Installation

Installing FireANTs is super easy! We recommend using a virtual environment using miniconda or venv.

We recommend getting a fresh conda environment for installing FireANTs.

conda create -n fireants python=3.10

Next, you can install FireANTs locally from source:

git clone https://github.com/rohitrango/fireants
cd fireants
pip install -e .

or install directly from PyPI (this may not be the latest iteration):

pip install fireants

It's as easy as that! No more superbuilds. No more cmake. No more waiting for hours to compile. No more esoteric errors. Just install and run! 🚀

Fused CUDA Operations

FireANTs also comes with a set of fused CUDA operations to improve memory and runtime performance of core operations in the registration pipeline. To install the fused CUDA operations, install the fireants_fused_ops package:

cd fused_ops
python setup.py build_ext && python setup.py install
cd ..

This will build the required CUDA extensions and install the fireants_fused_ops package. You can find the incredible memory savings and runtime improvements in the paper. Note that this package is only available for NVIDIA GPUs - sorry CPU and MPS users! If you are interested in implementing these fused ops for CPUs or other hardware, please open an issue or submit a pull request.

Docker

For Windows users, PyTorch installation might be tricky, especially NCCL support. We recommend using the Docker image to avoid these issues. See docker/README.md for more details.