Pose-robust personalized facial expression recognition through unsupervised multi-source domain adaptation

Shangfei Wang, Yanan Chang, Qiong Li, Can Wang, Guoming Li, Meng Mao

Abstract: Highlights•An unsupervised multi-source domain adaptation method is proposed for FER.•Adversarial learning is used to learn a source FER encoder avoiding pose variations.•Adversarial domain adaptation is used to train a personalized model for each subject.•The feature learned by the personalized model is pose-robust and subject-specific.