Facial Action Unit Recognition Based on Transfer Learning

Facial action unit recognition is an important task for facial analysis. Owing to the complex collection environ-ment, facial action unit recognition in the wild is still chal-lenging. The 3rd competition on affective behavior analysis in-the-wild (ABAW) has provided large amount of facial images with facial action unit annotations. In this paper, we introduce a facial action unit recognition method based on transfer learning. We first use available facial images with expression labels to train the feature extraction network. Then we fine-tune the network for facial action unit recognition.