Automatic Road Extraction Deep Learning

Deep learning experiments for road extraction from high resolution satellite imagery.

Getting Started

Describe how to get started.

Installing

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.

Adding a custom model

To add a model, navigate to networks/ directory and create <model_classname>.py, where model_classname is the class of the model.

Running train.py

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

Example

python -u train.py -ls BCESSIM -lr 1e-4 -b 16 -it 200 -dv 2 -au FCDenseNet > logs/FCDenseNet &