(a)
(b)
(c)
(d)
(e)
Figure 5: Comparison of different semantic segmentation algorithms on PASCAL VOC 2012: (a) original image; (b) FCN-8S; (c) DeepLab; (d) our method; (e) ground truth.