Occluded landmark about centerface

[bhack]

How are you handling the loss for occluded landmarks?
I was looking for this detail in the paper but I didn't find anything.

According to the face annotation of widerface, we filter the occluded face in the landmark task.

I meant more the case when you have one or two occluded landmarks so a very partial occlusion of the face.

[bhack]

So here:

Different from Box regression, the regression of the five facial landmarks adopts the target normalization method
based on the center position.

So if you have on a face just 4 visibile landmarks are you simply going to exclude that one from the regression?

[bhack]

This Is valid only if you have occluded points in GT.