This research presents an in-depth exploration of accurate and robust facial emotion recognition technology, with a focus on achieving effective performance across a diverse range of individuals in real-world settings. It addresses the challenges associated with the nature of training data and its ability to capture variations among subjects. This involves a critical examination of current affective datasets and their alignment with real-world scenarios, as well as considering the question, "Do all people express themselves in the same way?". Beyond subject differences, the research also delves into technical aspects, including image quality & lighting conditions, to understand their impact on emotion recognition accuracy. Furthermore, the potential of incorporating voice and prosody (patterns of stress and intonation) as an additional modality is investigated as a method to enhance current emotion recognition models. By exploring these aspects, the research aims to improve the fairness and generalization of emotion recognition technology, ensuring its effectiveness across diverse scenarios and populations.
Part of the IPAB Workshop Series
Note: Catering will be provided.