Kornia is a differentiable computer vision library that provides a rich set of differentiable image processing and geometric vision algorithms. Built on top of PyTorch, Kornia integrates seamlessly into existing AI workflows, allowing you to leverage powerful [batch transformations](), [auto-differentiation]() and [GPU acceleration](). Whether you're working on image transformations, augmentations, or AI-driven image processing, Kornia equips you with the tools you need to bring your ideas to life.

📢 Direction: Kornia is becoming the reference implementation and executable specification for differentiable computer vision and geometry in the PyTorch ecosystem — explicit conventions, conformance tests, and honest benchmarks over API growth. Read the Roadmap.

Key Components

• Filters: Gaussian, Sobel, Median, Box Blur, etc.
• Transformations: Affine, Homography, Perspective, etc.
• Enhancements: Histogram Equalization, CLAHE, Gamma Correction, etc.
• Edge Detection: Canny, Laplacian, Sobel, etc.
• ... check our docs for more.

• Augmentation Pipeline: AugmentationSequential, PatchSequential, VideoSequential, etc.
• Automatic Augmentation: AutoAugment, RandAugment, TrivialAugment.

• Face Detection: YuNet
• Feature Matching: LoFTR, LightGlue
• Feature Descriptor: DISK, DeDoDe, SOLD2
• Segmentation: SAM
• Classification: MobileViT, VisionTransformer.

Half-Precision Support

The status below comes from the Linux CPU half-precision CI jobs, which run the whole test suite in float16 and in bfloat16 against strict known-failure manifests (cpu_float16.txt, cpu_bfloat16.txt). The known failures column counts the manifest entries per module; the remaining work is tracked in #4153. No CI job covers CUDA or MPS half precision.

| Module | float16 | bfloat16 | Known failures (fp16 / bf16) | Notes | |--------|:-------:|:--------:|:----------------------------:|-------| | kornia.color | ⚠️ | ⚠️ | 3 / 9 | float16: HLS JIT/module and RGB255 round-trip accuracy; bfloat16: Lab, Luv and RGB255 accuracy | | kornia.filters | ⚠️ | ⚠️ | 18 / 9 | Accuracy misses in Canny magnitudes, discrete Gaussian kernels and Otsu; on CPU fft_conv runs its FFTs in float32 | | kornia.enhance | ✅ | ⚠️ | 0 / 2 | bfloat16: DiffJPEG and ZCA accuracy | | kornia.morphology | ✅ | ✅ | 0 / 0 | | | kornia.augmentation | ⚠️ | ⚠️ | 205 / 83 | float16: 108 entries are CutmixGenerator, whose Dirichlet sampling rejects float16 parameters; bfloat16: RandomJigsaw, RandomCutMixV2, RandomMixUpV2 and RandomMosaic raise KeyError: 'BFLOAT16' (#4467) | | kornia.geometry.transform | ⚠️ | ⚠️ | 43 / 58 | Accuracy misses in rotation matrices, affine/perspective warps, the homography warper and 3D crops | | kornia.geometry.camera | ⚠️ | ⚠️ | 13 / 23 | Pinhole cam2pixel/pixel2cam consistency, distortion round trips, StereoCamera reprojection; 12 bfloat16 entries are a test-side dtype assertion | | kornia.geometry.calibration | ⚠️ | ⚠️ | 13 / 12 | solve_pnp_dlt rejects half inputs (float32/float64 only); undistort_points misses its OpenCV reference values | | kornia.geometry.epipolar | ⚠️ | ⚠️ | 58 / 56 | find_fundamental, find_essential, decompose_essential_matrix, motion_from_essential* and KRt_from_projection raise NotImplementedError: CPU lu, eigh and QR have no half kernels | | kornia.geometry.homography | ⚠️ | ⚠️ | 11 / 16 | The DLT solvers run (SVD is cast to float32) but miss the clean-point accuracy checks | | kornia.geometry.liegroup | ⚠️ | ⚠️ | 36 / 130 | So2/Se2 use complex tensors: float16 hits missing ComplexHalf kernels, and most bfloat16 So2/Se2 tests raise (119 entries); So3/Se3 nearly all pass | | kornia.geometry.solvers | ⚠️ | ⚠️ | 2 / 2 | solve_quartic accuracy on random and one reference quartic | | kornia.geometry.subpix | ⚠️ | ⚠️ | 14 / 12 | ConvSoftArgmax3d raises (CPU avg_pool3d has no half kernel); the rest are accuracy | | kornia.geometry.conversions | ⚠️ | ⚠️ | 72 / 60 | Angle-axis, quaternion and rotation-matrix round trips lose accuracy | | kornia.geometry.ransac | ⚠️ | ⚠️ | 4 / 4 | The essential and fundamental models raise through the epipolar solvers | | kornia.geometry (other) | ⚠️ | ⚠️ | 8 / 17 | Accuracy in boxes, depth and line utilities; bfloat16 NamedPose construction raises | | kornia.image | ⚠️ | ⚠️ | 4 / 4 | draw_convex_polygon fill accuracy | | kornia.losses | ⚠️ | ⚠️ | 3 / 4 | Dice averaging overflows to inf/NaN in float16; mutual information range check; bfloat16 Dice weighting and total variation | | kornia.feature | ✅ | ✅ | 0 / 0 | Matching uses a manual cdist fallback for half dtypes; LightGlue's float16 tests are skipped, so that path is unmeasured | | kornia.metrics | ✅ | ⚠️ | 0 / 1 | bfloat16: ssim3d accuracy | | kornia.models | ⚠️ | ⚠️ | 15 / 7 | EfficientViT (float16) and Kimi-VL MoonViT raise dtype mismatches; bfloat16 RT-DETR RepVGG fusion accuracy | | contrib, core, io, onnx, sensors, tracking, utils | ✅ | ⚠️ | 0 / 3 | bfloat16: histogram matching, _torch_svd_cast, camera-model projection |

