search

marco-rudolph / differnet

This is the official repository to the WACV 2021 paper "Same Same But DifferNet: Semi-Supervised Defect Detection with Normalizing Flows" by Marco Rudolph, Bastian Wandt and Bodo Rosenhahn.
217 stars 68 forks source link

How to train anomaly data set to improve accuracy ? #28

Closed rfuk0503 closed 3 years ago

rfuk0503 commented 3 years ago

Is there any way to train anomaly images such as to modify Loss Function get_loss() ? Assuming that I got anomaly images after DifferNet trainning done.

marco-rudolph commented 3 years ago

This is not foreseen in this method. Theoretically, one could use the negative loss in this case to minimize the likelihood, but I guess in this case the model would become unstable and would need regularization for these cases.

rfuk0503 commented 3 years ago

Thank you for your quick reply.