A Workflow for Interactive Geospatial ML Mapping

Workshop Material for AMLD Africa

Getting Started

You can run the main notebook from Colab, or set up the repo locally:

git clone git@github.com:GeoAIAfrica/interactive_geospatial_mapping.git
cd interactive_geospatial_mapping
mamba create -n mapping python=3.10
conda activate mapping
pip install -r requirements.txt
mamba install -c conda-forge gdal
jupyter lab

Q/A

What do we want to solve?

• Map an object of interest in satellite imagery.
• Use as little resources as possible by focusing on: open-source software, 1-or-no GPU, publicly available satellite data.
• Go from no imagery/labels to vectorized predictions over the region of interest in a single notebook run.

Why is this Problem interesting?

Similar workflows can accelerate ML-assisted geospatial mapping by minimizing potential costs & adopting the best conventions.

How can I contribute?

You can join our community here.