Description: A framework that, through prior-knowledge-agnostic augmentations, fine-grained triplet relation learning, and a progressive learning controller, systematically mines and focuses on the most discriminative forgery clues to achieve state-of-the-art robustness and generalization in face forgery detection.
