Deep Learning for Deepfake Image Detection: Comparative Model Evaluation and Gradient-Based Visual Attribution

Key points

• Synthetic facial imagery has made the distinction between authentic and generated content increasingly important for digital forensics and online trust.1
• Experiments were conducted with the balanced 140k Real and Fake Faces dataset, comprising 140,000 facial images.1
• Under the recorded validation protocol, the augmented from-scratch model reached 84.15% accuracy (loss 0.3625), compared with 95.90% accuracy (loss 0.1074) for the fine-tuned pretrained model.1
• Gradient-weighted class activation mapping (Grad-CAM) was used to visualize image regions associated with predictions, and a Flask-based prototype was developed for side-by-side inspection.1

Sources (1)

• [1]Deep Learning for Deepfake Image Detection: Comparative Model Evaluation and Gradient-Based Visual Attribution
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 09:47 PM

  This study examines an explainable deep-learning workflow for image-level deepfake detection using two contrasting classification strategies: a compact convolutional model trained from scratch and an ImageNet-pretrained MobileNetV2 model adapted through two-stage transfer learning.

  Synthetic facial imagery has made the distinction between authentic and generated content increasingly important for digital forensics and online trust.

Extractive summary: sentences quoted from the sources.