The MMI Facial Expression Database includes 2,900 videos and 1,500 high-resolution static images of 75 subjects. The dataset annotates the AUs present in the videos (event coding) and partially encodes them at the frame level, indicating the phase (neutral, onset, apex, offset) of the AUs in each frame. A small part is annotated as audio-visual laughter.

The dataset is constantly being updated, aiming to continuously provide a large amount of reference data for facial expression analysis research.