Center for Spatial Studies

Image captioning and visual question answering are exciting problems that combine natural language processing and computer vision, currently attracting a significant interest. Some previous efforts have looked into these problems in the context of remote sensing imagery, opening a wide range of possibilities in terms of human interaction with these data through natural language. Still, the components that are involved in previously proposed models can be significantly improved, and evaluation has also mostly been carried out on relatively small datasets, often built automatically and without much diversity. This vision paper briefly surveys the current state-of-the-art in vision and language methods dealing with remote sensing data, also discussing some of the open challenges and possibilities for future work.