Installation

orthoseg is written in python, but it relies on several other libraries that have dependencies written in C/C++. Those dependencies can be difficult to install, but luckily the conda package management system also gives an easy alternative.

Python package/environment manager

If you don’t have conda installed yet, you can install it using the miniforge installer.

Install orthoseg and depencencies

If you want to run orthoseg on a GPU (optional), make sure you have installed the necessary drivers, and that the CUDA support is working.

If you run into problems anywhere in this procedure, it can be useful to also have a look at the tensorflow installation instructions on https://www.tensorflow.org/install

On Linux or WSL2

On Linux or WSL2, all dependencies can be installed from conda-forge, which should lead to the most reliable setup.

You can create a new environment with orthoseg and its dependencies installed with the following command:

conda env create -n orthoseg -f https://orthoseg.readthedocs.io/en/stable/_static/conda_envs/environment.yml
conda activate orthoseg

To use a CUDA GPU, also install the necessary CUDA packages in the environment:

conda install -y cudatoolkit=12.3 cudnn

On native Windows

For native Windows, no tensorflow packages are published on conda-forge. Hence, we need to install tensorflow using pip.

If you want to run orthoseg on CPU, you can create a new environment with orthoseg and its dependencies installed with the following command:

conda env create -n orthoseg -f https://orthoseg.readthedocs.io/en/stable/_static/conda_envs/environment-tf-pip.yml
conda activate orthoseg

If you want to run orthoseg on native Windows while using a GPU, things are a bit more complicated because builds of recent versions of tensorflow don’t support this. Notheless, it is possible to get it working, e.g. by using an older version of tensorflow (2.10) and some older versions of other dependencies. The combinations of (older) versions are a bit sensitive, and using old versions of software is never recommended, so another setup (linux or WSL2) is recommended, but these commands created a working environment at the time of writing:

You can create a new environment with orthoseg and its dependencies to run on a CUDA GPU installed with the following command:

conda env create -n orthoseg -f https://orthoseg.readthedocs.io/en/stable/_static/conda_envs/environment-win-gpu.yml
conda activate orthoseg

On MacOS

On MacOS, conda-forge packages exist for tensorflow, but they can lead to crashes in orthoseg, so installing tensorflow with pip is recommended.

You can create a new environment with orthoseg and its dependencies installed with the following command:

conda env create -n orthoseg -f https://orthoseg.readthedocs.io/en/stable/_static/conda_envs/environment-tf-pip.yml
conda activate orthoseg

To use a CUDA GPU, also install the necessary CUDA packages in the environment:

conda install -y cudatoolkit=12.3 cudnn