x-so2sat

So2Sat is a multimodal dataset developed for urban land use and land cover classification. This subset, m-So2Sat, contains 22,000 paired Sentinel-1 (SAR) and Sentinel-2 (multispectral) image patches, each of size 32×32 pixels. The data is annotated with 17 urban land use and land cover classes, enabling robust benchmarking for multimodal and multi-class classification tasks. In this version, the only modification applied to the original dataset is a preprocessing step to transform the Sentinel-1 VV and VH bands into their absolute values, enhancing interpretability and consistency across samples. The original pairing of Sentinel-1 and Sentinel-2 data is preserved.