1. Installing scikit-image#

How you should install scikit-image depends on your needs and skills:

1.1. Supported platforms#

  • Windows 64-bit on x86 processors

  • macOS on x86 and M (ARM) processors

  • Linux 64-bit on x86 processors

While we do not officially support other platforms, you could still try building from source.

1.2. Version check#

To see whether scikit-image is already installed or to check if an install has worked, run the following in a Python shell or Jupyter notebook:

import skimage
print(skimage.__version__)

or, from the command line:

python -c "import skimage; print(skimage.__version__)"

(Try python3 if python is unsuccessful.)

You’ll see the version number if scikit-image is installed and an error message otherwise.

1.3. Installation via pip and conda#

These install only scikit-image and its dependencies; pip has an option to include related packages.

1.3.1. pip#

Prerequisites to a pip install: You’re able to use your system’s command line to install packages and are using a virtual environment (any of several).

While it is possible to use pip without a virtual environment, it is not advised: virtual environments create a clean Python environment that does not interfere with any existing system installation, can be easily removed, and contain only the package versions your application needs. They help avoid a common challenge known as dependency hell.

To install the current scikit-image you’ll need at least Python 3.6. If your Python is older, pip will find the most recent compatible version.

# Update pip
python -m pip install -U pip
# Install scikit-image
python -m pip install -U scikit-image

To access the full selection of demo datasets, use scikit-image[data]. To include a selection of other scientific Python packages that expand scikit-image’s capabilities to include, e.g., parallel processing, you can install the package scikit-image[optional]:

python -m pip install -U scikit-image[optional]

Warning

Please do not use the command sudo and pip together as pip may overwrite critical system libraries which may require you to reinstall your operating system.

1.3.2. conda#

Miniconda is a bare-essentials version of the Anaconda package; you’ll need to install packages like scikit-image yourself. Like Anaconda, it installs Python and provides virtual environments.

Once you have your conda environment set up, you can install scikit-image with the command:

conda install scikit-image

1.4. System package managers#

Using a package manager (yum, apt-get, etc.) to install scikit-image or other Python packages is not your best option:

  • You’re likely to get an older version.

  • You’ll probably want to make updates and add new packages outside of the package manager, leaving you with the same kind of dependency conflicts you see when using pip without a virtual environment.

  • There’s an added risk because operating systems use Python, so if you make system-wide Python changes (installing as root or using sudo), you can break the operating system.

1.5. Downloading all demo datasets#

Some of the data used in our examples is hosted online and is not installed by default by the procedures explained above. Data are downloaded once, at the first call, but this requires an internet connection. If you prefer downloading all the demo datasets to be able to work offline, ensure that package pooch is installed and then run this command:

python -c 'from skimage.data import download_all; download_all()'

or call download_all() in your favourite interactive Python environment (IPython, Jupyter notebook, …).

1.6. Additional help#

If you still have questions, reach out through

To suggest a change in these instructions, please open an issue on GitHub.

2. Installing scikit-image for contributors#

We are assuming that you have a default Python environment already configured on your computer and that you intend to install scikit-image inside of it.

We also make a few more assumptions about your system:

  • You have a C compiler set up.

  • You have a C++ compiler set up.

  • You are running a version of Python compatible with our system as listed in our pyproject.toml.

  • You’ve cloned the git repository into a directory called scikit-image. You have set up the upstream remote to point to our repository and origin to point to your fork.

This directory contains the following files:

scikit-image
├── asv.conf.json
├── azure-pipelines.yml
├── benchmarks/
├── CITATION.bib
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.rst
├── CONTRIBUTORS.txt
├── doc/
├── INSTALL.rst
├── LICENSE.txt
├── MANIFEST.in
├── meson.build
├── meson.md
├── pyproject.toml
├── README.md
├── RELEASE.txt
├── requirements/
├── requirements.txt
├── skimage/
├── TODO.txt
└── tools/

All commands below are assumed to be running from the scikit-image directory containing the files above.

2.1. Build environment setup#

Once you’ve cloned your fork of the scikit-image repository, you should set up a Python development environment tailored for scikit-image. You may choose the environment manager of your choice. Here we provide instructions for two popular environment managers: venv (pip based) and conda (Anaconda or Miniconda).

2.1.1. venv#

# Create a virtualenv named ``skimage-dev`` that lives outside of the repository.
# One common convention is to place it inside an ``envs`` directory under your home directory:
mkdir ~/envs
python -m venv ~/envs/skimage-dev
# Activate it
# (On Windows, please use ``skimage-dev\Scripts\activate``)
source ~/envs/skimage-dev/bin/activate
# Install main development and runtime dependencies
pip install -r requirements.txt
# Install build dependencies of scikit-image
pip install -r requirements/build.txt
# Build scikit-image from source
spin build
# Test your installation
spin test
# Build docs
spin docs
# Try the new version in IPython
spin ipython

2.1.2. conda#

When using conda for development, we recommend adding the conda-forge channel for the most up-to-date version of many dependencies. Some dependencies we use (for testing and documentation) are not available from the default Anaconda channel. Please follow the official conda-forge installation instructions before you get started.

