Pretraining is All You Need for Image-to-Image Translation