KDGraph for Road Graph Extraction

The official code for KDGraph: A Keypoint Detection Method for Road Graph Extraction from Remote Sensing Images.

KDGraph

1. Prepare SpaceNet3 dataset for KDGraph in the spacenet_transform directory.

2. Use the following command for training on SpaceNet3.

   python main.py --config ./config.yaml --mode train

3. Inference on SpaceNet3.

   python main.py --config ./config.yaml --mode test --checkpoint_path ckpt_path

4. Metrics evaluation.

   cd metrics
   . main.bash
   

SoR dataset

This dataset is collected from 15 cities around the world to evaluate the ability of different road graph extraction methods in handling shadows and occlusions. The resolution is 0.6 m.

Baidu Netdisk Link: https://pan.baidu.com/s/1TtliEMDvaXgg3WGuL3KzLQ Code: 7ddc

Online road detector

This is a web application to conduct online road detection task worldwide. The code can be found in https://github.com/ruoyxue/Online-road-detection-application.

The demo video of KDGraph: https://www.youtube.com/watch?v=3bk0pOWXV4M&t=21s.

Citation

If you use the KDGraph code, please consider citing the following paper:

@article{he2024kdgraph,
  title={KDGraph: A Keypoint Detection Method for Road Graph Extraction from Remote Sensing Images},
  author={He, Wei and Xue, Ruoyao and Lu, Fangxiao and Xu, Jinjun and Zhang, Hongyan},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2024},
  publisher={IEEE}
}