Abstract: Face forgery detection suffers from cross-dataset generalization challenges, where performance degradation occurs due to distribution shifts between training and testing data. Recently, ...