COMPARISON OF FACE EMOTION RECOGNITION BASED ON EMOTION CATEGORY AND FACE ORIENTATION

Abstract

In this thesis, we try to compare face emotion recognition accuracy based on emotion category and face orientation. The main feature in this thesis concentrated on evaluating performance according to the face orientation with respect to the camera. As a data set, we use the AffectNet data set, which provides annotation for images and is available from the official website upon request. We use the pretrained HopeNet network to predict the facial orientation of each image. Three models from different families are implemented: ResNet, VGG, and CoAtNet. They are trained on one data set and tested. The result of our test is presented in the tables according to the emotion category and face orientation with respect to the camera. This work can give us an understanding of algorithm implementation classification for the emotion recognition task and the possibility to use them in outdoor video surveillance systems.