Kornia: geometric computer vision for spatial AI
Kornia is an Apache-2.0 Python library for geometric computer vision, used for image and video processing in spatial AI pipelines, with 11,821 GitHub stars.
TL;DR: Kornia is a geometric computer vision library for spatial AI, written in Python and licensed under Apache-2.0. The topics are
artificial-intelligence,computer-vision,deep-learning,image-processingandhacktoberfest, and the documentation site is kornia.readthedocs.io. The project was created in August 2018 and has 11,821 stars and 1,468 forks.
What it is
The repository description is short: “Geometric Computer Vision Library for Spatial AI”.
That phrasing narrows the scope in a useful way. Kornia is not a general image processing toolbox and not a model zoo. It is about geometry, the part of computer vision that deals with where things are in space, how they moved, and how to correct for viewpoint and lens effects.
That focus is what separates it from the usual pile of vision utilities. If your problem is homographies, camera calibration, pose estimation, optical flow or image warping, this is a library built around those specifically, with the kind of API consistency you get when a project has one job rather than thirty.
The “spatial AI” framing tells you the intended context. Spatial AI means systems that reason about physical space, which means robots, autonomous vehicles, AR devices and anything else that has to understand where a thing is and how the camera saw it.
PyTorch-native and differentiable
The language is Python, and the topic list includes deep-learning, which in practice means the library is built on PyTorch and its operations are differentiable. That is not a minor implementation detail.
If your vision pipeline is a PyTorch model and you need geometric operations inside it, differentiable geometry means you can put the warp inside the graph and backpropagate through it. A NumPy implementation of the same transform is a wall you cannot train through, and it is the reason many geometric CV pipelines end up with custom autograd functions written by hand.
Kornia exists so you do not have to write those yourself.
The module layout
Kornia is organised as a set of top-level namespaces, which is how you normally navigate it.
| Namespace | Covers |
|---|---|
kornia.augmentation | Differentiable image augmentations for training pipelines |
kornia.geometry | Camera models, calibration, pose, homography |
kornia.feature | Feature detection and descriptors |
kornia.filters | Image filters, including spatial and edge-aware ones |
kornia.color | Color space conversions |
kornia.losses | Loss functions for vision tasks |
kornia.utils | Grid creation, tensor helpers, image building |
A realistic pipeline might pull an augmentation from kornia.augmentation, a homography from kornia.geometry, and a loss from kornia.losses, all inside one training loop.
How to try it
pip install kornia
Then a first call is usually an augmentation in a training loop:
import torch
import kornia.augmentation as K
augment = K.RandomHorizontalFlip(p=0.5)
images = torch.rand(8, 3, 224, 224)
batch = augment(images)
Install the extras explicitly if you need the heavier parts, since some modules pull in optional dependencies:
pip install kornia[all]
The documentation host in the repository metadata is kornia.readthedocs.io, which is where the API reference and the gallery of examples live.
When to use it
Kornia fits when you need reliable geometric vision operations that play nicely with PyTorch, particularly inside a training loop or a differentiable pipeline. It is a strong choice for robotics and vision work where geometry is the core problem rather than an afterthought.
It is a poor fit if you want an end-to-end model. Kornia is a library of operations, not a segmentation or detection model, and it does not compete with those repositories.
It is also a poor fit for pure image processing on ordinary images where you want something quick, since OpenCV and Pillow are already there and lighter.
Source and Accuracy Notes
- Project page: https://github.com/kornia/kornia
- Source repository: kornia/kornia (verified via GitHub API
html_url) - License: Apache-2.0 (verified via GitHub API
license.spdx_id) - Stars at time of writing: 11821
- Homepage listed in repository metadata: https://kornia.readthedocs.io
- Repository created: 2018-08-22
Related Posts
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026
dev-tools
Baguette iOS Simulator Automation Guide
Set up Baguette for iOS Simulator automation, web dashboards, device farms, gesture input, streaming, and camera testing with Xcode caveats.
5/28/2026