Eliminating semantic discrepancy between different modalities is the ultimate goal of image text retrieval. However, most of the existing methods only focus on retrieval of the ground-truth instance while ignoring those semantically similar instances yet unlab... ...

请注册登录后继续浏览