Why are the preprocessing layers deepcopied?

[fepegar]

I'd like to keep track of the random augmentation parameters applied to each single sample. Every time the reader is called, the preprocessors are deepcopied, therefore the information I want is not stored.

Why are these layers deepcopied before being randomised?

[fepegar]

It seems to work fine without the deepcopy operation.

[wyli]

Hi @fepegar because that function could be called from multiple threads simultaneously, deepcopy ensures the 'randomness'.

[fepegar]

Thanks @wyli