Kornia
A differentiable geometric computer vision library built on PyTorch. It provides differentiable operators for image filtering, geometric transformations, camera models and 3D projection, and supports end-to-end training.
Timeline (last 90 days)
Kornia is a Python library based on PyTorch that provides differentiable geometric computer vision operators.
According to its GitHub page, Kornia has gained more than 11.5k stars and 1.4k forks.
Kornia recently gained 88 new stars in a single day.
Kornia's operators are built on top of PyTorch, gradients can propagate backward, and they can be seamlessly embedded into deep learning training pipelines.
At the image level, Kornia includes differentiable implementations of filtering, edge detection, and morphological operations.
At the geometry level, Kornia supports homography transformations, fundamental matrix estimation, camera calibration, and distortion correction.
At the 3D level, Kornia involves point cloud processing and conversion between depth maps and 3D projections.
Kornia may be less comprehensive than OpenCV in advanced algorithms for traditional image processing (such as certain specific feature descriptors).
A common combination in practice is to use OpenCV for offline calibration, traditional feature extraction, or visualization, and use Kornia for the differentiable geometry layer during training and inference.
The cost of Kornia's differentiable implementations is that some operators have extra overhead compared with C++ implementations, and the benefit is that they can directly participate in end-to-end optimization.
1 more timeline events
All Facts (11)
Kornia is a Python library based on PyTorch that provides differentiable geometric computer vision operators.
50%UnverifiedAccording to its GitHub page, Kornia has gained more than 11.5k stars and 1.4k forks.
50%UnverifiedKornia recently gained 88 new stars in a single day.
50%UnverifiedKornia's operators are built on top of PyTorch, gradients can propagate backward, and they can be seamlessly embedded into deep learning training pipelines.
50%UnverifiedAt the image level, Kornia includes differentiable implementations of filtering, edge detection, and morphological operations.
50%UnverifiedAt the geometry level, Kornia supports homography transformations, fundamental matrix estimation, camera calibration, and distortion correction.
50%UnverifiedAt the 3D level, Kornia involves point cloud processing and conversion between depth maps and 3D projections.
50%UnverifiedKornia may be less comprehensive than OpenCV in advanced algorithms for traditional image processing (such as certain specific feature descriptors).
50%UnverifiedA common combination in practice is to use OpenCV for offline calibration, traditional feature extraction, or visualization, and use Kornia for the differentiable geometry layer during training and inference.
50%UnverifiedThe cost of Kornia's differentiable implementations is that some operators have extra overhead compared with C++ implementations, and the benefit is that they can directly participate in end-to-end optimization.
50%UnverifiedKornia provides differentiable estimation and decomposition tools for homography matrices and fundamental matrices, including decomposing the essential matrix into rotation and translation.
50%