RPI ISL Enhanced Cohn-Kanade AU-coded Facial Expression Database


Our research group has manually created more ground-truth data for Cohn-Kanade AU-coded facial expression database:


o      AU labeling

In our own work, we modified the AU labeling in some image sequences according to the FACS description:

       Some of the AUs are relabeled with different intensity;

       We also labeled the initial frames of the image sequences that donít start from a neutral facial expression;

       More importantly, the image sequences are relabeled in a frame-by-frame manner to depict the temporal segments of displayed AUs for 14 common AUs.


o      Coordinates of facial feature points

For each image sequence, we have selected three images for three facial expression states (neutral, onset, and apex) respectively, and manually labeled the coordinates of 34 facial feature points on the face.