This looks awesome and somehow familiar to the poor man’s mocap video on youtube.
Strangely they don’t show how they get the data for the lower lids when adressing it as an emotional intensifier. You can clearly see that the actor is given wide extreme poses here.
It won’t work only being connected to the cheek markers. Either they will track the eyes volume in the 2D video plane or animate it later on.
Also lip rolls are not properly provided from the mocap data alone.