preprocessed ImageNet-C-bar dataset

[stsavian]

Greetings,

thanks for this great work! I have a question regarding the Berkeley prepocessed dataset. In the Berkeley preprocessed, the corruption severity spans between the integer 1-5. Hoever the corruption levels of the functions of augmentation-corruptions go up to 10.

In the preprocessed dataset, what are the corruption levels? Does the severity 5 reads as 5 according to your functions? Or do I have to double that value?

For example, if i have a gaussian distortion, in order to replicate the Berkeley preprocessed dataset, to I have to specify the severities:1,2,3,4,5 or do I have to specify 2,4,6,8,10?

EDIT: it seems that the file imagenet_c_bar.csv contains the corruption severities.

Thanks, Stefano