Cross-Resolution Weakly Supervised Water Extraction Dataset for WDTF-Net

This dataset supports the study “WDTF-Net: Wavelet-Guided Deformable Learning for High-Resolution Water Mapping from Low-Resolution Historical Water Products.” It contains three subsets constructed from Sentinel-2, Gaofen, and WorldView optical imagery, together with geographically matched weak water labels derived from the 30 m JRC Global Surface Water Occurrence product. The three subsets represent progressively increasing cross-resolution gaps, including 30 m-to-10 m, 30 m-to-2 m, and 30 m-to-0.5 m settings. The deposited files include dataset splits, GSW-derived weak labels, manually annotated test masks, and supporting metadata. Detailed file organization, preprocessing procedures, and data-use restrictions are provided in the accompanying README file. Original third-party imagery is included only where redistribution is permitted. Any restricted Gaofen or WorldView imagery remains subject to the licensing terms of the original data providers.