NA-SAR

NA-SAR NA-SAR is a North America synthetic aperture radar pretraining dataset built from NASA OPERA products. It is intended for self-supervised and masked-image pretraining of SAR foundation models. This release is an unsplit pretraining corpus. There are no train/validation/test partitions because the dataset is not intended to define an evaluation protocol. Downstream evaluations should define their own geographically and temporally appropriate splits. Dataset… See the full description on the dataset page:

Source: Hugging Face Hub (GFM-Bench/NA-SAR). Metadata imported from the dataset’s Hub tags.