Emotions in LatAm: A new dataset and benchmark for emotion recognition in Latin America

Pooja Kishore Kumar, Willams de Lima Costa, Renato Nogueira Ferraz e Oliveira, Veronica Teichrieb, Estefania Talavera Martinez; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025, pp. 41-47

Vision-based emotion recognition uses images or videos to analyze visual cues, such as facial expressions, to infer emotions. Researchers often explore how humans interpret these cues to develop more robust emotion recognition systems. Studies suggest that, while biological factors play a predominant role in allowing this capability, cultural influences shape and adapt universal emotions. Given the role of culture in this process, a major concern is that existing emotion recognition datasets predominantly feature content from North America and Europe, limiting their global representativeness. To bridge this gap, we introduce the Emotions in LatAm dataset (EiLA), a novel dataset comprising emotion recognition data collected exclusively in Latin America. Our goal is to enable future research on emotion recognition from a Responsible AI perspective. Additionally, we benchmark the performance of state-of-the-art and widely used open-source models on the task of Facial Expression Recognition (FER) using EiLA.