These tags are not known by the fine tuning of the model, but they might be known anyway due to training with normal images.