warmspringwinds
Cluster classification code
Can you please share the code to classify clusters (same issue as #7 )? Also, do you have any intuitive justification for why the technique works despite removing the downsampling layers?
Thank you
Hi, thank you for reaching out!
We will release an updated version of the code and paper soon!
It is a common technique in computer vision/image segmentation domain where you set stride of a layer to 1 (basically removing downsampling) and adjusting the dilation factor of convolution -- it allows to avoid downsampling feature maps while original convolutions still work at the same resolution (because of adjusted dilation factor).
For intuitive understanding please look at figure number 3 from this paper: https://arxiv.org/pdf/1606.00915.pdf
You can also read dilated residual networks paper for more information
