VecRoad

The official repo of the CVPR 2020 paper "VecRoad: Point-based Iterative Graph Exploration for Road Graphs Extraction"

[paper] [homepage]

Prepare Data

Datasets

OSM Dataset Baidu Netdisk with code: 2xoe

Released models

RP-Net Baidu NetDisk with code: 9nk4

Evaluation matrics

apls-visualizer-1.0 Baidu Netdisk with code: ugbz | Google Drive

Steps

1. Download OSM Dataset, RP-Net and APLS-visualizer into data/downloads.
2. sh prepare_data.sh

Requirements

conda could install the right version of rtree.

pytorch>=0.4.1
rtree==0.8.3
numpy==1.16.3
pillow==6.0.0
opencv-python==4.1.0.25
networkx==2.3
shapely==1.6.4
pickle==0.7.5
scikit-image==0.15.0
pyyaml==5.1

Inference

python infer.py --config configs/default.yml

Evaluation

Junction Metric (Golang needed). Please refer to roadtracer for more details.

python eval/eval_junction_metric.py \
    --graph_dir data/graphs/vecroad_4/graphs_junc/ \
    --gt_dir data/input/graphs/ \
    --save_dir data/graphs/vecroad_4/ \
    --file_name graphs_junc_jf1.csv

APLS Metric (Java needed). Please refer to Spacenet Challenge for more details.

python eval/graphs2wkt.py \
    --graph_dir data/graphs/vecroad_4/graphs_junc/ \
    --save_dir data/graphs/vecroad_4/graphs_junc_wkt/

python eval/eval_apls_metric.py \
    --apls_path eval/apls-visualizer-1.0/visualizer.jar \
    --wkt_dir data/graphs/vecroad_4/graphs_junc_wkt/ \
    --gt_dir data/input/graphs_test_wkt/ \
    --save_dir data/graphs/vecroad_4/ \
    --file_name graphs_junc_apls.csv

Pixel Metric (C++ needed). Please refer to ssai-cnn for more details.

python eval/graphs2seg.py \
    --graph_dir data/graphs/vecroad_4/graphs_junc/ \
    --save_dir data/graphs/vecroad_4/graphs_junc_seg/ \
    --region_file data/input/regions/test_regions.txt \
    --img_size 8192 \
    --thickness 8

python eval/eval_pixel_metric.py \
    --pred_dir data/graphs/vecroad_4/graphs_junc_seg \
    --gt_dir data/input/mask_test \
    --steps 1 \
    --relax 3 \
    --num_workers 12 \
    --thresh 128 \
    --crop 0

Citation

If you find this work or code is helpful in your research, please cite:

@inproceedings{VecRoad_20CVPR,
 title={VecRoad: Point-based Iterative Graph Exploration for Road Graphs Extraction},
 author={Yong-Qiang Tan and Shanghua Gao and Xuan-Yi Li and Ming-Ming Cheng and Bo Ren},
 booktitle={IEEE CVPR},
 year={2020},
}

License

Our source code is free for non-commercial usage. Please contact us if you want to use it for comercial usage. (yoqitan -AT- outlook -DOT- com)