Cross-modal retrieval aims to build correspondence between multiple modalities by learning a common representation space. Typically, an image can match multiple texts semantically, and vice versa, which greatly increases the difficulty of this task. To tackle this problem, probabilistic embeddings are proposed to quantify these many-to-many relationships. However, existing datasets (, MS-COCO) and metrics (, Recall@K) are hard to fully represent these diversity correspondences due to non-exhaus…