资源共享

	EnglishVersion

本站首页

小组动态

LSRV: The Large-Scale Road Validation Dataset

Abstract

The LSRV dataset is collected and shared by the RSIDEA research group of Wuhan University, and it was built for the validation of the road detection task. The LSRV dataset contains the images from Boston and its surrounding cities in the United States, Birmingham in the United Kingdom and Shanghai in China. Compared with the public road dataset, the LSRV dataset contains images of different resolutions from different regions, which can widely verify the generalization ability of the model.

1. The LSRV dataset

The LSRV dataset contains three large-scale images, which were collected from Google Earth and accurately labeled for the evaluation. Fig. 1 and Table I gives the details of the large-scale images. Among the images, the ground object distribution in the Shanghai image is quite different from that in the Boston and Birmingham images, and the buildings in Shanghai are relatively high and dense, with many narrow roads between them.

image
mask
1	Boston and its surroundings, in the U.S	Birmingham, in the UK	Shanghai, in China

Fig. 1. The large-scale road validation dataset.

Table. 1. The details of the three large-scale images

Image	Region	Range(pixels)	Resolution (m/pixel)	Area (km²)
1	Boston and its surroundings, in the U.S.	23104×23552	0.4411	105.87
2	Birmingham, in the UK	22272×22464	0.3640	66.29
3	Shanghai, in China	16768×16640	0.5109	72.83

2.Download

We provide two ways to download the LSRV dataset. We hope you can fill in a simple questionnaire before downloading, which will appear after clicking the download link.
● Download link: download
● Baidu Drive (extraction code: IDEA): download
3.Experiment

Table 2, Table 3, and Table 4, respectively, lists the quantitative results for the Boston, Birmingham, and Shanghai images. The training set for these experiments is the DeepGlobe road dataset [1], and the methods used are described in [2], [3].

Table. 2. Quantitative results of the six approaches on the Boston image.

Method	P	R	F1	IoU
LinkNet50 [2]	0.8992	0.5520	0.6841	0.5198
LinkNet50+SA [3]	0.8966	0.5771	0.7022	0.5411
LinkNet50+CA [3]	0.8907	0.6040	0.7199	0.5623
LinkNet50+GA [3]	0.8882	0.6423	0.7455	0.5942
LinkNet50+MS [3]	0.9014	0.6034	0.7229	0.5660
GAMS-Net [3]	0.8746	0.6936	0.7737	0.6309

Table. 3. Quantitative results of the six approaches on the Birmingham image.

Method	P	R	F1	IoU
LinkNet50 [2]	0.8024	0.5000	0.6161	0.4452
LinkNet50+SA [3]	0.8149	0.4685	0.5949	0.4234
LinkNet50+CA [3]	0.7603	0.6756	0.7155	0.5570
LinkNet50+GA [3]	0.7814	0.6691	0.7209	0.5636
LinkNet50+MS [3]	0.8125	0.5233	0.6366	0.4669
GAMS-Net [3]	0.7477	0.7013	0.7238	0.5671

Table. 4. Quantitative results of the six approaches on the Shanghai image.

Method	P	R	F1	IoU
LinkNet50 [2]	0.9199	0.3244	0.4797	0.3155
LinkNet50+SA [3]	0.9261	0.3030	0.4566	0.2958
LinkNet50+CA [3]	0.9095	0.3912	0.5471	0.3766
LinkNet50+GA [3]	0.9096	0.4419	0.5948	0.4233
LinkNet50+MS [3]	0.9263	0.3026	0.4562	0.2955
GAMS-Net [3]	0.8770	0.4831	0.6230	0.4525

Reference
1. Demir, I., Koperski, K., Lindenbaum, D., Pang, G., Huang, J., Basu, S., Hughes, F., Tuia, D., Raska, R., 2018. Deepglobe 2018: A challenge to parse the earth through satellite images, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE, pp. 172-17209.
2. Chaurasia, A., Culurciello, E., 2017. Linknet: Exploiting encoder representations for efficient semantic segmentation, 2017 IEEE Visual Communications and Image Processing (VCIP). IEEE, pp. 1-4.
3. Lu, X., Zhong, Y., Zheng, Z., and Zhang, L., 2021. “GAMSNet: A Novel Globally Aware Road Detection Network with Multi-Scale Residual Learning.” ISPRS Journal of Photogrammetry and Remote Sensing.

4.Copyright

The copyright belongs to Intelligent Data Extraction, Analysis and Applications of Remote Sensing (RSIDEA) academic research group, State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing (LIESMARS), Wuhan University. The LSRV road dataset can be used for academic purposes only and need to cite the following paper, but any commercial use is prohibited. Otherwise, RSIDEA of Wuhan University reserves the right to pursue legal responsibility.

Reference:
[1] Lu, X., Zhong, Y., Zheng, Z., and Zhang, L., 2021. GAMSNet: A Novel Globally Aware Road Detection Network with Multi-Scale Residual Learning. ISPRS J. Photogramm. Remote Sens. 175, 340-352.

5.Contact

If you have any the problem or feedback in using LSRV road dataset, please contact:
Ms. Xiaoyan Lu: [email protected]
Prof. Yanfei Zhong: [email protected]