- Conducting ML-assisted geospatial mapping starting with no labels require a lot of effort.
- Steps: image acquisition, image preparation/preprocessing, setting up a labeling environment & budget, preparing a compute environment, and finally training a deep learning model.
- The required resources makes it difficult to use deep learning models in regions where it is needed most, especially in Africa.
The need for accelerating such workflows using open-source software, publicly available satellite imagery, and free compute is essential in low-resource environments.
In this session, we aim to showcase an end-to-end workflow for mapping objects in satellite imagery using publicly available resources with minimal costs on the user.
You can adapt & customize the workflow to your needs. You can build on it to make it easier to be used to empower geospatial machine learning applications in the continent.
You can run the notebook from Colab, or set up the repo locally:
git clone git@github.com:Akramz/flexible_geospatial_mapping.git
cd flexible_geospatial_mapping
mamba create -n mapping python=3.10
conda activate mapping
pip install -r requirements.txt
jupyter notebook