The author proposes a local non-local joint network to adaptively enhance facial crucial regions in feature learning for facial expression recognition.
