Training on own data

[desaibhargav]

Hey there, thanks for sharing this work and many congratulations ❤️

I have a problem statement to register 3D organs (one kidney to another).

I tried the shape_transfer.py but it did not give me a good overlap ratio between the two kidneys I was trying to register.

I was wondering if there was a way to train on my own data?

Thank you and congratulations again ❤️

[Yang-L1]

Hi. For unsupervised shape registration, a good starting position is essential: maybe try to roughly align two organs using rigid registration before running DeformationPyramid. If you want to use supervised correspondence, for model training, you may need a large dataset of organs with ground-truth deformation data. Good luck.