Improving the Perturbation-Based Explanation of Deepfake Detectors Through the Use of Adversarially-Generated Samples
This work addresses the need for better interpretability in deepfake detection for security and media verification, though it is incremental as it modifies existing explanation methods.
The paper tackled the problem of improving visual explanations for deepfake detectors by using adversarially-generated samples to create perturbation masks, resulting in mostly positive performance gains in quantitative assessments and more accurate demarcation of manipulated regions in qualitative analysis.
In this paper, we introduce the idea of using adversarially-generated samples of the input images that were classified as deepfakes by a detector, to form perturbation masks for inferring the importance of different input features and produce visual explanations. We generate these samples based on Natural Evolution Strategies, aiming to flip the original deepfake detector's decision and classify these samples as real. We apply this idea to four perturbation-based explanation methods (LIME, SHAP, SOBOL and RISE) and evaluate the performance of the resulting modified methods using a SOTA deepfake detection model, a benchmarking dataset (FaceForensics++) and a corresponding explanation evaluation framework. Our quantitative assessments document the mostly positive contribution of the proposed perturbation approach in the performance of explanation methods. Our qualitative analysis shows the capacity of the modified explanation methods to demarcate the manipulated image regions more accurately, and thus to provide more useful explanations.