of the 30 sources its claims cite were checked for support.

Cross-dataset video deepfake detection using Transformer and CNN architectures

Open access
Petmezas, Georgios · 2026 · Machine Vision and Applications

Abstract

The growing sophistication of deepfake generation techniques poses serious challenges to the authenticity of digital media, with potential risks spanning privacy, security and misinformation. Deep learning (DL) methods have shown significant promise in detecting such manipulations; however, inconsistencies in application, the absence of standardized pipelines and limited cross-dataset generalization hinder their reliable deployment in real-world scenarios. This work presents a comprehensive evaluation of Transformer- and CNN-based architectures for video deepfake detection. Multiple benchmark datasets, along with a novel facial-reenactment dataset, are used to investigate cross-dataset generalization and pretraining with limited fine-tuning on small target subsets (10–30%). Additionally, we analyze the impact of temporal window length on detection performance. Experimental results demonstrate that TimeSformer consistently achieves the highest performance, reaching 78.4% accuracy, 0.801 area under the curve (AUC) and 77.0% F1-score with 96-frame clips and 30% fine-tuning, confirming the advantage of joint spatiotemporal modeling. All models benefit from moderate fine-tuning, with gains plateauing beyond 20%. Increasing clip length enhances performance for temporally aware models, highlighting the importance of extended temporal context. Overall, this study provides empirical evidence into the strengths and limitations of current architectures, offering guidance for future research and practical deployment of robust and generalizable deepfake detectors.

Overview

CoChat Summary

TL;DR

This paper compares CNN- and Transformer-based video deepfake detectors in a cross-dataset setting, using FF++ and Celeb-DF for training and DFD plus ReenactFaces for testing. TimeSformer performs best overall, reaching 78.4% accuracy, 0.801 AUC, and 77.0% F1-score with 96-frame clips and 30% fine-tuning.

Key findings

  • Across all configurations, TimeSformer achieved the strongest results; in the 96-frame, 30% fine-tuning setting it reached 78.4% accuracy / 0.801 AUC / 77.0% F1-score.
  • Fine-tuning improved every model, with the paper stating the largest gains were between 0% and 20% adaptation and that performance then tended to plateau.
  • Longer temporal windows generally helped temporally aware models: the paper reports better performance for 32, 64, and 96-frame clips in CNN-LSTM, CNN-Attention, and TimeSformer, while ResNet models improved only modestly.
  • CNN-Attention outperformed the ResNet variants and CNN-LSTM; for example at 96 frames and 30% fine-tuning, it reached 75.6% accuracy / 0.752 AUC / 73.6% F1-score, compared with 72.8% / 0.744 / 70.9% for ResNet101.

Methods

  • Cross-dataset evaluation with two source datasets (FF++ and Celeb-DF) and two target datasets (DFD and ReenactFaces), testing both direct transfer (0% fine-tuning) and partial adaptation (10%, 20%, 30% fine-tuning).
  • Datasets were preprocessed by frame extraction, RetinaFace-based face detection/alignment, resizing to 112 × 112, normalization, and organization into temporal windows of 32, 64, or 96 frames.
  • Compared six architectures: ResNet50, ResNet101, CNN-LSTM, CNN-Attention, ViT, and TimeSformer; models were trained with binary cross-entropy and Adam (1 × 10^-4), evaluated with accuracy, AUC, and F1-score, and results were averaged across three runs.

Limitations

  • The paper states that the datasets may not fully represent emerging generative techniques, particularly diffusion or neural rendering models.
  • The experimental setup uses visual information alone, without audio or physiological signals, which the paper says could strengthen detection in harder cases.
  • The paper notes that it did not include explicit domain adaptation modules and did not include some task-specific Transformer methods (e.g., GenConViT, DeepFake-Adapter, Swin-Fake), which may limit robustness; this is a limitation stated in the paper.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s00138-026-01809-w

Cites

60 of 60 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Overview of the proposed deepfake detection framework
Fig. 1 Overview of the proposed deepfake detection frameworkFrom the PDF
Image 2
From the PDF

Verification

Claims the paper makes

13 Valid · 3 Not valid · 5 Not checked · 47 Own results and interpretations

21 more claims

Create a free account to see all 68 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 30
  • Probably not a substantial claim. 14
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 55 references

Research works printed but not linked: 2

  1. Sabir, E., Cheng, J., Jaiswal, A., AbdAlmageed, W., Masi, I., Natarajan, P.: Recurrent convolutional strategies for face manipulation detection in videos. In: CVPR Workshops (2019)
  2. Agarwal, S., Farid, H., Gu, Y., He, M., Nagano, K., Li, H.: Protecting world leaders against deep fakes. In: CVPR Workshops (2019)

Cited by

See all 6

Metadata

Authors
Petmezas, G.
Venue
Machine Vision and Applications
Published
2026-04-07
Volume
37
Issue
3
ISSN
  • 0932-8092 (ISSN-L)
  • 1432-1769
OpenAlex
W7151598350
Cited by
6
References
55
Citations by year
2026: 6
FWCI
17.15
Type
journal article
Language
en
Open access
Licence
Locations
Machine Vision and Applications (publishedVersion, cc-by)
Funders
University of Western Macedonia
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
Keywords
cross-dataset generalization, TimeSformer, transformer models, convolutional neural network architecture, fine-tuning, spatio-temporal modeling, misinformation, temporal context

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Cross-dataset video deepfake detection using Transformer and CNN architectures. https://app.cochat.ai/papers/10.1007/s00138-026-01809-w
The paper
Petmezas, G. (2026). Cross-dataset video deepfake detection using Transformer and CNN architectures. Machine Vision and Applications. https://doi.org/10.1007/s00138-026-01809-w

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.