Face AI Detector (ViT Specialist)
This model is a fine-tuned version of the Vision Transformer (ViT) specialized in detecting AI-generated faces (StyleGAN/ThisPersonDoesNotExist).
Performance Summary
- AUC-ROC: 0.91
- Real-Face Accuracy: 91%
- AI-Face Accuracy: 81%
Project Notes
The model was trained using Sequential Transfer Learning:
- Initial training on CIFAKE for general AI artifacts detections.
- Specialist refinement on high-resolution facial datasets to identify StyleGAN-specific inconsistencies in skin texture and facial geometry.
