Learning to rematch mismatched pairs for robust cross-modal retrieval using the L2RM framework based on Optimal Transport.