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

Authors: Saad Korchi, Abderrahim Waga

Published: 2026-10-15 21:47:45+00:00

Comment: 8 pages, 3 figures, 1 table

AI Summary

This study evaluates two deep-learning strategies for image-level deepfake detection: a compact convolutional model trained from scratch and an ImageNet-pretrained MobileNetV2 model. It highlights the significant performance advantage of the transfer learning approach and uses Grad-CAM for visual attribution to provide explainability.

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.

Key findings

The fine-tuned pretrained MobileNetV2 model achieved significantly higher accuracy (95.90%) compared to the augmented from-scratch model (84.15%) on the validation set. Grad-CAM provided visual explanations of model decisions, showing different patterns of attention between the two models. The study emphasizes the practical value of pretrained visual representations but also highlights limitations regarding generalization to unseen deepfake generation methods and the interpretability of Grad-CAM outputs as causal explanations.

Approach

The authors compare two deep-learning approaches for deepfake image detection: a custom convolutional neural network trained from scratch and a MobileNetV2 model fine-tuned using transfer learning. Both models classify images as real or fake, and Grad-CAM is employed to visualize the regions influencing their predictions.

Datasets

140k Real and Fake Faces dataset

Model(s)

Custom Convolutional Neural Network, MobileNetV2

Author countries

Morocco

← Previous