Fig 1.
Road regions with occlusion problems.
Fig 2.
The proposed network structure.
Table 1.
Experimental environments.
Fig 3.
Road map (left), the output features of the optimized pix2pix (middle) and the output features of the classical pix2pix (right) at the beginning of training(Created by Microsoft power point).
Table 2.
Experimental parameters in the learning stage.
Table 3.
Quantitative evaluation of results on DeepGlobe and Massachusetts dataset.
Table 4.
Ablation tests of different loss functions.
Fig 4.
Display of road extraction results in our data set.
The first column is the collected remotes sensing, the second column is the extraction results and the third is the road images based on the second column by using morphological processing.
Table 5.
Regional classification results.