3d_registration/learn2reg_oasis_unpaired_brain_mr.ipynb weights are missing

[lvora]

Where do I find the weights for the VoxelMorph pretrained network in learn2reg_oasis_unpaired_brain_mr? I don't seem to find them in the GIT as well as anywhere else on the repo

Section on "Load pretrained model and perform forward pass" has a way to load the pretrained model, however there are no weights. Has anyone found them?
/~https://github.com/Project-MONAI/tutorials/blob/main/3d_registration/learn2reg_oasis_unpaired_brain_mr.ipynb

[KumoLiu]

Hi @lvora, thank you for your interest! It appears that the original contributors did not provide the pretrained weights. The argument in the notebook is meant to demonstrate how to include a pretrained weight if you have one.

Also cc the contributor @kvttt

[kvttt]

Hi @lvora, I don't think we have pretrained weights for VoxelMorph in MONAI. I am sure there are pretrained weights for VoxelMorph out there somewhere. But it might take some effort to get these weights to work with the re-implementation of VoxelMorph in MONAI.

The argument in the notebook is meant to demonstrate how to include a pretrained weight if you have one.

And yes, the flag load_pretrained_model_weights = True is solely for demonstration purpose.

With that being said, you should be able to get a reasonably good model by simply running the entire notebook. And it shouldn't take too long, especially with amp_enabled = True. When I ran the notebook, I got reasonably good result at epoch 50, which took less than 2 hours on a RTX 3090.

Let me know if you have additional questions and I am happy to help!

[lvora]

Thank you, @kvttt makes sense.