How to do inverse transformation in the test image data

I do some augmentation techniques in my image dataset ```
RandomRotation,RandomVerticalFlip, and others at the same time. after training I want to recover the original image with no augmentation in the test part. It there a flag or something that identifies which image was transformed and with which transformation do to the inverse transformation ?

No, the transformations do not track their randomly samples arguments, but you could track them manually and by sampling the random arguments first and then by applying them with the functional transformation API.
This post gives you an example how this API can be used.