GH-PGD v1.0: A Geographically Diverse Very High-Resolution Benchmark Dataset for Fine-Scale Plastic Greenhouse Mapping

GH-PGD v1.0 is a geographically diverse very high-resolution benchmark dataset for fine-scale plastic greenhouse (PG) mapping. It covers 16 study areas comprising 18 sub-regions across nine countries and contains 37,732 indexed 512 × 512 patches at a nominal spatial resolution of 0.5 m.The public release provides 37,732 semantic masks, 37,732 instance masks, 198,247 patch-level instance annotations, predefined training/validation/test splits, patch-level geographic extents, source-image valid footprints, acquisition dates, imagery attribution, and supporting documentation.The predefined dataset split comprises 18,866 training patches, 9,433 validation patches, and 9,433 test patches.Due to licensing restrictions associated with the source Google Earth imagery, the original very high-resolution RGB imagery and derived RGB image patches are not redistributed. The released geographic metadata, source-image valid footprints, acquisition information, and reconstruction documentation are provided to facilitate the preparation of corresponding imagery obtained from authorized sources. Users should verify the spatial alignment between reconstructed imagery and the released masks before model training or evaluation.The Creative Commons Attribution 4.0 International (CC BY 4.0) license applies only to the author-created annotations, masks, metadata, and documentation included in this release and does not grant rights to the underlying third-party source imagery.