Training only tan model

Hi,

I would like to know if it is enough to train only the tan model (without the end part) to generate de augmented samples. Then I want to apply and indpendente classifier (resnet) to the transformed dataset. Is it necessary the end model for this case of use?