search

DeepRegNet / DeepReg

Medical image registration using deep learning
Apache License 2.0
564 stars 76 forks source link

Use separable kernels for ResizeCPTransform #722

Closed mathpluscode closed 3 years ago

mathpluscode commented 3 years ago

Subject of the feature

Currently, in ResizeCPTransform we are using 3D Gaussain kernel https://github.com/DeepRegNet/DeepReg/blob/main/deepreg/model/layer.py#L435

I think the 1D separable kernel is equivalent to this and it's more efficient? As in LNCC https://github.com/DeepRegNet/DeepReg/blob/main/deepreg/loss/image.py#L182

So I suggest to uniform this bit, what do you think @acasamitjana

acasamitjana commented 3 years ago

Hey, I'm looking at that. Just one comment: I'm using functions from loss.kernel.py file. Should we move this outside of the loss directory?

mathpluscode commented 3 years ago

Hey, I'm looking at that. Just one comment: I'm using functions from loss.kernel.py file. Should we move this outside of the loss directory?

@acasamitjana entirely up to you! 😃I'm fine with both.

acasamitjana commented 3 years ago

Ok, let's leave as is, then.

stale[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.