✅ No known CPU failures   ⚠️ Runs, with known failures (mostly accuracy; notes name the ops that raise)

Test results:

| Run | Passed | Failed | Skipped | Pass% | Measured | |-----|-------:|-------:|--------:|------:|----------| | CPU float32 (baseline) | 10398 | 0 | 3737 | 100.0% | ca5021eb, 2026-09-14 | | CPU float16 | 9795 | 522 | 3821 | 94.9% | ca5021eb, 2026-09-14 | | CPU bfloat16 | 9849 | 512 | 3774 | 95.1% | ca5021eb, 2026-09-14 | | CUDA float32 (baseline) | 7634 | 3 | 3280 | 99.9% | 6131e98, 2026-03-21 | | CUDA float16 (KORNIA_TEST_IN_SUBPROCESS=1) | 6727 | 643 | 3556 | 91.3% | 6131e98, 2026-03-21 | | CUDA bfloat16 (KORNIA_TEST_IN_SUBPROCESS=1) | 6695 | 713 | 3518 | 90.4% | 6131e98, 2026-03-21 |

Pass% = passed ÷ (passed + failed). The CPU rows are the nightly main CI jobs (Linux x86_64, Python 3.11, PyTorch 2.9.1, no --runslow). In the half jobs, Failed is the manifest's entry count: CI reports those tests as strict xfails, and it fails if any of them passes or fails differently. Tests marked xfail in the source are excluded from every row. Reproduce a CPU half row in that environment with KORNIA_TEST_OPTIMIZER= pixi run test-module tests/ --verify-known-failures --known-failure-profile=cpu-float16 (or cpu-bfloat16), and the baseline with pixi run test-f32. pixi run test-half is an unseeded sweep of both dtypes whose counts can drift slightly from the manifests. The CUDA rows have not been re-measured since March 2026 and predate the CPU half-precision fixes merged since then.

See the full precision guide for details.

Sponsorship

Kornia is an open-source project that is developed and maintained by volunteers. Whether you're using it for research or commercial purposes, consider sponsoring or collaborating with us. Your support will help ensure Kornia's growth and ongoing innovation. Reach out to us today and be a part of shaping the future of this exciting initiative!

Installation

From pip

  pip install kornia
  

Some features (ONNX, Stable Diffusion dissolving) need extra packages; see Optional extras.

  pip install -e .
  

