Land cover classification (or semantic segmentation in the CV context), is one of the most important applications of machine / deep learning models in remote sensing image analysis. There are numerous benchmark datasets with different features, designed and published for LULC classification task. Although radar-derived and optical imagery are widely available at similar timescales and spatial resolutions, some issues make their combined processing more complicated. These issues include coregistration between satellite missions, processing of SAR imagery to correct for ground geometry and incidence angle; and the most important one, lack of reliable labeled ground truth pixels appropriate for research purposes. Here, I'm going to introduce SEN12TS; a very large satellite image dataset (1.69 TB in storage!), designed by University of Colombia and Descartes Lab, specifically for land cover classification.