Augmented Deep Representations for Unconstrained Still/Video-based Face Recognition