  # Install pixi (if not already installed)
  curl -fsSL https://pixi.sh/install.sh | bash
# Create the Pixi environment and install development dependencies
  pixi install
  pixi run install
# Run tests
  pixi run test
# For CUDA development
  pixi run -e cuda install
  pixi run -e cuda test-cuda
  

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

Quick Start

Kornia is not just another computer vision library — it's your gateway to effortless Computer Vision and AI.

import numpy as np
import kornia_rs as kr
from kornia.augmentation import AugmentationSequential, RandomAffine, RandomBrightness
from kornia.filters import StableDiffusionDissolving
Load and prepare your image
img: np.ndarray = kr.read_image_any("img.jpeg")
img = kr.resize(img, (256, 256), interpolation="bilinear")
alternatively, load image with PIL
img = Image.open("img.jpeg").resize((256, 256))
img = np.array(img)
img = np.stack([img] * 2)  # batch images
Define an augmentation pipeline
augmentation_pipeline = AugmentationSequential(RandomAffine((-45.0, 45.0), p=1.0), RandomBrightness((0.0, 1.0), p=1.0))
Leveraging StableDiffusion models
dslv_op = StableDiffusionDissolving()
img = augmentation_pipeline(img)
dslv_op(img, step_number=500)
dslv_op.save("Kornia-enhanced.jpg")

import numpy as np
from kornia.onnx import ONNXSequential
Chain ONNX models from HuggingFace repo and your own local model together
onnx_seq = ONNXSequential(
    "hf://operators/kornia.geometry.transform.flips.Hflip",
    "hf://models/kornia.models.detection.rtdetr_r18vd_640x640",  # Or you may use "YOUR_OWN_MODEL.onnx"
)
Prepare some input data
input_data = np.random.randn(1, 3, 384, 512).astype(np.float32)
Perform inference
outputs = onnx_seq(input_data)
Print the model outputs
print(outputs)
Export a new ONNX model that chains up all three models together!
onnx_seq.export("chained_model.onnx")

Call For Contributors

If kornia is useful to you and you would like to help, contributions of many kinds are welcome: code, bug reports, benchmarks, documentation, questions, answers, and examples. The maintainers have limited time, so we cannot promise that every proposal or pull request will be reviewed or merged.

Strengthen the Core (Priority)

Kornia's differentiated value is its geometry core: warping and sampling, homographies, cameras, epipolar geometry, rotations and Lie groups, and geometry-consistent augmentation. The highest-impact contributions make that core more trustworthy — see the Roadmap for the full picture. Great entry points:

• Benchmark results from your hardware: run the benchmark suite with --contribute and send the JSON — CUDA numbers from diverse GPUs are especially useful.
• Convention pinning tests and conformance vectors for core geometry operations.
• Corrective error messages: upgrade bare shape assertions into errors that state what was wrong, what was expected, and which convention applies.
• Classical vision in the core domain: camera intrinsic calibration, fiducial markers (ArUco/ChArUco), classical tracking, dense stereo, and Hough transforms. An early design discussion can be useful for work of this size.

AI Models

The model zoo is currently frozen for expansion while maintainer bandwidth concentrates on the core. Shipped models (LoFTR, LightGlue, DISK, DeDoDe, SAM, and friends) stay available and maintained, and model work approved before the freeze (Efficient LoFTR, SANDesc) will be completed under its existing scope. New integrations — including VLM/VLA models — require a named maintainer sponsor who accepts ongoing ownership of the integration; a contributor implementation alone cannot reopen the surface. See the Roadmap for the reasoning and the reopen condition.

Documentation And Tutorial Optimization

Kornia's foundation lies in its extensive collection of classic computer vision operators, providing robust tools for image processing, feature extraction, and geometric transformations. We continuously seek for contributors to help us improve our documentation and present nice tutorials to our users.

Cite

If you are using kornia in your research-related documents, it is recommended that you cite the paper. See more in CITATION.

  @inproceedings{eriba2019kornia,
    author    = {E. Riba, D. Mishkin, D. Ponsa, E. Rublee and G. Bradski},
    title     = {Kornia: an Open Source Differentiable Computer Vision Library for PyTorch},
    booktitle = {Winter Conference on Applications of Computer Vision},
    year      = {2020},
    url       = {https://arxiv.org/pdf/1910.02190.pdf}
  }
  

Contributing

See CONTRIBUTING.md for our social contract, development setup, and technical guidelines. Participation is subject to the Code of Conduct.

Community

• Discord: talk with people who use and develop kornia. Join the server
• GitHub Issues: report bugs and propose concrete changes. Open an issue
• GitHub Discussions: ask questions and discuss implementations, research, and ideas. Join a discussion

Made with contrib.rocks.

License

Kornia is released under the Apache 2.0 license. See the LICENSE file for more information.