Carnegie Mellon University, the U.S. Department of Defense, and U.S. startup CrowdAI have released a dataset of 22,000 satellite images to advance the development of systems that can automatically assess building damage. The dataset includes and pre and post-natural disaster images and includes 850,000 building annotations. The researchers used a zero to three annotation scale to categorize the level of a building’s level of damage.
Image: Marco Dormino/The United Nations Development Programme
