Abstract: Existing studies have shown that learning rich and precise semantic concepts helps to improve the performance of image-text cross-modal retrieval. However, existing models are trained using ...