Why does the order in the upsampling path of the U-Net developed by Monai deviate from standard u-net?

1

Hi guys,
I have a question about the U-Net developed by @MONAI . They refer to this link: https://link.springer.com/chapter/10.1007/978-3-030-12029-0_40. for explanation about the residual units they use. In every layer of their upsampling path, they first apply a transposed convolution, then a ‘normal’ convolution and then they concatanate the result with the result of the same layer of the downsampling path. However, in the link they refer to, the order is transposed convolution --> concatenation --> normal convolution. Does anyone know why they deviate from this usual order, and does the order have any influence on the result?

Thanks in advance!

2

Hi @Sabien,

I would recommend to post this question in the @MONAI GitHub repository, as more MONAI devs would see it. :wink: