ok, you have an object, this defines two points, two corners of a cubeoid, one intersecting all the minimum dimentions, another intersecting all the maximum dimentions.
it's quite posable that the parseing code simply does not work, (it might be easier for you to make a box in your modeling program and apply the invisable texture to it, and use the no colide invisable flag), just make sure there in the same subobject (or less effecently but functional, the visable geometry is a child of the box object)
what it does is it takes the position of the eye relitive to the object beiing evaluated, and on an element by element, basis it makes sure none of the values are bigger than the max, or less than the min, if any are it won't render