Interpretable Explainability for Face Expression Recognition

Training facial emotion recognition models requires large sets of data and
costly annotation processes. Additionally, it is challenging to explain the operation
principles and the outcomes of such models in a way that is interpretable and
understandable by humans. In this paper, we introduce a gamified method of
acquiring annotated facial emotion data without an explicit labeling effort by
humans. Such an approach effectively creates a robust, sustainable, and continuous
machine learning training process. Moreover, we present a novel way of providing
interpretable explanations for facial emotion recognition using action units as
intermediary features and translating them into natural language descriptions of
facial expressions of emotions.