# Create a conda environment named ``skimage-dev``
conda create --name skimage-dev
# Activate it
conda activate skimage-dev
# Install main development and runtime dependencies
conda install -c conda-forge --file requirements/default.txt
conda install -c conda-forge --file requirements/test.txt
conda install -c conda-forge pre-commit
# Install build dependencies of scikit-image
pip install -r requirements/build.txt
# Build scikit-image from source
spin build
# Test your installation
spin test
# Build docs
spin docs
# Try the new version
spin python

For more information about building and using the spin package, see meson.md.

2.2. Updating the installation#

When updating your installation, it is often necessary to recompile submodules that have changed. Do so with the following commands:

# Grab the latest source
git checkout main
git pull upstream main
# Update the installation
pip install -e . -vv

2.3. Testing#

scikit-image has an extensive test suite that ensures correct execution on your system. The test suite must pass before a pull request can be merged, and tests should be added to cover any modifications to the code base.

We use the pytest testing framework, with tests located in the various skimage/submodule/tests folders.

Our testing requirements are listed below:

asv
matplotlib>=3.5
pooch>=1.6.0
pytest>=7.0
pytest-cov>=2.11.0
pytest-localserver
pytest-faulthandler

Run all tests using:

pytest skimage

Or the tests for a specific submodule:

pytest skimage/morphology

Or tests from a specific file:

pytest skimage/morphology/tests/test_gray.py

Or a single test within that file:

pytest skimage/morphology/tests/test_gray.py::test_3d_fallback_black_tophat

Use --doctest-modules to run doctests. For example, run all tests and all doctests using:

pytest --doctest-modules skimage

2.3.1. Warnings during testing phase#

Scikit-image tries to catch all warnings in its development builds to ensure that crucial warnings from dependencies are not missed. This might cause certain tests to fail if you are building scikit-image with versions of dependencies that were not tested at the time of the release. To disable failures on warnings, export the environment variable SKIMAGE_TEST_STRICT_WARNINGS with a value of 0 or False and run the tests:

export SKIMAGE_TEST_STRICT_WARNINGS=False
pytest --pyargs skimage

2.4. Platform-specific notes#

Windows

A run-through of the compilation process for Windows is included in our setup of Azure Pipelines (a continuous integration service).

Debian and Ubuntu

Install suitable compilers:

sudo apt-get install build-essential

2.5. Full requirements list#

Build Requirements

# Also update `tools/pyproject.toml.in`, [build-system] -> requires
meson-python>=0.13
wheel
setuptools>=67
packaging>=21
ninja
Cython>=0.29.32
pythran
numpy>=1.21.1

# Developer UI
spin==0.3
build

Runtime Requirements

numpy>=1.21.1
scipy>=1.8
networkx>=2.8
pillow>=9.0.1
imageio>=2.27
tifffile>=2022.8.12
PyWavelets>=1.1.1
packaging>=21
lazy_loader>=0.2

Test Requirements

asv
matplotlib>=3.5
pooch>=1.6.0
pytest>=7.0
pytest-cov>=2.11.0
pytest-localserver
pytest-faulthandler

Documentation Requirements

sphinx>=5.0
sphinx-gallery>=0.11
numpydoc>=1.5
sphinx-copybutton
pytest-runner
matplotlib>=3.5
dask[array]>=2022.9.2
pandas>=1.5
seaborn>=0.11
pooch>=1.6
tifffile>=2022.8.12
myst-parser
ipywidgets
# Needed until https://github.com/jupyter-widgets/ipywidgets/issues/3731 is resolved
ipykernel
plotly>=5.10
kaleido
scikit-learn>=0.24.0
sphinx_design>=0.3
pydata-sphinx-theme>=0.13

Developer Requirements

pre-commit
rtoml

Data Requirements

The full selection of demo datasets is only available with the following installed:

pooch>=1.6.0

Optional Requirements

You can use scikit-image with the basic requirements listed above, but some functionality is only available with the following installed:

  • SimpleITK

    Optional I/O plugin providing a wide variety of formats. including specialized formats using in medical imaging.

  • Astropy

    Provides FITS I/O capability.

  • PyAMG

    The pyamg module is used for the fast cg_mg mode of random walker segmentation.

  • Dask

    The dask module is used to speed up certain functions.

SimpleITK
astropy>=5.0
# cloudpickle is necessary to provide the 'processes' scheduler for dask
cloudpickle>=0.2.1
dask[array]>=2021.1.0
matplotlib>=3.5
pooch>=1.6.0
pyamg
scikit-learn>=0.24.0

2.6. Help with contributor installation#

See Additional help above.