DFEW

Dynamic Facial Expression in-the-Wild (DFEW) is a large-scale facial expression database with 16372 very challenging video clips taken from movies. Clips in the DFEW database are of various challenging interferences, such as extreme illumination, occlusions, and capricious pose changes. Based on the crowdsourcing annotations, we hired 12 expert annotators, and each clip has been independently labeled ten times by them. DFEW database has enormous diversities, large quantities, and rich annotations, including:

16372 number of very challenging video clips from movies,

a 7-dimensional expression distribution vector for each video clip,

single-labeled expression annotation for classic seven discrete emotions

baseline classifier outputs based on single-labeled annotation

DFEW Code Link: https://github.com/jiangxingxun/DFEW