When we use view frustum, do we need to first convert back the coordinate in normalized space to world coordinate before we multiply the point position by camera projection matrix?