Given a one-shot image with a reference mask, the models are required to segment the indicated target object in any other images.