A method that learns shared embedding spaces by contrasting similar and dissimilar image pairs, then ranks candidates by similarity.