kornia/kornia-rs

🦀 Low-level 3D Computer Vision library in Rust

What it solves

Kornia-rs provides a high-performance, low-level foundation for computer vision tasks in Rust. It addresses the need for fast, thread-safe image processing that can seamlessly bridge the gap between raw camera data and machine learning inference engines like PyTorch or TensorRT without unnecessary data copying.

How it works

It uses a unified API that automatically dispatches operations to either the CPU or an NVIDIA GPU based on where the data resides. By leveraging SIMD (NEON/AVX2) for CPU tasks and custom CUDA kernels for GPU tasks, it enables real-time pipelines. It supports zero-copy interoperability through standards like DLPack and the CUDA Array Interface, allowing processed frames to be passed directly into deep learning models.

Who it’s for

This library is designed for developers building real-time computer vision applications, such as camera-to-inference pipelines, robotics, or scientific imaging, who require both the memory safety of Rust and the raw performance of hardware-accelerated processing.

Highlights

  • Written in Rust for memory and thread safety with no Global Interpreter Lock (GIL).
  • Unified API that handles both CPU and GPU execution automatically.
  • Zero-copy integration with PyTorch and TensorRT.
  • Real-time ready with V4L2 camera capture and fused CUDA kernels for preprocessing.
  • Supports a wide range of image formats including JPEG (via libjpeg-turbo), PNG, TIFF, and WebP.
  • Python bindings available, supporting both standard and free-threaded Python builds.

Related

  • Project
    image-rs/imageproc

    imageproc is a Rust library that provides a suite of image processing functions designed to serve as a reliable foundation for computer vision and graphics applications.

  • Project
    jamjamjon/usls

    A cross-platform Rust library powered by ONNX Runtime for efficient inference of vision and vision-language models under 1B parameters.

  • Project
    robertknight/ocrs

    A Rust library and CLI tool for extracting text from images using neural network models to minimize the need for image preprocessing.

  • Project
    kornia/kornia

    Kornia is a differentiable computer vision library for PyTorch, offering over 500 image processing, geometric, and augmentation operators that support auto-differentiation, batching, and GPU acceleration, along with pre-trained models for faces, features, and segmentation.

  • Project
    SomeB1oody/RustyML

    RustyML is a pure Rust library for machine learning and deep learning that provides tools for data preprocessing, model training, and performance evaluation.