I understand that based on your paper, it is mentioned that the CIFAR-10 is evaluated in one-class setting, where one class is normal and the others are considered as anomaly.
Others: one class as normal and others as anomaly, at testing: the whole test set is used
I wonder for the final reported performance, did you conduct experiments for each of the classes considered as an anomaly, i.e., did 10 experiments for 10 classes, and averaged the results from the 10 experiments? Or did you specifically choose only one class? If so, which class did you choose for the experiment? Was it '3' as used in the config file here?
I understand that based on your paper, it is mentioned that the CIFAR-10 is evaluated in one-class setting, where one class is normal and the others are considered as anomaly.
I wonder for the final reported performance, did you conduct experiments for each of the classes considered as an anomaly, i.e., did 10 experiments for 10 classes, and averaged the results from the 10 experiments? Or did you specifically choose only one class? If so, which class did you choose for the experiment? Was it '3' as used in the config file here?
https://github.com/rohban-lab/Knowledge_Distillation_AD/blob/main/configs/config.yaml#L13