Why is the data volume of 33k downloaded from the four scenarios corresponding to vkitti, and how is the data volume of 20k in the paper obtained?