LLMs as Bridges: Reformulating Grounded Multimodal Named Entity Recognition