Implementation predication to repo HDenseUet

I try train repo that working on LiTs dataset, the repo is this: GitHub - thangylvp/HDenseUet: H-DenseUnet pytorch implementation
This is a convert code from source keras repo (GitHub - xmengli/H-DenseUNet: TMI 2018. H-DenseUNet: Hybrid Densely Connected UNet for Liver and Tumor Segmentation from CT Volumes) to pytorch.
The train working fine, i get 91%+ success, but when i try write the predict i don’t success.
when i try implementation predict the input only time return tensor with other values to what I enter.
Can someone please to:

  1. understand who the input tensor looks like?
  2. And who i can a get the summary?
  3. How i can Implementation predication for this repo?
  4. And why the source code in keras Implementation strange predication with window slice?

I would be very happy for your help,