A diagram that compares hand-digitization versus human-in-the-loop image segmentation

A diagram that compares hand-digitization versus human-in-the-loop image segmentation workflows. The image (a) is the first in data set F, captured by Landsat 8 on 15 February 2015. The hand-drawn polygons (b) are rasterized to create a label image (c). Subplots (d) and (e) show details from the two regions identified in (c). The same image is segmented using sparse annotations, or "doodles" (f), resulting in label image (g). The same regions highlighted in (d) and (e) are shown in (h) and (i), respectively, for the image segmented using Doodler.