(a)
(b)
(c)
(d)
Figure 6: Comparison of different semantic segmentation algorithms on Cityscapes: (a) original image; (b) FCN-8S; (c) our method; (d) ground truth.