AffectNet is a large-scale facial expression dataset. The dataset uses 6 languages and 1,250 emotion-related keywords to collect more than 1 million facial images from 3 major search engines, of which about 440,000 images are manually annotated with 7 expressions (classification model) and the intensity of valence and arousal (dimensional model).

This dataset is the largest database of facial expressions, valence and arousal to date, and can automatically recognize facial expressions in two different emotion models.