Self-Cropped FaceForensics++ Dataset for Face Fraud Detection

The facial image data used in this study is based on the FaceForensics++ (FF++) public dataset (Rössler et al., 2019), which contains 1000 real videos and 4000 forged videos, covering five typical deep forgery methods (DeepFakes, Face2Face, FaceSwap, NeuralTextures, FaceShifter). We use the MTCNN (Multi-task Convolutional Neural Network) face detection algorithm to extract frames (with a sampling rate of 2 frames per second) from the FF++ videos and automatically crop the face region images. All cropped face images are uniformly adjusted to 256×256 pixels.