617618.fig.002a
(a)
617618.fig.002b
(b)
617618.fig.002c
(c)
Figure 2: Illustration of the RNN training inputs: (a) a training image (red and blue are differently labeled regions). (b) An adjacency matrix of image segments. (c) A set of correct trees which is oblivious to the order in which segments with the same label are merged.