Implementation of SynthMorph in PyTorch.
This project mainly adapts the source code from libraries VoxelMorph and Neurite.
Initially, this project is intended to reproduce the SynthMorph demo created by the paper authors. The original demo was done in TensorFlow (TF), and synthmorph_demo.ipynb aims to reproduce it using PyTorch.
Although the voxelmorph package provides the required modules in PyTorch, the registration output in PyTorch does not match the output from the TF implementation. Therefore, this repo also reimplements the required modules in PyTorch, faithfully following the TF implementation. In addition, this repo also provides weight transfer steps in tf2torch.ipynb.
Python 3.8.x versions are recommended when using code from this repo, but there is no promise of compatibility with other Python versions. The dependencies in requirements.txt are based on Linux installations for GPU. For Windows and/or CPU only packages, you might need to change how to install some packages, such as TensorFlow.
In your Python environment:
pip install -r requirements.txt- Implemented affine registration model (only for default parameters).
- Implemented affine data generation.