FACIAL EMOTION RECOGNITION USING VISION TRANSFORMER

Abstract

Image recognition, which is the ability of machine detect specific objects in digital images, is utilized for various tasks such as facial recognition for security purpose or visual search in Google search engine by detect the similar object in multiple images. In this thesis, we perform experiment with Vision in Transformer (ViT) model to automatically recognize and classify eight human emotions: Anger, contempt, disgust, fear, happy, neutral, sad and surprise. Particularly, we will fine tune ViT model on a custom dataset has 800 images, which includes 100 images for each emotion of American. Furthermore, the evaluation is performed on Accuracy metric, which is built-in Transformer library. As a result, the Accuracy score from the experiment is 0.65 with approximate 1.08 loss.