Compare CNN-based landmark processing and fusion strategies for facial expression recognition using action unit aligned facial regions.
