r20220512

Trained cascaded models (PyTorch)

Registration models trained with the SynthMorph method and converted from .h5 (TensorFlow/Keras) to .pt (PyTorch). The two models provided in this release (in cascaded_models_pytorch.zip) are the same as the ones provided in the release r20220329.

The configuration files used to train these models originally are present in cascaded_models_pytorch.zip. And the files used to convert the models from TensorFlow/Keras to Pytorch are also included in cascaded_models_pytorch.zip

---

An example script + example data to test these models is included in cascaded-registration-example.zip. To test this example follow these steps:

1. Download and extract cascaded-registration-example.zip.
2. Download and extract cascaded_models_pytorch.zip to the model/ folder inside cascaded-registration-example/.
3. Run the following commands:

cd cascaded-registration-example

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

python cascaded-registration.py

r20220329

Trained cascaded models

Registration models trained with the SynthMorph method. The two models provided in this release (in cascaded_models.zip) can be used successively in the cascaded approach to perform coarse registration and accommodate images that are far apart with the first model, and refine registration with the second model to ensure high accuracy of registration results. This can be done with the pipeline pipeline_bids_register_evaluate_two_steps.sh.

The configuration files used to train these models are also present in cascaded_models.zip.