MagDR: Mask-guided Detection and Reconstruction for Defending Deepfakes
This addresses the threat of deepfakes for ensuring the authenticity of visual content, presenting an incremental improvement over prior adversarial perturbation methods.
The paper tackles the problem of defending deepfakes from adversarial attacks by proposing MagDR, a mask-guided detection and reconstruction pipeline that defends three main deepfake tasks and shows promising performance against both black-box and white-box attacks.
Deepfakes raised serious concerns on the authenticity of visual contents. Prior works revealed the possibility to disrupt deepfakes by adding adversarial perturbations to the source data, but we argue that the threat has not been eliminated yet. This paper presents MagDR, a mask-guided detection and reconstruction pipeline for defending deepfakes from adversarial attacks. MagDR starts with a detection module that defines a few criteria to judge the abnormality of the output of deepfakes, and then uses it to guide a learnable reconstruction procedure. Adaptive masks are extracted to capture the change in local facial regions. In experiments, MagDR defends three main tasks of deepfakes, and the learned reconstruction pipeline transfers across input data, showing promising performance in defending both black-box and white-box attacks.