ASTRA KernelKit is part of the ASTRA Toolbox project.
Note
Last update is August 8th to verison 1.0.0rc2. This contains updated 3D conebeam forward and backprojectors and some documentation. Parallel beam and more examples are to follow soon.
Find documentation at https://kernelkit.gra.as/.
ASTRA KernelKit is an all-Python tomographic reconstruction package, leveraging the GPU-accelerated tomographic projection kernels of the ASTRA Toolbox using CuPy.
- Develop, explore and debug algorithms and projectors with the Python ecosystem.
- Use CUDA features such as graphs and unified memory for maximum performance.
- Kernels are compiled during the script with NVRTC and Jinja2. No complicated build system, and yet easy customization of kernels.
- Easy to install and integrates well with PyTorch and other GPU frameworks.
KernelKit depends on CuPy, which we recommend to be installed first. Refer to our installation instructions.
See the documentation for more information.
The ASTRA KernelKit and ASTRA Toolbox projects are freely available and open source under the GPLv3 license. See the LICENSE.md file for details.
Please use Github pull requests (contributions welcome!) and issue tracker for bug reports and feature requests. For other questions, please contact the authors.
Copyright (c) 2010-2023, imec Vision Lab, University of Antwerp 2014-2023, CWI, Amsterdam.