EMOTIC stands for EMOTions In Context, which is a dataset of human facial images in natural environments. The images have obvious facial emotion annotations, which are annotated by an extended list of 26 emotion categories and 3 common continuous dimensions: Valence, Arousal and Dominance.

This dataset is used to train different CNN models for emotion recognition. The researchers combined the annotation information of the portraits with the contextual information extracted from the scenes and concluded that the scene context provides important information for the automatic recognition of emotions and promotes research in this direction.