Stacked U-Nets With Multi-Output for Road Extraction

Tao Sun, Zehui Chen, Wenxiang Yang, Yin Wang; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 202-206

Abstract


With the recent advances of Convolutional Neural Networks(CNN) in computer vision, there have been rapid progresses in extracting roads and other features from satellite imagery for mapping and other purposes. In this paper, we propose a new method for road extraction using stacked U-Nets with multiple output. A hybrid loss function is used to address the problem of unbalanced classes of training data. Post-processing methods, including road map vectorization and shortest path search with hierarchical thresholds, help improve recall. The overall improvement of mean IoU compared to the vanilla VGG network is more than 20%.

Related Material


[pdf]
[bibtex]
@InProceedings{Sun_2018_CVPR_Workshops,
author = {Sun, Tao and Chen, Zehui and Yang, Wenxiang and Wang, Yin},
title = {Stacked U-Nets With Multi-Output for Road Extraction},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}