CESBIO_AI4QC dataset

This dataset was used in the AI4QC project (Artificial Intelligence for Quality Control), in the context of cloud detection through semantic segmentation. It consists of a set of reference cloud masks generated by an active learning method for 30 Sentinel-2 scenes over 10 sites: Alta Floresta (Brazil), Arles (France), Ispra (Italy), Munich (Germany), Orleans (France), Gobabeb (Namibia), Marrakech (Morocco), Railroad Valley (USA), Pretoria (South AFrica) and Mongu (Zambia). The cloud masks are globally distributed and cover the year 2017-2018. The scenes are collected for different seasons and cloud cover types. The dataset is a smaller version of the CESBIO dataset, where the S2 L1C images have been added and the classes modified. The new classes are the following:

• 0 - clear
• 1 - thick cloud
• 2 - thin cloud
• 3 - cloud shadow

The "classification_maps" folder contains the cloud masks. The folder "S2_images_true_color" contains RGB images combining the bands B04, B03 and B02. These images are of size 1830 x 1830 pixels with a resolution of 60m, matching the size and resolution of the cloud masks. The folders "S2_images_B*" contain the Sentinel-2 scenes for one single band. Their resolution follows the Sentinel-2 specifications: 

• B02, B03, B04, B08: 10m
• B05, B06, B07, B8A, B11, B12: 20m
• B01, B09, B10: 60m

The original CESBIO dataset is available at zenodo.org.