ML Security PapersDeiTFake: Deepfake Detection Model using DeiT Multi-Stage Training
Saksham Kumar 1, Ashish Singh 2, Srinivasarao Thota 1, Sunil Kumar Singh 3, Chandan Kumar 4
Published on arXiv
2511.12048
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Two-stage progressive training with DeiT achieves 99.22% accuracy and AUROC of 0.9997 on OpenForensics, surpassing all reported baselines on that dataset.
DeiTFake
Novel technique introduced
Deepfakes are major threats to the integrity of digital media. We propose DeiTFake, a DeiT-based transformer and a novel two-stage progressive training strategy with increasing augmentation complexity. The approach applies an initial transfer-learning phase with standard augmentations followed by a fine-tuning phase using advanced affine and deepfake-specific augmentations. DeiT's knowledge distillation model captures subtle manipulation artifacts, increasing robustness of the detection model. Trained on the OpenForensics dataset (190,335 images), DeiTFake achieves 98.71\% accuracy after stage one and 99.22\% accuracy with an AUROC of 0.9997, after stage two, outperforming the latest OpenForensics baselines. We analyze augmentation impact and training schedules, and provide practical benchmarks for facial deepfake detection.
Key Contributions
- DeiTFake: a DeiT vision transformer adapted for facial deepfake detection using knowledge distillation
- Two-stage progressive training framework: standard augmentations in stage one followed by advanced affine and deepfake-specific augmentations in stage two
- Achieves 99.22% accuracy and AUROC of 0.9997 on OpenForensics (190,335 images), outperforming prior baselines
🛡️ Threat Analysis
DeiTFake proposes a novel architecture and training strategy for detecting AI-generated facial deepfakes — a direct contribution to output integrity and AI-generated content detection.