Improving Cross-Modal Retrieval With Set of Diverse Embeddings | Read Paper on Bytez