Efficient Deepfake Detection using EfficientNet-B2 with Selective Layer-wise Fine-Tuning: A Study on the 140k Faces Benchmark
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Langue: | anglais |
| Publié: |
Zenodo
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866902305959313408 |
|---|---|
| author | Mariya, Desingh |
| author_facet | Mariya, Desingh |
| contents | <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]"><strong>"Efficient Deepfake Detection using EfficientNet-B2 with Selective Layer-wise Fine-Tuning"</strong> presents a lightweight deepfake detection system designed to run on consumer-grade hardware. The core idea is a two-phase fine-tuning strategy applied to the EfficientNet-B2 architecture — first training only the classification head, then selectively unfreezing the upper convolutional blocks (Blocks 5 and 6) to capture manipulation artifacts specific to AI-generated faces.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The model was tested on 140,000 face images (real FFHQ vs. synthetic StyleGAN) and achieved an AUC of 0.9624 and 88% accuracy, with a fake-detection precision of 0.94 — all on a modest 4GB GPU. The work demonstrates that targeted layer unfreezing outperforms both full freezing and full retraining, offering a practical and efficient alternative for digital forensics applications.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19809442 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Efficient Deepfake Detection using EfficientNet-B2 with Selective Layer-wise Fine-Tuning: A Study on the 140k Faces Benchmark Mariya, Desingh Deepfake Detection, EfficientNet-B2 Transfer Learning Layer-wise Fine-Tuning Convolutional Neural Networks Binary Classification StyleGAN Image Forensics <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]"><strong>"Efficient Deepfake Detection using EfficientNet-B2 with Selective Layer-wise Fine-Tuning"</strong> presents a lightweight deepfake detection system designed to run on consumer-grade hardware. The core idea is a two-phase fine-tuning strategy applied to the EfficientNet-B2 architecture — first training only the classification head, then selectively unfreezing the upper convolutional blocks (Blocks 5 and 6) to capture manipulation artifacts specific to AI-generated faces.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The model was tested on 140,000 face images (real FFHQ vs. synthetic StyleGAN) and achieved an AUC of 0.9624 and 88% accuracy, with a fake-detection precision of 0.94 — all on a modest 4GB GPU. The work demonstrates that targeted layer unfreezing outperforms both full freezing and full retraining, offering a practical and efficient alternative for digital forensics applications.</p> |
| title | Efficient Deepfake Detection using EfficientNet-B2 with Selective Layer-wise Fine-Tuning: A Study on the 140k Faces Benchmark |
| topic | Deepfake Detection, EfficientNet-B2 Transfer Learning Layer-wise Fine-Tuning Convolutional Neural Networks Binary Classification StyleGAN Image Forensics |
| url | https://doi.org/10.5281/zenodo.19809442 |