Installation

GDAL

This project requires GDAL, which can be a pain to install. Pip is unable to install it, so there are two recommended ways:

1. Install it system wide. e.g. sudo apt-get install gdal-bin
2. Use conda. e.g. conda install gdal (after creating a conda environment for this project)

Both methods seem to work just about as well. If using UV later, Conda will have some redundancy with python environments. (Conda has a python for the environment, then UV has another python for the project venv. Kind of icky but seems to work okay)

With UV

Otherwise, the rest of the dependencies can be installed via pip. The following steps uses the pip-based environment/dependency manager uv. Install uv with:

pip install uv

Clone the project:

git clone git@github.com:alexmitchell/map_dataset.git
cd map_dataset
uv sync

The pyproject.toml was created following instructions on UV's guide to facilitate installing pytorch with CPU or CUDA GPU processing. If you need a CUDA version not listed below, follow the guide to add it to the pyproject.toml.

To install all dependencies with the CPU version of Pytorch, run:

uv sync --extra cpu

To install all dependencies with the CUDA version of Pytorch, run (NOTE: I don't have a GPU so can't test the CUDA install):

uv sync --extra cu124

Without UV

Dependencies are listed in pyproject.toml. Good luck :P

For pytorch, follow the instructions on the Pytorch website.

Running

If you used uv to install the project, you can run scripts from the command line with:

uv run extract_tile.py 1234

Or can run scripts in interactive mode in your IDE like VSCode. (If it recognizes the # %% syntax for cells)