AI RESEARCH
RemoteShield: Enable Robust Multimodal Large Language Models for Earth Observation
arXiv CS.CV
•
ArXi:2604.17243v1 Announce Type: new A robust Multimodal Large Language Model (MLLM) for Earth Observation should maintain consistent interpretation and reasoning under realistic input variations. However, current Remote Sensing MLLMs fail to meet this requirement. Trained on carefully curated clean datasets, they learn brittle mappings that do not generalize to noisy conditions in operational Earth Observation. Consequently, their performance degrades when confronted with imperfect inputs in deployment.