Abstract: Multimodal (MM) large language models (MLLMs) have achieved remarkable success in image- and region-level remote sensing (RS) image understanding tasks, such as image captioning (IC), visual ...