Installation#

Kornia is distributed as pure-Python wheels on PyPI and on conda-forge. It requires PyTorch 2.5.1 or newer; the only other dependencies are numpy and kornia-rs (the Rust image I/O backend used by kornia.io). Install PyTorch first if you need a specific CUDA build.

pip

pip install kornia

conda

conda install -c conda-forge kornia

From source

pip install git+https://github.com/kornia/kornia

or, from a local clone, an editable install for development:

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

Once the installation has finished, check that you can import the package:

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

Pretrained models (RT-DETR, LoFTR, DISK, SAM, …) download their checkpoints on first use, so no extra installation step is needed for them.

Optional extras#

A few Kornia features wrap third-party packages that are not installed with the base wheel. They are declared as extras, so you only pay for the ones you use. If one of the extras below is missing, the corresponding Kornia object raises an ImportError naming the extra to install. The installation mode changes this: set kornia.config.kornia_config.lazyloader.installation_mode, or the KORNIA_INSTALLATION_MODE environment variable before Kornia is imported, to "ask" to be asked on an interactive terminal whether to install the extra, or to "auto" to install the declared extra with pip install "kornia[<extra>]" without asking. Without an interactive terminal, "ask" raises the same ImportError: for example in a CI job, when output is redirected to a file, or in a Jupyter notebook, whose kernel’s stdin is not a terminal (use "auto" there). The default is "raise".

Extra	What it enables	Install command
`image`	Pillow-backed PIL input/output and display helpers, plus remote image decoding in `kornia.io.get_sample_images()`	`pip install "kornia[image]"`
`onnx`	kornia.onnx, ONNX export of Kornia modules, and `OnnxLightGlue`	`pip install "kornia[onnx]"`
`sd`	`kornia.filters.StableDiffusionDissolving`	`pip install "kornia[sd]"`
`dev`	Contributor environment: the test and lint toolchain (includes `kornia[onnx]`)	`pip install -e ".[dev]"`
`docs`	Documentation toolchain, on top of `dev`	`pip install -e ".[dev,docs]"`

Next steps#

What is Kornia? – what Kornia is and what each module contains.
Conventions & Pitfalls – the tensor layout, coordinate and angle conventions to know before writing code.
Applications – end-to-end guides, or the API reference.