kornia
🐍 Geometric Computer Vision Library for Spatial AI
What is kornia?
🐍 Geometric Computer Vision Library for Spatial AI
Overview of kornia
What it is
Kornia is a differentiable computer vision library built on top of PyTorch. It provides a collection of differentiable image processing and geometric vision algorithms, designed to integrate into existing AI workflows with batch transformations, auto-differentiation, and GPU acceleration.
Key features
- Differentiable image processing operators: filters (Gaussian, Sobel, Median), transformations (Affine, Homography, Perspective), enhancements (Histogram Equalization, CLAHE, Gamma Correction), and edge detection (Canny, Laplacian, Sobel).
- Advanced augmentations: AugmentationSequential, PatchSequential, VideoSequential, AutoAugment, RandAugment, TrivialAugment.
- AI models: face detection (YuNet), feature matching (LoFTR, LightGlue), feature descriptors (DISK, DeDoDe, SOLD2), segmentation (SAM), classification (MobileViT, VisionTransformer).
- Batched minimal solvers and robust estimation: find_fundamental, find_essential, solve_pnp_dlt, RANSAC for homography, fundamental, and essential models.
- Differentiable geometry: depth to 3D and back, warp_frame_depth, remap, camera models with Brown–Conrady and Kannala–Brandt distortion, Lie groups (So3, Se3).
- Local features on GPU: SIFT, HardNet, DISK, ALIKED, XFeat, DeDoDe, LoFTR, LightGlue with mutual-nearest-neighbour and ratio-test matchers.
Use cases
Kornia is suited for scenarios where a hand-written implementation becomes long or fragile, or where alternatives leave the GPU unused. It is used for breadth (1,200+ public operators), per-sample parameters in one batched call, batched minimal solvers and robust estimation, differentiable geometry, annotated augmentation (2D and 3D), local feature extraction and matching on GPU, morphology with arbitrary structuring elements, and compiling/exporting to ONNX. It is designed to be fast on batched tensors on a GPU.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
License: Apache-2.0. Pricing: open source.
Based on the project's public README and repository metadata.