Consistent Video-to-Video Transfer Using Synthetic Dataset