GRBD: a global-scale multiple-type river barrier annotation database for global river barrier mapping with deep learning

GRBD is a global-scale, multi-category annotated dataset designed for deep learning–based river barrier detection. Constructed from 2022–2024 Google Earth remote-sensing imagery at 1 m spatial resolution, the dataset comprises 20,770 standardized image tiles (680 × 680 pixels) and 27,764 finely annotated instances spanning six major barrier categories: hydropower dams, earthen dams, locks, sluice gates, tidal barrages, and weirs. RT-DETRv2-x-bestbt.zip contains the best-performing RT-DETRv2-X model trained using the GRBD dataset. Barrier image training set contain the training images and corresponding JSON annotation files used for model training. Barrier image validation set contain the validation images and corresponding JSON annotation files used for model validation and parameter selection. Barrier image test set contain the test images and corresponding JSON annotation files used for independent model evaluation. All compressed files are password-protected. The extraction password is the name of the corresponding author in lower case