Remote Sensing Visual Question Answering

Remote sensing VQA asks models to interpret Earth imagery and answer questions about objects, land use, spatial relationships, change over time, and measured quantities, supporting applications such as mapping and disaster monitoring. Common approaches adapt vision-language models through fine-tuning or lightweight adapters, while newer systems use localization, image tiling, retrieval, segmentation, or other tools to gather visual evidence. Benchmarks increasingly test high-resolution scenes, multiple sensors, video, temporal comparisons, and quantitative or spatial reasoning. A central challenge is making answers reliably grounded in the right image evidence: models still struggle with fine detail, numerical reasoning, and long-range or cross-view relations, while tool use and larger inputs can increase inference cost.