What problem it solves

AI workflows need image-processing and geometric-vision routines that integrate with PyTorch while preserving differentiability. Hand-written implementations can become long or fragile, while implementations that leave the GPU can limit batched workflows.

Key features

• Differentiable image processing — The library provides image-processing operators for filters, transformations, enhancements, edge detection, morphology, color, and losses, with many designed for integration into deep-learning pipelines.

  Source: README, kornia

• Geometric vision and robust estimation — Its geometry components cover camera models, epipolar geometry, pose estimation, depth projection, homography, Lie groups, minimal solvers, and device-side RANSAC estimation.

  Source: README

• Annotated augmentation pipelines — AugmentationSequential applies one sampled transform consistently to images, masks, boxes, and keypoints. Its 2D pipelines can also be inverted.

  Source: README

• Local feature detection and matching — The feature component includes local detectors and descriptors, along with mutual-nearest-neighbor and ratio-test matchers for image correspondence.

  Source: README

• Compilation and ONNX export — Operator configurations are tested with torch.compile, and the library also supports exporting operators to ONNX through dedicated interfaces.

  Source: README, kornia

Quick start

Copied verbatim from the project's README.

  pip install kornia

  pip install -e .

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")

Who it may suit

This may suit PyTorch developers and researchers building differentiable image-processing, augmentation, geometric-vision, feature-matching, or ONNX-export workflows.

Topics to discuss

• Which geometric-vision contracts should receive additional conformance tests and reference implementations?

• How should the community organize coverage for CUDA and MPS half-precision support?

• Which augmentation and image-processing workflows would benefit from new annotated pipeline capabilities?

Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.

Read the original README https://github.com/kornia/kornia#readme

ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.