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

A study posted to arXiv (2610.22676v1) found that an ImageNet-pretrained MobileNetV2 model adapted through two-stage transfer learning reached 95.90% validation accuracy (loss 0.1074) on deepfake face detection, versus 84.15% accuracy (loss 0.3625) for a compact convolutional model trained from scratch. The experiments used the balanced 140k Real and Fake Faces dataset of 140,000 facial images, with gradient-weighted class activation mapping (Grad-CAM) used to visualize prediction regions and a Flask-based prototype built for side-by-side inspection. The authors caution that the figures are validation results rather than held-out test estimates, and that validation performance should be read alongside dataset provenance and explanation limitations.

arXiv:2610.22676v1 Announce Type: new Abstract: Synthetic facial imagery has made the distinction between authentic and generated content increasingly important for digital forensics and online trust. 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. Experiments were conducted with the balanced 140k Real and Fake Faces dataset, comprising 140,000 facial images. 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. 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. These observations support the practical value of pretrained visual representations in the studied setting while underscoring the need to interpret validation performance alongside dataset provenance, explanation limitations, and evaluation scope. The reported results are validation results, not held-out test estimates.