Use to extract structured values (scores, labels, numbers) from screenshots or UI captures that come in varying resolutions — locate regions relatively, then OCR only those regions.
