Installation

You can install fastrad and its core dependencies easily using standard Python package managers. We recommend using a virtual environment (e.g., venv, conda, or uv).

Prerequisites

• Python 3.11+

• A compatible PyTorch installation (if using hardware acceleration)

Installing from Source

Currently, fastrad is actively developed on GitHub. To install the latest stable version:

# Clone the repository
git clone https://github.com/helloerikaaa/fastrad.git
cd fastrad

# Install standard CPU dependencies
pip install .

Hardware Acceleration (CUDA)

For NVIDIA GPU hardware acceleration, which unlocks the massive 10x-70x speedups, you should ensure that your environment has a CUDA-compatible version of PyTorch installed.

To install with CUDA-specific optional dependencies:

pip install ".[cuda]"

cuCIM for GLSZM Acceleration

The Gray Level Size Zone Matrix (GLSZM) relies on Connected-Component labeling. fastrad uses a hyper-optimized hybrid pipeline for this.

If running on CPU, fastrad uses the highly optimized C-backed scipy.ndimage.label. However, if you are executing on a CUDA GPU, the cuCIM (RAPIDS) dependency is automatically installed via the [cuda] extension, allowing fastrad to natively keep the connected-component computations on the GPU without memory transfer penalties.