Skip to content

installation-and-setup.md

Latest commit

 

History

History
162 lines (111 loc) · 3.45 KB

installation-and-setup.md

File metadata and controls

162 lines (111 loc) · 3.45 KB

Installation and Setup

This guide consolidates the setup steps needed for a first nnU-Net v2 run.

1. Check the basics

  • Use Python 3.10 or newer.
  • Linux is the primary target, but Windows and macOS are also supported.
  • GPU is strongly recommended for training.
  • Apple mps can be used, but 3D convolutions may still require CPU fallback.

2. Install PyTorch first

Install PyTorch for your hardware before installing nnunetv2:

https://pytorch.org/get-started/locally/

Choose the build that matches your environment:

  • cuda for NVIDIA GPUs
  • mps for Apple Silicon
  • cpu if no accelerator is available

Do not install nnunetv2 before PyTorch is in place.

3. Install nnU-Net

For normal use:

pip install nnunetv2

If you want a local editable checkout for development:

git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
pip install -e .

On Intel (non-Apple-Silicon) macOS, add the intel_macos extra to pin compatible versions of torch and numpy:

pip install "nnunetv2[intel_macos]"

or with an editable install:

pip install -e ".[intel_macos]"

4. Create the three storage locations

nnU-Net needs three locations:

  • nnUNet_raw: raw datasets in nnU-Net format
  • nnUNet_preprocessed: preprocessed data used during training
  • nnUNet_results: trained models and installed pretrained models

Recommended layout:

/path/to/nnUNet_raw
/path/to/nnUNet_preprocessed
/path/to/nnUNet_results

Inside nnUNet_raw, each dataset lives in its own DatasetXXX_Name folder.

Example:

nnUNet_raw/
└── Dataset001_MyDataset
    ├── dataset.json
    ├── imagesTr
    ├── imagesTs
    └── labelsTr

5. Set environment variables

Linux and macOS

For a persistent setup, add this to your shell profile such as .bashrc or .zshrc:

export nnUNet_raw="/path/to/nnUNet_raw"
export nnUNet_preprocessed="/path/to/nnUNet_preprocessed"
export nnUNet_results="/path/to/nnUNet_results"

For a temporary setup, run the same commands in the current shell before using nnU-Net.

Windows PowerShell

$Env:nnUNet_raw = "C:/path/to/nnUNet_raw"
$Env:nnUNet_preprocessed = "C:/path/to/nnUNet_preprocessed"
$Env:nnUNet_results = "C:/path/to/nnUNet_results"

Windows Command Prompt

set nnUNet_raw=C:/path/to/nnUNet_raw
set nnUNet_preprocessed=C:/path/to/nnUNet_preprocessed
set nnUNet_results=C:/path/to/nnUNet_results

6. Verify the setup

Check that the variables are visible in your shell.

Linux and macOS:

echo "$nnUNet_raw"
echo "$nnUNet_preprocessed"
echo "$nnUNet_results"

PowerShell:

echo $Env:nnUNet_raw
echo $Env:nnUNet_preprocessed
echo $Env:nnUNet_results

Command Prompt:

echo %nnUNet_raw%
echo %nnUNet_preprocessed%
echo %nnUNet_results%

7. Optional extras

hiddenlayer enables network topology plots:

pip install --upgrade git+https://github.com/FabianIsensee/hiddenlayer.git

If you train on a fast GPU, you may also want to tune nnUNet_n_proc_DA for data augmentation throughput.

8. Next steps

  1. Prepare a dataset
  2. Plan and preprocess
  3. Train models

Related reference pages