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.
215 stars 67 forks source link

How is θ given #19

Closed renqiqi closed 3 years ago

renqiqi commented 3 years ago

Hi， How is θ given? Is it artificially set or calculated? If it is calculated, how is it calculated? Looking forward to your reply.

marco-rudolph commented 3 years ago

There is no fix theta. Firstly, it depends on the dataset. Secondly it depends on the desired false positive rate. You could use a validation set (by splitting the test set) to get a feeling for the distribution of anomaly scores of normal data.

renqiqi commented 3 years ago

There is no fix theta. Firstly, it depends on the dataset. Secondly it depends on the desired false positive rate. You could use a validation set (by splitting the test set) to get a feeling for the distribution of anomaly scores of normal data.

I got it, thanks for your reply.