Tag
This paper introduces Modl, a technique for creating cross-lingual remote-sensing multimodal large language models by composing domain and language LoRAs with mutual orthogonality, achieving superior performance without paired multilingual data.