CRIS: CLIP-Driven Referring Image Segmentation