What AI says about XAI-Deepfakes Explainability Toolkit

XAI-Deepfakes is a PyTorch-based project that provides code and data to evaluate explainable AI (XAI) methods for deepfake detection. It evaluates several explanation techniques—GradCAM++, RISE, SHAP, LIME, and Sobol—on top of a state-of-the-art EfficientNet-based deepfake detector to identify image regions most influential to the detector's decision. The framework includes dataset information and experiments that test how explanations hold up under adversarial attacks intended to flip or degrade the detector's predictions.