kornia
Kornia AI is a non-profit building and maintaining an open-source stack for spatial AI: differentiable computer vision for research and real-time perception for robots. Its goal is to accelerate the research, development and deployment of computer-vision solutions and democratize access to AI, computer vision and free software.
At a Glance
- Computer-vision and machine-learning researchers
- Universities and academic labs
- AI/ML engineers and software developers using PyTorch
- Robotics, autonomous-systems and spatial-AI companies
- +2 more
AI Tools by kornia
(1)
Kornia
Differentiable Computer Vision for PyTorch
Discussions
No discussions yet
Be the first to start a discussion about kornia
Latest News
Kornia 0.9.0 released with correctness and performance improvements, better half precision, CUDA/MPS support, compile-first guidance and benchmark documentation.
Kornia 0.9.0rc1 released as a compile-first release candidate focused on deployable augmentations and torch.export/torch.compile testing.
Kornia 0.8.3 release activity published, including XFeat, ALIKED, VisualPrompter batched inference and expanded model/geometry improvements.
Kornia AI selected as a Google Summer of Code 2026 organization, continuing its Spatial AI, Rust, robotics and computer-vision mentoring program.
Products & Services
Apache-2.0 open-source differentiable computer-vision library for PyTorch, covering image processing, geometry, augmentation, feature detection/matching, losses, pretrained models and 3D vision.
Native Rust low-level 3D computer-vision library and edge runtime for high-performance, memory-efficient spatial-AI and robotics pipelines.
Open-source real-time visual-inertial SLAM and mapping project in the Kornia ecosystem.
Camera and stereo-driver project in the Kornia ecosystem.
Market Position
Kornia positions itself as a differentiable, batched and GPU/device-agnostic computer-vision stack for PyTorch, spanning classical geometry and learned vision rather than only inference. Its official FAQ contrasts it with OpenCV (Kornia focuses on differentiable operators for training; OpenCV primarily targets inference) and notes inspiration/behavioral compatibility with OpenCV and TorchVision. Its differentiable geometry, per-sample batched parameters, PyTorch integration, compile/export support and emerging Rust edge stack distinguish it from conventional CPU-centric libraries, while OpenCV, TorchVision and specialized feature/SLAM libraries remain competitors or alternatives.
Leadership
Founders
Edgar Riba
Creator and project leader of Kornia; a computer-vision research scientist at IRI-CSIC-UPC, PhD candidate at the Computer Vision Center at Universitat Autònoma de Barcelona, and a member of the OpenCV.org Technical Committee. The project’s package metadata lists him as author, and the original Kornia paper credits him as first author.
Dmytro Mishkin
Computer-vision and deep-learning researcher/consultant and co-author of the original Kornia paper; he is listed among Kornia’s project maintainers.
Daniel Ponsa
Computer-vision researcher and co-author of the original Kornia paper.
Ethan Rublee
Computer-vision engineer/researcher and co-author of the original Kornia paper.
Gary Bradski
Computer-vision researcher and OpenCV community leader; mentor/technical collaborator to Edgar Riba and co-author of the original Kornia paper.
Executive Team
Edgar Riba
Creator, project lead and Kornia AI co-founder/president
Computer-vision research scientist and PhD candidate; former/associated OpenCV technical-community member and leader of Kornia’s main library activities and core development.
Founding Story
The project began as PyTorch Geometry (torchgeometry), a small differentiable geometric-computer-vision package released around PyTorch DevCon 2018. It evolved into a broader computer-vision library and was renamed Kornia because of a naming conflict with PyTorch Geometric; the initial vision was to make computer-vision operators differentiable and usable inside PyTorch training pipelines.
Business Model
Revenue Model
The core software is free, open source and Apache-2.0 licensed. Kornia AI is funded by individual and corporate sponsorships through Open Collective and GitHub Sponsors, with optional logo placement, support contracts, priority support, roadmap access and advisory/partnership retainers for companies using Kornia in production.
Pricing Tiers
Name listed in BACKERS.md.
Logo on the organization README, repository READMEs, documentation site and PyPI.
Supporter benefits plus documentation banner, priority issues and a quarterly roadmap call.
Partner benefits plus roadmap input, support SLA and invoicing.
Exclusive single-company tier with top placement, launch shoutout and direct roadmap access.
Target Markets
- Computer-vision and machine-learning researchers
- Universities and academic labs
- AI/ML engineers and software developers using PyTorch
- Robotics, autonomous-systems and spatial-AI companies
- Industrial and edge-device perception teams
- Open-source projects and production teams needing differentiable or GPU-native vision
- Differentiable computer vision research and academic prototyping
- Training deep-learning models with GPU-native image processing and augmentation
- Image matching, registration, stitching and visual prompting
- Camera calibration, stereo, depth, 3D reconstruction, pose estimation and SLAM
- Robotics and spatial AI perception on embedded/Jetson-class edge devices
- Production deployment through torch.compile, ONNX Runtime or Rust edge pipelines
- Farm-ng
- Rerun.io
- Copper Robotics
- 17,565 public GitHub repositories and 321 packages declared kornia as a dependency as of 2026-09-02