In this paper, we make the first attempt on adapting the image-text matching model well-trained on a labeled source domain to an unlabeled target domain in the absence of source data, namely, ...