[Seeking information] Implementation as a OOD detector

[W-XuWX]

Hi, after training the G-ODIN model on my own data, I would like to evaluate its ability to predict whether a test sample is within or out of distribution.

Q1. When preparing a test set, is the right approach to first create two classes (within_distribution and outside_distribution) ?

Q2. From resnet20_odin.py, the output of the model is (None, n_classes). How can I translate this scoring into a binary one (within vs out of distribution) ?

Thank you so much!