RSICD (Remote Sensing Image Captioning Dataset)

RSICD 用于遥感图像字幕任务。从谷歌地球、百度地图、MapABC、天地图收集了上万幅遥感图像。图像固定为具有各种分辨率的 224X224 像素。遥感影像总数为10921幅，每幅影像有5个句子描述。据我们所知，该数据集是最大的遥感字幕数据集。数据集中的样本图像具有较高的类内多样性和较低的类间相异性。因此，该数据集为研究人员提供了推进遥感字幕任务的数据资源。

RSICD is designed for remote sensing image captioning tasks. It collects over ten thousand remote sensing images from Google Earth, Baidu Maps, MapABC, and Tianditu. All images are fixed to 224×224 pixels with various resolutions. The total number of remote sensing images is 10921, and each image is paired with 5 descriptive sentences. To the best of our knowledge, this dataset is the largest remote sensing image captioning dataset available to date. The sample images in the dataset exhibit high intra-class diversity and low inter-class dissimilarity. Therefore, this dataset provides researchers with a valuable data resource to advance the research on remote sensing image captioning tasks.