SEN2VENµS, a Dataset for the Training of Sentinel-2 Super-Resolution Algorithms

Boosted by the progress in deep learning, Single Image Super-Resolution (SISR) has gained a lot of interest remote sensing community, who sees it as an opportunity to compensate for satellites’ ever-limited spatial resolution with respect end users’ needs. This is especially true Sentinel-2 because its unique combination resolution, revisit time, global coverage and free open data policy. While there been great amount work on network architectures recent years, deep-learning-based SISR still limited availability large training sets requires. The lack publicly available datasets required variability terms landscapes seasons pushes researchers simulate their own means downsampling. may impair applicability trained model real-world at target input resolution. paper presents SEN2VENµS, open-data licensed dataset composed 10 m 20 cloud-free surface reflectance patches from Sentinel-2, reference spatially registered 5 acquired same day VENµS satellite. covers 29 locations earth total 132,955 256 × pixels can be used comparison super-resolution algorithms bring 8 bands up m.