Cross Domain Image Retrieval

Cross-domain image retrieval (CDIR) aims to find matching images across different visual domains (e.g., sketches and photos) despite significant visual differences. Recent research heavily emphasizes unsupervised approaches, avoiding the need for costly labeled data, and focuses on developing robust feature learning techniques, often incorporating contrastive learning and optimal transport methods to align representations across domains. These advancements are crucial for improving the efficiency and scalability of image search in diverse applications, such as e-commerce and multimedia retrieval, where labeled data is scarce and domain variations are common.