We present our dataset containing images with labeled polygons, annotated over Sentinel-2 L1C imagery from snow and ice-covered regions. We use similar labels as the Fmask cloud detection algorithm, i.e., clear-sky land, cloud, shadow, snow, and water. We annotated the labels manually using the QGIS software. The dataset consists of 45 scenes divided into validation (22 scenes) and test datasets (23 scenes). The source images were captured by the satellite between October 2019 and December 2020. We provide the list of '.SAFE' filenames containing the satellite imagery and these files can be downloaded from the Copernicus Open Access Hub. The dataset can be used to test and benchmark deep neural networks for the task of cloud, shadow, and snow segmentation.

The zip file contains the following:- label_mapping.txt - This file contains the mapping from labels used in the annotation to the class names.- test_SAFE_filenames.txt - This file contains the '.SAFE' filenames that are part of the test dataset.- test_labels.zip - This zip file contains 23 '.TIF' images that contain the labeled polygons for the Sentinel-2 imagery listed test_SAFE_filenames.txt- validation_SAFE_filenames.txt - This file contains the '*.SAFE' filenames that are part of the test dataset.- validation_labels.zip - This zip file contains 22 '.TIF' images that contain the labeled polygons for the Sentinel-2 imagery listed validation_SAFE_filenames.txt