The labeled regions are: blue ground truth GT (first column), red machine segmentation8 MS (second), GT\MS (third), and MS\GT (last). These are shown for three different slices (a)–(c) of k1. On close inspection, these slices present examples of undersegmentation, oversegmentation, and also random errors, each of the order of a few pixels. The random errors are similar to our simulated salt-and-pepper and are visible in the second and third columns in (c).