Deep learning experiments for road extraction from high resolution satellite imagery.
Describe how to get started.
git clone https://github.com/aznboystride/automatic-road-extraction
cd automatic-road-extraction/
mkdir weights optimizers # training and testing program will read and write to these folders.
To add a model, navigate to networks/
directory and create <model_classname>.py
, where model_classname
is the class of the model.
Run python train.py
to see the parameters required.
usage: train.py [-h] -lr LR -b BATCH -it ITERATIONS -dv DEVICES [-lw LWEIGHTS]
[-ls LOSS] [-e EPOCH] [-au]
model
python -u train.py -ls BCESSIM -lr 1e-4 -b 16 -it 200 -dv 2 -au FCDenseNet > logs/FCDenseNet &