Code for the paper:"Spatial information inference net: Road extraction using road-specific contextual information" by Chao Tao, Ji Qi, Yansheng Li, Hao Wang and Haifeng Li.
python3
pytorch >= 1.1
The DEEPGLOBE-CVPR 2018 road extraction sub-challenge dataset (CVPR dataset) can be download from the competition website. Thanks!
The Massachusetts road dataset was presented by Mnih and Hinton et al.. We collected the dataset from their website. Thanks!
The RoadTracer dataset was presented by Bastani et al. in RoadTracer: Automatic Extraction of Road Networks from Aerial Images. We collected this dataset and conduct comparison experiments by using the code provided by the author. Thanks!
Images and GT_labels for training and testing after cropping shoulded be organized as follows:
├── root
| ├── train
| ├── train_labels
| ├── val
| ├── val_labels
In addition, the values of GT_labels are 0 and 1 (0-bg, 1-Road).
Once the data set is prepared, set the dir of your dataset at config\opt_cvpr.py
, config\opt_rbdd.py
and config\opt_mit.py
. Then, training with main_road_train.py
.
Do prediecting (or evaluating) use main_road_reval.py
for the CVPR and Massachusetts dataset;
Do prediecting (or evaluating) use main_road_reval_mit.py
and eval_roadtracer.py
for the RoadTracer dataset;
[1] C. Tao, J. Qi, Y. Li, H. Wang, and H. Li, "Spatial information inference net: Road extraction using road-specific contextual information," ISPRS-J. Photogramm. Remote Sens., vol. 158, pp. 155-166, 2019.
@article{siinet2019, author = {Tao, Chao and Qi, Ji and Li, Yansheng and Wang, Hao and Li, Haifeng}, title = {Spatial information inference net: Road extraction using road-specific contextual information}, journal = {ISPRS Journal of Photogrammetry and Remote Sensing}, volume = {158}, pages = {155-166}, ISSN = {0924-2716}, year = {2019}, type = {Journal Article} }