about training loss

[zjz19960805]

hi
thanks for your code sharing
and when I try to train a model used VCOCO datasets, I was doubt why the loss is so small from the beginning, like this:

I want to know if I make some mistake to do the training step or the loss is same as yours?
thanks again~