Set of approx 10,000 image patches from ExtremeEarth Polar Use Case Training Data (https://zenodo.org/record/3695276#.YBfpIJP7RhG) with training labels.
Ground truth labels extracted from ShapeFiles in format:
image_id x_position y_position label
Where label is one of "W" - water, "I" - ice, "L" - land, and (x_pos, y_pos) are both image pixel coordinates.