Unsupervised Multimodal Video-to-Video Translation via Self-Supervised Learning.