Forensic-Aware Continual Adaptation for Image Forgery Localization
This work introduces the first continual learning framework for IFL and establishes a comprehensive benchmark under two realistic data-evolution protocols: cross-dataset and cross-content continual learning.