CoChat

Enhancing Deepfake Detection Through Quantum Transfer Learning and Class-Attention Vision Transformer Architecture

Open access

The DOI matches its Crossref record; 39 of the 39 references it cites were found in a registry or an index; 13 of the 33 sources its claims cite were checked for support.

Abstract

The widespread use of the internet, coupled with the increasing production of digital content, has caused significant challenges in information security and manipulation. Deepfake detection has become a critical research topic in both academic and practical domains, as it involves identifying forged elements in artificially generated videos using various deep learning and artificial intelligence techniques. In this dissertation, an innovative model was developed for detecting deepfake videos by combining the Quantum Transfer Learning (QTL) and Class-Attention Vision Transformer (CaiT) architectures. The Deepfake Detection Challenge (DFDC) dataset was used for training, and a system capable of detecting spatiotemporal inconsistencies was constructed by integrating QTL and CaiT technologies. In addition to existing preprocessing methods in the literature, a novel preprocessing function tailored to the requirements of deep learning models was developed for the dataset. The advantages of quantum computing offered by QTL were merged with the global feature extraction capabilities of the CaiT. The results demonstrated that the proposed method achieved a remarkable performance in detecting deepfake videos, with an accuracy of 90% and ROC AUC score of 0.94 achieved.

Overview

CoChat Summary

TL;DR

The paper proposes a deepfake detection model that combines Quantum Transfer Learning (QTL) with a Class-Attention Vision Transformer (CaiT) and evaluates it on DFDC-based data. It reports 90% accuracy and 0.94 ROC AUC, but notes weaker performance on real-class classification and limited validation on a simulated quantum device.

Key findings

  • On the DFDC test set, the model achieved 90% accuracy and 0.94 ROC AUC.
  • Class-level results in Table 2 show better performance on fake videos than real videos: fake precision/recall/F1 = 0.96/0.91/0.93 vs. real precision/recall/F1 = 0.69/0.83/0.75.
  • Weighted-average metrics were precision 0.91, recall 0.90, and F1 0.90; macro-average F1 was 0.84.
  • In a cross-dataset check on 10 real and 10 fake CelebDF-v2 videos, the model achieved 85% overall accuracy, macro-average F1 0.85, and AUC 0.97.

Methods

  • The study built a hybrid deepfake detector using a pretrained CaiT (cait_s24_224) as the feature extractor and a Quantum Neural Network / DressedQuantumNet classifier layer with quantum gates (Hadamard, RY, CX).
  • Data came from the DFDC dataset and the Deepfake and Real Images Dataset on Kaggle; the DFDC set was split into 50 parts and the paper says 25,000 videos were used for training because of session/time constraints.
  • Preprocessing included frame extraction, face detection with MTCNN, cropping, resizing to 224 × 224, and tensor conversion; training used Adam with cross-entropy loss and a learning rate of 3 × 10^-5.

Limitations

  • The paper states the model was evaluated only in a simulated quantum environment, with no real quantum device testing.
  • It says only 50% of the DFDC dataset was used because of hardware and time constraints; it also notes that Colab session limits made full-epoch training impractical.
  • Evident from the reported results: performance is substantially weaker on the real class than on the fake class, suggesting class imbalance or class-specific optimization issues.

