Upsample Convolutions in UNet

[dyollb]

I am working MR to CT synthesis and was reading about checkerboard artifacts and transposed convolutions vs upsample convolutions.

• I see monai has an Upsample block
• I also found some discussion about refactoring the generative module to use upsample convolutions

Does monai currently provide any UNet (or UNet-like) model that can use upsample convolutions in the decoder layers?

[dyollb]

Hmm. I see for many networks you can set upsample="nontrainable"

• BasicUnet
• BasicUnetPluPlus
• FlexibleUnet
• SegResNet
• and probably others

But not in UNet (which is what I was using) or DynUNet (I think).

Btw. I know this is unrelated and doesn't belong in this discussion, but is there any overview or guidance to help choose one the many versions of U-Net in MONAI?