installation.md

Installation

MeshGrow requires Python 3.10+. Use a virtual environment so pipeline dependencies stay isolated:

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

Core package

pip install -e .

Pipeline dependencies

MeshGrow orchestrates external tools. Install them in this order:

pip install seqseg linflonet nnunetv2

# PyTorch — required by nnU-Net, SeqSeg, and LinFlo-Net
# Linux + NVIDIA GPU: install the CUDA wheel from https://pytorch.org/get-started/locally/
pip install torch

# pytorch3d — required by LinFlo-Net (builds from source; torch must already be installed)
pip install --no-build-isolation \
  "git+https://github.com/facebookresearch/pytorch3d.git@stable"

--no-build-isolation is needed because pytorch3d’s build imports torch. There is no universal PyPI wheel; building from the @stable tag usually works on macOS and Linux.

On macOS, use the default CPU torch build. MeshGrow sets nnU-Net to CPU when no CUDA device is available (runtime.device: auto in the default config). Expect long runtimes on large CT volumes.

See the LinFlo-Net quick start for additional platform notes.

Model weights

meshgrow download-weights --dest models/
meshgrow doctor

Weights are stored under models/ (gitignored). Set MESHGROW_ROOT or MESHGROW_WEIGHTS_DIR to override the default location.

Verify setup

meshgrow doctor
python -c "import torch; import pytorch3d; print('torch', torch.__version__, 'pytorch3d', pytorch3d.__version__)"

meshgrow doctor checks Python imports for the pipeline packages, CLI tools on PATH, and that expected checkpoint files exist under models/. It does not import pytorch3d — use the one-liner above to confirm LinFlo-Net will run.