Cross-Modal Representation Learning for Lightweight and Accurate Facial Action Unit Detection | IEEE Journals & Magazine | IEEE Xplore