Multiple Land-use / Land-cover Dataset (MLULC)

This dataset covers the French metropolitan territory (500,000km²). It includes

Six open access land-cover maps from various providers (CLC, CGLS-LC100, OSO, OCS-GE cover, OCS-GE use, MOS).

 A May 2019 Sentinel-2 L3A mosaic ( cloudless image using all maps available during a month) including RVB and NIR. Provided by Theia.

A Manually built ground truth of 2300 random points annotated with their labels in each map nomenclature.

A consolidated ground truth with the original 2300 and 400 non-random points focusing on rare classes.

A suggested train/val/test split (60%,5%,35%). Note that all ground truth points belong to patches of the suggested test set.

Since this dataset is intended to be used with a deep learning algorithm, the data is split into tiles of 6x6km² following a grid given with the dataset.

More information is provided in README.

Note that exception made of the ground truth, all the data (Land covers and Sentinel-Images) aren't our property and are only shared as authorized by their respective original license.