tist0bsc / SGCN

Split Depth-wise Separable Graph Convolution Network for Road Extraction in Complex Environment from High-resolution Remote Sensing Imagery

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SGCN

Split Depth-wise Separable Graph Convolution Network for Road Extraction in Complex Environment from High-resolution Remote Sensing Imagery

Gaodian Zhou, Weitao Chen, Qianshan Gui, Xianju Li and Lizhe Wang,Split Depth-wise Separable Graph Convolution Network for Road Extraction in Complex Environment from High-resolution Remote Sensing Imagery(IEEE TGRS, 2021)

image

quick start

requirements

python 3.6 CUDA10.1 GPU:2080Ti*1

pytorch==1.1.0 tqdm==4.49.0 Pillow==6.1.0 opencv-python==4.1.0.25

parameters

You can change epoch,batch_size,lr and decay in train_config.json

train

1.Download the files mentioned in "dataset/gansu/readme"
2.python3 main.py

It will create a folder, named 'logs', and a log file. This log file will record the training process.

And the trained model, with maximum OA in validation set, will be saved in a folder, named 'saved', and record the epoch num in 'best_epoch.txt' when saving the model.

eval

1.python3 eval.py

It will evalute this model in test dataset, and print the metrics, including OA, IOU, precision, recall, F1, then save the confusion matrix in 'saved' folder.

perdict

1.python3 predict.py

You can find the visual results in 'predict/gansu/',grays value of roads in 'vis' is 255 ,while 1 in 'mask'.

cite

@ARTICLE{9614130,
author={Zhou, Gaodian and Chen, Weitao and Gui, Qianshan and Li, Xianju and Wang, Lizhe},
journal={IEEE Transactions on Geoscience and Remote Sensing}, 
title={Split Depth-wise Separable Graph-Convolution Network for Road Extraction in Complex Environments from High-resolution Remote-Sensing Images}, 
year={2021},
volume={},
number={},
pages={1-1},
doi={10.1109/TGRS.2021.3128033}}

About

Split Depth-wise Separable Graph Convolution Network for Road Extraction in Complex Environment from High-resolution Remote Sensing Imagery


Languages

Language:Python 100.0%