tensorflow-nufft 0.12.0

A fast, native non-uniform FFT op for TensorFlow.

TensorFlow NUFFT

TensorFlow NUFFT is a fast, native non-uniform fast Fourier transform op for TensorFlow. It provides:

• Fast CPU/GPU kernels. The TensorFlow framework automatically handles device placement as usual.
• A simple, well-documented Python interface.
• Gradient definitions for automatic differentiation.
• Shape functions to support static shape inference.

The underlying algorithm is based on the NUFFT implementation by the Flatiron Institute. Please refer to FINUFFT and cuFINUFFT for more details.

Installation

You can install TensorFlow NUFFT with pip:

pip install tensorflow-nufft

Note that only Linux wheels are currently being provided.

TensorFlow compatibility

Each TensorFlow NUFFT release is compiled against a specific version of TensorFlow. To ensure compatibility, it is recommended to install matching versions of TensorFlow and TensorFlow NUFFT according to the table below.

TensorFlow NUFFT Version	TensorFlow Compatibility	Release Date
v0.12.0	v2.11.x	Nov 27, 2022
v0.11.0	v2.10.x	Oct 12, 2022
v0.10.1	v2.10.x	Sep 26, 2022
v0.10.0	v2.10.x	Sep 7, 2022
v0.9.0	v2.9.x	Sep 5, 2022
v0.8.1	v2.9.x	Jun 23, 2022
v0.8.0	v2.9.x	May 20, 2022
v0.7.3	v2.8.x	May 4, 2022
v0.7.2	v2.8.x	Apr 29, 2022
v0.7.1	v2.8.x	Apr 6, 2022
v0.7.0	v2.8.x	Feb 8, 2022
v0.6.0	v2.7.x	Jan 27, 2022
v0.5.0	v2.7.x	Dec 12, 2021
v0.4.0	v2.7.x	Nov 8, 2021
v0.3.2	v2.6.x	Aug 18, 2021
v0.3.1	v2.6.x	Aug 18, 2021
v0.3.0	v2.6.x	Aug 13, 2021

Usage

Once installed, you can perform NUFFTs in your TensorFlow code simply as:

import tensorflow_nufft as tfft

outputs = tfft.nufft(inputs, points)

See the documentation for the tfft.nufft function to learn more about the different parameters.

Documentation

Visit the docs for the API reference and examples of usage.

Issues

If you use this package and something does not work as you expected, please file an issue describing your problem. We're here to help!

Credits

If you find this software useful in your research, please cite us.

Contributors

See CONTRIBUTORS for a list of people who have contributed to this project. Thank you!

All contributions are welcome. If there is an issue you would like to address or a feature you would like to add, you might want to begin by commenting on the corresponding issue (or if it doesn't exist yet, by creating a new issue).