SkyScript
收藏资源简介:
SkyScript是由斯坦福大学创建的一个大规模且语义多样的遥感视觉语言数据集,包含260万个图像-文本对,覆盖2.9万个不同的语义标签。该数据集通过地理坐标将Google Earth Engine上的无标签遥感图像与OpenStreetMap中的丰富语义信息相连接,实现了全球覆盖。SkyScript不仅包含广泛的对象类别,还包含子类别和详细属性,如农田作物类型、道路表面材料等。该数据集旨在支持遥感领域中多种多模态任务的发展,如开放词汇分类、检索、图像描述和文本到图像合成。
SkyScript is a large-scale, semantically diverse remote sensing vision-language dataset developed by Stanford University. It comprises 2.6 million image-text pairs and covers 29,000 distinct semantic labels. By leveraging geographic coordinates, this dataset connects unlabeled remote sensing imagery hosted on Google Earth Engine with rich semantic information from OpenStreetMap, achieving full global coverage. SkyScript not only includes a broad spectrum of object categories, but also features subcategories and fine-grained attributes such as farmland crop types, road surface materials and others. This dataset is designed to facilitate the advancement of diverse multimodal tasks in the remote sensing domain, including open-vocabulary classification, retrieval, image captioning and text-to-image synthesis.




