Our work uses two datasets that are derived from EnMAP L2 data:
- Large scale unsupervised EnMAP dataset
- Supervised EnMAP-DFC dataset
To obtain the datasets, you first have to download the original source data from via the EnMAP data portal. This is possible after registration once your account has been approved for the cat1distributor role. The names of the data products used for our pre-training are listed in enmap_products.txt. The unsupervised dataset consists of 64x64 pixel patches and can be created with the create_enmap_dataset.py script.
To create the labeled EnMAP-DFC dataset, you also need a local version of the GRSS IADF Data Fusion Contest 2020 dataset. Set the path at the top of create_enmap_dfc_dataset.py to your local data and execute the script to create the dataset of matched patches.