A Comparative Study: Deepfake Detection Using Deep-learning
Abstract
In recent decades, we have seen significant advancement in fields like Artificial Intelligence, Machine Learning, and Deep Learning, resulting in the developing of new technologies such as deepfake. Deepfakes are a form of digital media that replaces one identity’s likeness with another or creates a synthetic personality: in the form of high-quality realistic fake video, image, or audio. Deepfakes can be helpful in education, art, activism, and self-expression; however, some subjects can use deepfakes to harm the portrayal of people, create pornographic content, and spread misleading information. High-quality deepfakes are easy to build but incredibly difficult to detect, creating a need to explore technologies which can be helpful in deepfake detection. Therefore, we present a comparative study of deep-learning models that can benefit deepfake detection. We have explored four deep-learning models, namely, VGG16, MobileNetV2, XceptionNet, and InceptionV3 and trained these models on the FaceForensics++ dataset. Finally, we evaluate the performance of these models for deepfake detection and conclude the study with our observations and future scope for improvement in this field.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Our cross-comparison of XCeption, ResNet-50, VGG16, and MobileNet V2 underscores the multifaceted nature of real-world deployment.
“While many prior works concentrate on individual metrics or specific datasets, our cross-comparison of XCeption, ResNet-50, VGG16, and MobileNet V2 underscores the multifaceted nature of real-world deployment [19,20].”
From Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks (2025) · cites this paper as [19]
No single architecture uniformly dominates in every scenario.
“Our analysis suggests that no single architecture uniformly dominates in every scenario, reflecting a broader consensus in the literature [19].”
From Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks (2025) · cites this paper as [19]
Comparable accuracy ranges for VGG16 (above 85%) were also observed by Khatri et al. [19].
“Comparable accuracy ranges for VGG16 (above 85%) were also observed by Khatri et al. [19], reinforcing that it remains a robust model for deepfake detection, albeit at the cost of efficiency.”
From Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks (2025) · cites this paper as [19]
Smaller architectures, while efficient, can be more prone to targeted adversarial perturbations [19].
“This vulnerability also aligns with findings in [19] indicating that smaller architectures, while efficient, can be more prone to targeted adversarial perturbations.”
From Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks (2025) · cites this paper as [19]
By comparing these results with similar studies [11,19], we observe consistent patterns in model performance.
“By comparing these results with similar studies [11,19], we observe consistent patterns in model performance and highlight that advanced defenses, such as adversarial training and domain adaptation, are indispensable for real-world adoption.”
From Comprehensive Evaluation of Deepfake Detection Models: Accuracy, Generalization, and Resilience to Adversarial Attacks (2025) · cites this paper as [19]
Cites
A Probabilistic Interpretation of Precision, Recall and F-Score, with Implication for Evaluation
Goutte, C., & Gaussier, E. (2005) · Lecture Notes in Computer Science · doi:10.1007/978-3-540-31865-1_25 · Cited by 2,285 · see citing works
Celeb-DF: A Large-Scale Challenging Dataset for DeepFake Forensics
Li, Y., Yang, X., Sun, P., Qi, H., & Lyu, S. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.00327 · Cited by 1,846 · cited 2× · see citing works
“Datasets such as CELEB-DF [4], FaceForensics++, and Wilddeepfake [5], amongst others, have surfaced, which can be utilized for the problem.”
MethodDeep learning for deepfakes creation and detection: A survey
Nguyen, T. T., Nguyen, Q. V. H., Nguyen, D. T., Nguyen, D. T., Huynh-The, T., Nahavandi, S., Nguyen, T. T., Pham, Q.-V., & Nguyen, C. M. (2022) · Computer Vision and Image Understanding · doi:10.1016/j.cviu.2022.103525 · Cited by 464 · cited 1× · see citing works
“, a higher resolution image requires equally low accuracy of the GAN that is employed to generate the synthetic image [10].”
Deep Learning for Deepfakes Creation and Detection: A Survey
Nguyen, T. T., Nguyen, Q. V. H., Nguyen, D. T., Nguyen, D. T., Huynh-The, T., Nahavandi, S., Nguyen, T. T., Pham, Q.-V., & Nguyen, C. M. (2019) · arXiv · doi:10.48550/arxiv.1909.11573 · Cited by 464
Flagged first, then by citations
Cited by
Sorted by citations
Deepfake video detection methods, approaches, and challenges
Alrashoud, M. (2025) · Alexandria Engineering Journal · doi:10.1016/j.aej.2025.04.007 · Cited by 49 · see citing works
Abbasi, M., Váz, P., Silva, J., & Martins, P. (2025) · Applied Sciences · doi:10.3390/app15031225 · Cited by 32 · see citing works
- Authors
- Khatri, N., Borar, V., & Garg, R.
- Affiliations
- Amity University (IN)
- ORCID
- Garg, Rakesh
- Venue
- 2023 13th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
- Publisher
- IEEE
- Published
- 2023-01-19
- Pages
- 1-5
- OpenAlex
- W4321488528
- Cited by
- 29
- References
- 4
- Citations by year
- 2026: 6 · 2025: 13 · 2024: 9 · 2023: 1
- FWCI
- 3.11
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- International Conference on Cloud Computing, Data Science & Engineering (Confluence) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Keywords
- deepfake detection, deep learning models, FaceForensics++, VGG16, MobileNetV2, Xception, InceptionV3, synthetic media, image classification, model comparison
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.
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
