Investigating the Generalizability of Affect Detectors from Facial Expressions

Abstract

The goal of this paper was to investigate the generalizability of affect detectors created from facial expressions. Videos of students were captured while they were playing an educational game in a natural computer laboratory setup. Trained observers annotated the learning-centered affective states which served as affect labels for training detectors. Detectors were trained using data from students in the northern part of the Philippines and were tested from data of students from the southern part of the Philippines. We discuss the results, challenges and future work of face-based affect detectors from facial expressions taken in the wild.