Switching from dino to stego requires different hyper parameter

[JonasFrey96]

When switching from:
seg -> stego - random
feature -> stego - dino

Completly different hyperparameters are needed for the anomaly detection:

• confidence_std_factor (stego) 1 - and with dino minimum (5)

There maybe exist multiple reasons for it but certainly, this is not ideal.

[mmattamala]

This might be related to the stuff I was discussing here: #242 (comment)