Ensemble of Multi-task Learning Networks for Facial Expression Recognition In-the-Wild with Learning from Synthetic Data

Jae-Yeop Jeong, Yeong-Gi Hong, Sumin Hong, Jiyeon Oh, Yuchul Jung, Sang-Ho Kim, Jin-Woo Jeong

Abstract: Facial expression recognition in-the-wild is essential for various interactive computing applications. Especially, “Learning from Synthetic Data” is an important topic in the facial expression recognition task. In this paper, we propose a multi-task learning-based facial expression recognition approach where emotion and appearance perspectives of facial images are jointly learned. We also present our experimental results on validation and test set of the LSD challenge introduced in the 4th affective behavior analysis in-the-wild competition. Our method achieved the mean F1 score of 71.82 on the validation and 35.87 on the test set, ranking third place on the final leaderboard.