Abnormal contact prediction

[YoloZyk]

Hi,

I tried to predict SMPL-X contacts on my own dataset using the trained model provided in the repository, but I got some strange predictions..

The pattern it shows seems to always take the front as the contact surface, but the xyz direction of the mesh shown in the figure above is consistent with the description in the document.

So this result confused me, have you encountered the same situation? Or do you have any insights?