Question about InpFormer ++ Segmentation head

[unilin-pietergobin]

Hi,

First off, thank you for open sourcing such a nice universal anomaly detection model.
I am currently trying to work out how you made the segmentation head for the inpformer ++ paper.

I currently have two questions:

1. Could you share the outline of this segmentation head: kernel sizes, strides, where to put the bilinear upscaling?
2. Did you ever find a use for the distance map to better guide segmentation during training of the segmentation head?

Thanks in advance!