The FAIR1M dataset, developed at the Chinese Academy of Sciences, contains more than 15,000 satellite images with 1,000,000 detailed annotations, including specific aircraft models, ship types, and vehicles. The images are collected by China’s Gaofen satellites and cover hundreds of airports, ports, cities and towns around the world.
The dataset was prepared by the Aerospace Information Research Institute of China to promote the development of remote sensing image interpretation technology. FAIR1M allows you to train neural networks to recognize objects on high-resolution satellite images. A special feature of the dataset is the detailed annotations of objects divided into 5 groups (roads, sports fields, planes, ships, and cars), which contain 37 subgroups (see Fig.). For objects in the images, their borders are specified. Each image has a size ranging from 1000 × 1000 to 10,000 × 10,000 pixels.
The FAIR1M dataset, in addition to training machine learning models, can be used for geographical research, image processing, shooting and mapping using remote sensing methods. The International Society of Photogrammetry and Remote Sensing has chosen the dataset as a benchmark for evaluating the effectiveness of object detection algorithms. The dataset is available here. For more information about FAIR1M, see the article on arXiv.