CLIP-ReID: Exploiting Vision-Language Model for Image Re-Identification without Concrete Text Labels