Towards Compressing Efficient Generative Adversarial Networks for Image Translation via Pruning and Distilling