Currently the plugin's implementation is very primitive in how it interprets the two dimensional capture data: it performs minimal corrections to translation changes by calculating delta distances from the neutral pose face, but doesn't factor in pitch, yaw, or roll of the face (moving & rotating in three dimensions).