Test time Augmentation Method?

In test time augmentation, is it simply to convert a tensor to a PIL image, then convert it (Flip, Rotation) and then convert it to a tensor and put it into the model?

You shouldn’t apply any transformation but normalization at the time of testing

1 Like

Thank you for answering my question.
What is mean ‘shouldn’t apply any transformation’?
Could you tell me a little more specifically?