Self-Learning Transformations for Improving Gaze and Head Redirection