Attend to where and when: Cascaded Attention Network for Facial Expression Recognition

Abstract

Recognizing an expression in videos is a challenging problem due to dynamic changes of facial action in the whole video. We propose a novel Cascaded Attention Network(CAN) based on sptiotemporal attention, tailored to the facial expression recognition in videos. The proposed model(CAN) got state-of-the-art result on three common datasets.

Publication
IEEE Transaction on Affective Computing (Major Revision)
Date
Links