MIPT campus road network mapping from satellite images

Road maps of the MIPT campus (Moscow Institute of Physics and Technology) and its surroundings are generated from satellite imagery using segmentation models (UNet and DeepLabV3) and road network graph reconstruction techniques.

Segmentation dataset

For model training the DeepGlobe Road Extraction Dataset was used, which contains approximately 6k images, each 1024x1024, with a scale of 50cm per pixel.

Usage

• The first option is to open and run the notebook /notebooks/road_mapping.ipynb with comments and visualizations in Kaggle or Google Colab.

• The second option is cloning the repo, installing the needed requirements, and working locally:

git clone https://github.com/RadyaSRN/MIPT-campus-satellite-road-network-mapping.git
cd MIPT-campus-satellite-road-network-mapping
conda create -n roadnet python=3.10
conda activate roadnet
pip install -r requirements.txt