search

DeepRegNet / DeepReg

Medical image registration using deep learning
Apache License 2.0
569 stars 79 forks source link

Adding Point-Set registration method(s) #475

Closed zacbaum closed 3 years ago

zacbaum commented 4 years ago

Subject of the feature

It could be beneficial to DeepReg packages to include some form of point-set registration (either classical [ICP, CPD] or new learning-based methods).

This will also likely require useful methods for quick extraction of point-sets from image labels/masks.

NMontanaBrown commented 4 years ago

Do we have to re-implem this or just point to existing packages/integrate with packages?

I'm aware of CPD at:

And I think the sksurgery libraries have ICP? @MattClarkson

I opened a separate issue for FPT #511.

YipengHu commented 4 years ago

Do we have to re-implem this or just point to existing packages/integrate with packages?

I'm aware of CPD at:

And I think the sksurgery libraries have ICP? @MattClarkson

I opened a separate issue for FPT #511.

the said implementation is not "buggy" but having only partial implementation... i'm more interested in adding Zac's learning-based stuff.

NMontanaBrown commented 4 years ago

The ticket does say "either classical [ICP, CPD] or new learning-based methods" so I was trying to be thorough 😄 @YipengHu

mathpluscode commented 4 years ago

@NMontanaBrown is this ticket duplicated with #511 or #511 is a small part of this?

NMontanaBrown commented 4 years ago

@NMontanaBrown is this ticket duplicated with #511 or #511 is a small part of this?

Hmm, there is some overlap, but I had thought it might be worth separating into two, since #511 is specifically a new architecture and this refers to classical point set registration in addition. Let's modularise our tickets as it will be easier to review, if FPT is not dependent on ICP or CPD implems? @zacbaum

zacbaum commented 4 years ago

Probably no use in reimplementing CPD or ICP as there are existing versions out there (was just a thought when I typed this out during the meeting a few weeks ago). I'd say we can close one of them, just a matter of which...

NMontanaBrown commented 4 years ago

Probably no use in reimplementing CPD or ICP as there are existing versions out there (was just a thought when I typed this out during the meeting a few weeks ago). I'd say we can close one of them, just a matter of which...

I can close #511, but I'm wondering if there is any use integrating the classical point set packages via a demo for ease of use with DeepReg? This is a completely different ticket, but thoughts?

zacbaum commented 4 years ago

Probably no use in reimplementing CPD or ICP as there are existing versions out there (was just a thought when I typed this out during the meeting a few weeks ago). I'd say we can close one of them, just a matter of which...

I can close #511, but I'm wondering if there is any use integrating the classical point set packages via a demo for ease of use with DeepReg? This is a completely different ticket, but thoughts?

Could be of use in the future, but let's get Point Set registration in first, then worry about how to showcase it in a demo 😄

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.