Face Recognition on Facebook: What Is it? Why Bother?

DeepFace face alignment pipeline: (a) the detected face with 6 fiducial points; (b) the cropped image; (c) the face with 67 fiducial points; (d) the reference 3D face shape transformed to respectively aligned image; (e) the image showing visibility of the different areas on the face, darker areas are less visible; (f) the 67 fiducial points induced by the 3D model that are used to “turn” the face straight; (g) the final frontalized image; (h) a new view generated using the 3D model.