After training on an extended version of the
PhraseCut dataset, our system generates a binary segmentation map for an image based on a free-text prompt or on
an additional image expressing the query.