Modelling Emotions Recognition from Facial Expression using Vision Transformer with IMED Dataset

Humans use facial expressions to express their emotions. Facial expressions can convey hidden feelings that are not verbalized. This research aims to find out, implement, and fine-tune Vision Transformer architecture to recognize basic emotions through human faces using the Indonesian Mixed Emotion Dataset (IMED). The research methods used are literature review, performance comparison alongside similar models that used the same dataset, and interviewing with a psychology student, concluding with Fear and Disgust being the hardest to predict as both have similarities in eye, mouth, or eyebrows, and develop an application using Flutter framework. The experiment results show that the proposed model of this research predicts emotions faster than most models by 21.5% on average. Furthermore, it produces 98.9% test accuracy and can compete against state-of-the-art models such as CNN. YOLOv3, and ARCH-I.