[ErikBjare]

You would have to do the same fine-tuning as if you had the training data.