S2Looking: A Satellite Side-Looking Dataset for Building Change Detection

Building-change detection underpins many important applications, especially in the military and crisis-management domains. Recent methods used for change have shifted towards deep learning, which depends on quality of its training data. The assembly large-scale annotated satellite imagery datasets is therefore essential global building-change surveillance. Existing almost exclusively offer near-nadir viewing angles. This limits range changes that can be detected. By offering larger observation ranges, scroll imaging mode optical satellites presents an opportunity to overcome this restriction. paper introduces S2Looking, a building-change-detection dataset contains side-looking images captured at various off-nadir consists 5000 bitemporal image pairs rural areas more than 65,920 instances throughout world. train deep-learning-based change-detection algorithms. It expands upon existing by providing (1) angles; (2) large illumination variances; (3) added complexity images. To facilitate {the} use dataset, benchmark task has been established, preliminary tests suggest deep-learning algorithms find significantly challenging closest-competing LEVIR-CD+. S2Looking may promote advances available https://github.com/S2Looking/.