IGraSS-Iterative-Graph-constrained-Semantic-Segmentation-From-Sattelite-Imagery

Usage

Dataset

We used PlanetScope optical satellite imagery to map the irrigation canals infrastructure network. The canal waterway data was obtained from the National Hydrography(NHD) Dataset, which contains shapefiles of linestrings representing canal waterways.

Training IGraSS

To train a IGraSS framework to get a refined ground truth:

/Framework/run_framework.py --iterations 5 --process_type f --model_type resnet --output_path /output_path/ --from_scratch --dilation --k 4 --R 150 --th 0.5 --r_th 0.1 --epoch 10

To see all optional arguments for training:

/Framework/run_framework.py -h

Testing the Model

Visualization