Unused edge info in the E2HTransform

[borkowski1110]

Hi!
I am currently researching the Facial Landmark Detection problem and have stumbled upon your architecture which is very compelling!
It would be of great help if you could elighten me in one matter. In the E2HTransform we strive for utilizing the information whether a certain keypoint is covered by an edge.

It all makes sense, although the indicator whether the said holds or not seems to remain unused.

Could you explain the underlying idea to me?

Thank you,
Jakub

Permalink: https://github.com/ZhenglinZhou/STAR/blob/9b125749b0d35766ed83d047036d1aa5e384984c/lib/backbone/stackedHGNetV1.py#L160C13-L160C22