Optimal Hyperparameter Tuning of Convolutional Neural Networks for Visual Sentiment Analysis

Abstract

Study of visual sentiment examines complex emotional reaction and reflex behavior of visual expressions, such as
pictures and videos. Our research aims to explain the high-level visual information content and obtain recognition
results as seven emotional states (neutral, excitement, sorrow, surprise, outrage, fear, disgust), based on facial
emotions. The project is divided into three phases: Face identification, which is the ability to recognize facial
orientation in any input image or frame inside boundary box coordinates; Facial recognition, which deals with
analyzing multiple faces together to recognize the faces belong to the same individual by matching facial embedding
vectors and Emotion Detection to define the expression on the face and classify them as happy, neutral, surprise,
disgust, fear, outrage or sad