π Geometric Computer Vision Library for Spatial AI
Python artificial-intelligencecomputer-visiondeep-learninghacktoberfestimage-processingmachine-learningneural-networkpythonpytorchroboticsspatial-ai
Learn
=====
.. meta::
:description: Learn Kornia: install the library, follow a quickstart application guide, learn the conventions every function follows, and see what makes the library different.
New to Kornia? Three steps take you from zero to a working pipeline:
.. grid:: 1
:gutter: 2
:class-container: kornia-steps
.. grid-item-card:: 1 Β· Install
:link: installation
:link-type: doc
``pip install kornia`` β plus conda and from-source options, and how to check it worked.
.. grid-item-card:: 2 Β· QuickStart
:link: ../applications/image_augmentations
:link-type: doc
Pick the application closest to your task β augmentation, matching, stitching, registration,
detection, prompting or denoising β and adapt its pipeline.
.. grid-item-card:: 3 Β· Learn the conventions
:link: conventions
:link-type: doc
``(B, C, H, W)`` floats in ``[0, 1]``, ``(x, y)`` points vs ``(h, w)`` sizes, degrees vs radians.
Nearly every subtle Kornia bug is a convention mismatch β read this before writing code.
Looking for a specific function instead? The :doc:`API reference </api>` documents every module, and
:doc:`What is Kornia? <introduction>` gives the one-page overview of the whole library.
.. toctree::
:caption: Getting started
:hidden:
installation
introduction
conventions
camera-conventions
.. toctree::
:caption: Why Kornia?
:hidden:
gpu-acceleration
differentiability
onnx
edge
multi-framework-support
precision
Performance on Mac chips <performance>
.. toctree::
:caption: Applications
:hidden:
../applications/image_augmentations
../applications/image_matching
../applications/image_stitching
../applications/image_registration
../applications/face_detection
../applications/visual_prompting
../applications/image_denoising
.. toctree::
:caption: Resources
:hidden:
Tutorials <https://www.kornia.org/tutorials/>
ONNX, torch.compile & torch.export support <export-support>