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

Integrity signals

Registry

Crossref · 10.3390/app15020525

Cites

39 of 39 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Model flowchart for deepfake detection based on CaiT and QTL. Figure 1. Model flowchart for deepfake detection based on CaiT and QTL.
Figure 1 Model flowchart for deepfake detection based on CaiT and QTL. Figure 1. Model flowchart for deepfake detection based on CaiT and QTL.CC-BY · From the PDF
Figure 2 Diagram of the process for extracting faces from videos and data transformation.
Figure 2 Diagram of the process for extracting faces from videos and data transformation.CC-BY · From the PDF
Figure 3 Illustrates the stages of superposition, entanglement, and measurement in a qu circuit. The numbers 0, 1, 2, and 3 represent the qubits in the circuit. The 'H' denotes the Had Figure 3. Illustrates the stages of superposition, entanglement, and measurement in a quantum circuit. The numbers 0, 1, 2, and 3 represent the qubits in the circuit. The 'H' denotes the Hadamard Gate, 'RY' represents the Rotation-Y Gate, and the circuit also includes CNOT Gates and measurement symbols.
Figure 3 Illustrates the stages of superposition, entanglement, and measurement in a qu circuit. The numbers 0, 1, 2, and 3 represent the qubits in the circuit. The 'H' denotes the Had Figure 3. Illustrates the stages of superposition, entanglement, and measurement in a quantum circuit. The numbers 0, 1, 2, and 3 represent the qubits in the circuit. The 'H' denotes the Hadamard Gate, 'RY' represents the Rotation-Y Gate, and the circuit also includes CNOT Gates and measurement symbols.CC-BY · From the PDF
Figure 4 Model construction and integration process. Figure 4. Model construction and integration process.
Figure 4 Model construction and integration process. Figure 4. Model construction and integration process.CC-BY · From the PDF
Figure 5 Data processing flowchart for the model training process. Figure 5. Data processing flowchart for the model training process.
Figure 5 Data processing flowchart for the model training process. Figure 5. Data processing flowchart for the model training process.CC-BY · From the PDF
Figure 6 Confusion matrix for test data. Figure 6. Confusion matrix for test data. accurately classifying fake content.
Figure 6 Confusion matrix for test data. Figure 6. Confusion matrix for test data. accurately classifying fake content.CC-BY · From the PDF
Figure 7 ROC curve and AUC of the model. The dashed blue line represents the performance of a random classifier, where predictions are no be tt er than chance, with equal true and false positive Figure 7. ROC curve and AUC of the model. The dashed blue line represents the performance of a random classifier, where predictions are no better than chance, with equal true and false positive rates.
Figure 7 ROC curve and AUC of the model. The dashed blue line represents the performance of a random classifier, where predictions are no be tt er than chance, with equal true and false positive Figure 7. ROC curve and AUC of the model. The dashed blue line represents the performance of a random classifier, where predictions are no better than chance, with equal true and false positive rates.CC-BY · From the PDF
Figure 8 ( a ) Confusion matrix showing classification performance on the cross-dataset evaluation; ( b ) ROC curve representing the model's performance for the cross-dataset evaluation. The dashed blue line represents the performance of a random classifier, where predictions are no be tt er than chance, with equal true and false positive rates. Figure 8. ( a ) Confusion matrix showing classification performance on the cross-dataset evaluation; ( b ) ROC curve representing the model's performance for the cross-dataset evaluation. The dashed blue line represents the performance of a random classifier, where predictions are no better than chance, with equal true and false positive rates.
Figure 8 ( a ) Confusion matrix showing classification performance on the cross-dataset evaluation; ( b ) ROC curve representing the model's performance for the cross-dataset evaluation. The dashed blue line represents the performance of a random classifier, where predictions are no be tt er than chance, with equal true and false positive rates. Figure 8. ( a ) Confusion matrix showing classification performance on the cross-dataset evaluation; ( b ) ROC curve representing the model's performance for the cross-dataset evaluation. The dashed blue line represents the performance of a random classifier, where predictions are no better than chance, with equal true and false positive rates.CC-BY · From the PDF

Verification

Claims the paper makes

8 Valid · 5 Not valid · 18 Not checked · 68 Own results and interpretations

23 more claims

Create a free account to see all 99 claims and to verify the 10 references still open.

See all claims — free account
  • Probably not a substantial claim. 53
  • It may state two separate claims as one. 28
  • The quote only partly states this claim. 4

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.

2 of 2

Citations

Cites

Show all 39 references

Research works printed but not linked: 1

  1. Rijsbergen, 1979, Information retrieval; Butterworth, 1978, J. Libr., 11, 237

Cited by

See all 12

Metadata

Authors
Katı, B. E., Küçüksille, E. U., & Sarıman, G.
Affiliations
  • Süleyman Demirel Üniversitesi (TR)
  • Muğla University (TR)
Corresponding author
Katı, Bekir Eray
Venue
Applied Sciences
Publisher
MDPI AG
Published
2025-01-08
Volume
15
Issue
2
Pages
525
ISSN
2076-3417 (ISSN-L)
OpenAlex
W4406174629
Cited by
12
References
32
Citations by year
2026: 6 · 2025: 6
FWCI
4.19
Type
journal article
Language
en
Open access
Locations
  • Applied Sciences (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
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
deepfake detection, quantum transfer learning, video forgery detection, data preprocessing, quantum computing

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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and 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: Enhancing Deepfake Detection Through Quantum Transfer Learning and Class-Attention Vision Transformer Architecture. https://app.cochat.ai/papers/10.3390/app15020525
The paper
Katı, B. E., Küçüksille, E. U., & Sarıman, G. (2025). Enhancing Deepfake Detection Through Quantum Transfer Learning and Class-Attention Vision Transformer Architecture. Applied Sciences. https://doi.org/10.3390/app15020525

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.