Training Strategies and Data Augmentations in CNN-based DeepFake Video Detection

The fast and continuous growth in number and quality of deepfake videos calls\nfor the development of reliable detection systems capable of automatically\nwarning users on social media and on the Internet about the potential\nuntruthfulness of such contents. While algorithms, software, and smartphone\napps are getting better every day in generating manipulated videos and swapping\nfaces, the accuracy of automated systems for face forgery detection in videos\nis still quite limited and generally biased toward the dataset used to design\nand train a specific detection system. In this paper we analyze how different\ntraining strategies and data augmentation techniques affect CNN-based deepfake\ndetectors when training and testing on the same dataset or across different\ndatasets.\n