Lightweight Mood Estimation Algorithm For Faces Under Partial Occlusion

The latest advancements in Machine Learning have led to impressive capabilities in distinguishing emotions from facial expressions, allowing computers and smart devices to accurately detect and interpret human emotions through computer vision. While a lot of work has been conducted on understanding human expressions by utilizing visual information, most of them assume that the faces are fully exposed. In this work, we present the implementation of a lightweight mood estimation deep learning model in the presence of partial occlusion where the user is wearing eyewear equipment that completely covers the area around their eyes. Examples of such eyewear are glasses for visually impaired people or a head-mounted display in a virtual reality setting. Rather than collecting a new dataset of images illustrating individuals wearing such eyewear or virtual reality equipment, we utilized a dataset based on a previous work of ours, where the occlusion arising from such headsets was obtained through simulation. That way, we were able to make use of the transfer learning approach by fine-tuning an efficient model that was pre-trained on a typical Facial Expression Recognition task.