Question: No New Net (2019)

stalhabukhari commented 4 years ago

Hi! My question is about your paper No New Net which cites your previous paper, corresponding to this repository. Are you using Concatenation to supply feature maps from encoder pathway to decoder pathway of your UNet (as in this repository), or simple Addition which is maybe more memory efficient?

FabianIsensee commented 4 years ago

Hi, I use concatenation. Addition could work just as well, though. I have not compared the two. Best, Fabian

stalhabukhari commented 4 years ago

Thank you so much for the prompt response. Just have a few more to clear out:

It seems that there are no skips connections/residual blocks or pre-activation blocks in No New Net. Also there is use of Max-Pooling in contrast to Strided Convolution in this repository. Am I on track?
In No New Net, the term Trilinear Up-Sampling is used. Does this refer to Up-Sampling by repetition (as in this repository) or Up-Sampling by linear interpolation?
If it was recorded during implementation, in both cases, what was the memory footprint of your architectures on GPU?

Best Regards, Talha

MIC-DKFZ / BraTS2017

Question: No New Net (2019) #33