Towards Lifelong Self-Supervision For Unpaired Image-to-Image Translation.