Abstract: Semisupervised deep facial expression recognition (FER) tries to learn better representations from both labeled and unlabeled data to avoid the huge manual labeling cost in the supervised ...