GhalamClouds - a dataset for semantic cloud segmentation based on KazSTSat imagery archive

GhalamClouds dataset is a collection of images from the KazSTSat earth observation satellite, manually annotated for semantic cloud segmentation tasks. Presented data are part of a larger archive and include images selected from 2023-2024 of Kazakhstan territory. The image has 22 m GSD (ground sample distance) for 1 pixel at 686 km altitude.The dataset is divided into two parts: GhalamClouds80 and GhalamClouds220, representing a number of scenes used in its creation. Different experts in the field annotated each subset. (The data presented here contains 73 and 204 scenes only). Each scene has a size of 14456x1800 pixels which is then divided into patches of size 384x384 pixels presented here. For each scene, there are 7 images representing spectral channels and cloud masking (gt). The spectral channels are Near-Infrared (nir), Red (red), Red Edge (rde), Green (green), Blue (blue), and Coastal Blue (cbl).In GhalamClouds80 for each spectral channel, there are 12350 patches for training and 1140 patches for testing. For GhalamClouds220 there are 34200 and 4180 patches respectively. Each channel was additionally aligned with Near-Infrared channel and provided as well, marked as ‘a’.