Network diagram interpretation

1

Hi guys,

I am trying to implement the segmentation network proposed in [1].

image
image758×335 49.2 KB

I have two questions regarding the interpretation of that diagram:

  1. Following the down convolution, there is always a new cube added in the back. How would you interpret this? I am guessing that it’s the
nfilters * 1                                   nfilters * 1   
   |_ nfilters * 2                       nfilters * 2_|
      |_ nfilters * 3              nfilters * 3_|
          |_ nfilters * 4 | nfilters * 4 _|
  1. How would I execute the concatenation step? I was thinking about the following:
 |_ nfilters * 3 (left side) ---> compress(squeze to one filter) and concat  --> nfilters * 2 (right side) --> resulting in nfilters * 3

What do you guys think?

Thanks a lot,

Best
Christian

[1] Anatomically Constrained Neural Networks (ACNNs): Application to Cardiac Image Enhancement and Segmentation | IEEE Journals & Magazine | IEEE Xplore