I do ML research and had to work on applied remote sensing for a project. My daily driver was NixOS for years, and this project required GDAL and a bunch of GIS libraries. Tried to get it working, eventually gave up and switched to Fedora just to move on with the project.
My use case was just creating datasets from KMLs/GeoJSONs. At the time I patched together a system that could do the job, but I'd wanted to turn it into a proper package for a while.
So I wrote mapcv. CLI + Python library, tile math and rasterization in Rust, no GDAL. It’s extremely lean and lightweight. The only non-trivial dependencies are numpy, shapely, and Pillow; the rest is just type stubs, a YAML parser, and CLI helpers.
Try it via PyPI:
pip install mapcv
Github: https://github.com/tahamukhtar20/mapcv
Docs: https://tahamukhtar20.github.io/mapcv
Just released v0.1.0. If you try it, feedback and questions are welcome. If you hit any issues, feel free to open one on the repo.