Frequency-Aware Robustness Analysis of Deepfake Detection Models

Puntos clave

• The aim is to analyze the robustness of deepfake detection models against specific attacks and masks.
• Conducted robustness assessment on ResNet50, FreqNet, and Xception v1 models.
• Applied controlled perturbation attacks to evaluate model performance.
• Tested the impact of frequency masking on detection accuracy.
• Identified varying degrees of robustness among the models under different attack scenarios.
• Xception v1 showed improved resilience compared to ResNet50 and FreqNet.
• Frequency masking significantly affected model effectiveness.