A different color is automatically assigned to each label by the training algorithm, i.e., there is
no control of a user. Therefore, the same region may be represented by a different color at each
run of the segmentation algorithm though the image is segmented by the same feature extraction
method and the same neural network.