Different Loss Functions

[TobiArndt]

Hi, I really like your work! I have got a question with respect to the loss functions: Is there a chance to use loss functions from the unsupervised regime? Like Student-Teacher Losses, where the final loss is the difference between the intermediate layer between a frozen teacher and a trainanble student network? This could be important for filtering anomalies from a dataset, so that only the non-anomaly images can be used for training the anomaly detector. Thank in advanced!

[Dref360]

Hello, sorry for the delay.

Do you mean that you take the loss as a measure of uncertainty?