Training linemod or custom dataset

[Yipingggg]

In the code, seems all the input data are normalized, but they are simulated data.
When training linemod or custom dataset, how to deal with the input correspondence? What normalization should I do, or I don't need to normalize? Thanks!