MAVA: Multi-Level Adaptive Visual-Textual Alignment by Cross-Media Bi-Attention Mechanism | IEEE Journals & Magazine | IEEE Xplore