search

basveeling / keras-gcnn

Roto-reflection equivariant CNNs for Keras as presented in B. S. Veeling, J. Linmans, J. Winkens, T. Cohen, M. Welling. "Rotation Equivariant CNNs for Digital Pathology".
Other
85 stars 42 forks source link

Generalization to higher dimensions #8

Closed clementpoiret closed 4 years ago

clementpoiret commented 4 years ago

Hi, and thanks for your work.

I've a little question : I'm interested in using gcnn on higher dimensional volumes. Do you have any idea on how to make it 4d or at least 3d?

Thanks, Clement.

basveeling commented 4 years ago

Hi Clément,

Thanks for reaching out! There does exist follow up work on extending G-CNNs to 3D. The tricky thing is that with that extra dimension, the number of orientations under 90 degree rotations explodes (22 orientations if I recall correctly). This can get a bit memory intensive. Here is a reference:

https://arxiv.org/abs/1804.04458

Good luck,

Bas

On Sun, 19 Jan 2020 at 08:46, Clément POIRET notifications@github.com wrote:

Hi, and thanks for your work.

I've a little question : I'm interested in using gcnn on higher dimensional volumes. Do you have any idea on how to make it 4d or at least 3d?

Thanks, Clement.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/basveeling/keras-gcnn/issues/8?email_source=notifications&email_token=AAEDDDZB4NI3XXPLBBCVUVLQ6QALPA5CNFSM4KIXWAYKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IHFBOGQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAEDDD43JEPD4D42UKF6L6DQ6QALPANCNFSM4KIXWAYA .

clementpoiret commented 4 years ago

Thanks for the paper, I'll try to implement it. Wish you the best, Clément.