Progressive Multi-granular Alignments for Grounded Reasoning in Large Vision-Language Models

Quang-Hung Le, Long Hoang Dang, Ngan Le +2

Existing Large Vision-Language Models (LVLMs) excel at matching concepts across multi-modal inputs but struggle with compositional concepts and high-level relationships between ent…