DT-TransUNet: A Dual-Task Model for Deepfake Detection and Segmentation

The swift advancement of deepfake technology has given rise to concerns regarding its potential misuse, necessitating the detection of fake images and acquisition of supportive evidence. In response to this need, we present a novel dual-task network model called DT-TransUNet, which concurrently performs segmentation for both deepfake detection and deepfake segmentation. Additionally, we propose a new Multi-Scale Spatial Frequency Feature (MSSFF) module that employs the Stationary Wavelet Transform (SWT) to extract multi-scale high-frequency components and enhance these features using a texture activation function. When evaluated on multiple datasets, DT-TransUNet surpasses comparable methods in performance and visual segmentation quality, thereby validating the effectiveness and capability of the MSSFF module for deepfake detection and segmentation tasks.