we simplify the four-order tensor into 2 dimension by using transaltion invariance.
    Why?
    because the feature of the object we detect doesn’t depend on its position.