pytorch-complex 0.1.1

Complex Modules for PyTorch

pytorch-complex

Install it using pip:

pip install pytorch-complex

Usage: Similar to PyTorch. For using the Complex features of this library, just change the regular torch imports with torchcomplex imports. For example: import torchcomplex.nn as nn instead of import torch.nn as nn Then, simply nn.Conv2d for both torch and torchcomplex, for 2D Convolution

Credits

If you like this repository, please click on Star!

If you use this package or benift from the codes of this repo, please cite the following in your publications:

Soumick Chatterjee, Chompunuch Sarasaen, Alessandro Sciarra, Mario Breitkopf, Steffen Oeltze-Jafra, Andreas Nürnberger, Oliver Speck: Going beyond the image space: undersampled MRI reconstruction directly in the k-space using a complex valued residual neural network (ISMRM, May 2021)

BibTeX entry:

@inproceedings{mickISMRM21ksp,
      author = {Chatterjee, Soumick and Sarasaen, Chompunuch and Sciarra, Alessandro and Breitkopf, Mario and Oeltze-Jafra, Steffen and Nürnberger, Andreas and                     Speck, Oliver},
      year = {2021},
      month = {05},
      pages = {1757},
      title = {Going beyond the image space: undersampled MRI reconstruction directly in the k-space using a complex valued residual neural network},
      booktitle={2021 ISMRM \& SMRT Annual Meeting \& Exhibition}
}

Thank you so much for your support.

History

0.1.1 (2023-06-16)

• modReLU implementation changed
• New activation functions added: Hirose and modSigmoid

0.1.0 (2022-11-06)

• Default behaviour of complex_weights has been changed to True
• Basic max and avg functional callers added
• Bug fix with the new pooling functionals

0.0.8 (2020-12-08)

• Complex Weight Initializations added (Few basics and trabelsi_standard, trabelsi_independent)

0.0.1 (2020-11-06)

• First release (of the package) on PyPI.
• torchcomplex.nn -> Convolutions, Linears, Dropout, Pooling, BatchNorm, Few Actications
• Untested first version, could be buggy

0.0.0 (2020-11-06)

• First release (basic structure, not the actual package code) on PyPI.