kornia/kornia

Features

• Computer Vision Libraries - Provides a comprehensive set of computer vision tools that operate across multiple tensor backends like PyTorch, TensorFlow, JAX, and NumPy.
• Cross-Framework Tensor Dispatch - Abstracts tensor operations to allow the same vision logic to run across PyTorch, TensorFlow, JAX, and NumPy.
• Differentiable Vision Operations - Implements computer vision operations as differentiable tensors to enable seamless integration into deep learning pipelines.
• 3D Spatial AI - Estimates camera pose, calibration, and depth from stereo images to reconstruct three dimensional scenes.
• Augmentation Pipelines - Creates sequences of random geometric transformations and noise injection to increase dataset diversity for training.
• Differentiable Operations - Implements vision algorithms as differentiable tensors, enabling their direct integration into deep learning training pipelines.
• Image Augmentation - Applies random spatial and color transformations as differentiable layers for neural network training pipelines.
• Feature Extraction - Detects keypoints and extracts descriptors to match image features and analyze topology.
• Framework Transpilers - Translates vision operations to run on different backends including JAX, TensorFlow, and NumPy.
• Differentiable Computer Vision - Integrates geometric vision operations into deep learning pipelines using tensors that support gradient computation for model training.
• Geometric Vision Tools - Offers a specialized toolbox for spatial AI, including camera calibration and 3D scene reconstruction using PyTorch tensors.
• Differentiable Vision Operators - Implements computer vision algorithms as differentiable functions to enable gradient flow through geometric transformations.
• Feature Detection - Detects and describes keypoints in images using algorithms for feature matching tasks.
• Camera Calibration - Provides tools for camera calibration and pose estimation to determine spatial relationships between images.
• Local Feature Matching - Identifies keypoints in images and matches them across frames using descriptors and geometric algorithms.
• Epipolar Geometry - Calculates homographies and epipolar geometry to resolve spatial relationships between multiple image perspectives.
• Camera Projection Systems - Transforms coordinates between 3D world space and 2D image planes using perspective projection techniques.
• Stereo Vision Reconstruction - Computes disparity and depth estimation from stereo image pairs to reconstruct three-dimensional scenes.
• Camera Geometry Estimation - Calculates camera calibration and epipolar geometry to analyze spatial relationships between images.
• Cross-Framework Tensor Execution - Runs vision operations across different tensor libraries including TensorFlow, JAX, and NumPy.
• Depth Estimation - Includes algorithms for computing disparity maps and reconstructing 3D scenes from stereo image pairs.
• Homography Estimation - Computes perspective transformations between images using line correspondences or iterative RANSAC methods.
• Image Segmentation - Isolates specific objects or image regions using connected component labeling and other segmentation techniques.
• ONNX Model Exporters - Converts complex vision pipelines into the standardized ONNX format for cross-platform hardware inference.
• Vision Model Training - Provides specialized training pipelines for computer vision tasks such as object detection and semantic segmentation.
• Vision-Language Models - Combines computer vision operations with large language models to process both images and text.
• Vision Loss Functions - Calculates photometric and structural similarity metrics to optimize the accuracy of image-based deep learning models.
• Optical Flow Computation - Determines the motion of pixels between frames using both dense and sparse optical flow algorithms.
• Image Registration - Aligns multiple images into a single coordinate system using registration and stitching tools.
• Edge Detection Algorithms - Identifies boundaries within images using Canny edge detection and Laplacian pyramids.
• 3D Geometry Utilities - Provides calculations for transformations, scaling, and rotations within three-dimensional coordinate systems.
• Image Filters - Applies smoothing and denoising operations using bilateral and unsharp mask filters.
• Lie Group Operations - Performs algebraic operations on SO2 and SE2 groups including adjoints and translations.
• Lie Group Algebra - Uses SO(n) and SE(n) group theory to handle rotations and translations within a consistent mathematical framework.
• Quaternion Rotation Utilities - Executes spherical linear interpolation and rotational operations using quaternion algebra.
• Computer Vision - Differentiable computer vision library for spatial AI.
• Computer Vision Frameworks - Differentiable library for image processing and geometric vision.
• Deep Learning and Computer Vision - Differentiable computer vision library for PyTorch.
• PyTorch Ecosystem - Differentiable computer vision library.
• Image Processing and Manipulation - Differentiable computer vision library for GPU-accelerated image transformations.

Star history