Question about Chars74k dataset as OOD

hendrycks / outlier-exposure

Deep Anomaly Detection with Outlier Exposure (ICLR 2019)

Apache License 2.0

541 stars 107 forks source link

Closed igormq closed 5 years ago

igormq commented 5 years ago

You said in the paper:

``Chars74K is a dataset of photographed characters in various styles; letters such as “O” and “l” were removed since they can look like numbers''

So, did you also remove the labels between 0 and 9? Because, Chars74k also contains digits (+ uppercase characters and lowercase characters).

Best, Igor.

hendrycks commented 5 years ago

Yes, those were removed too. I'll make note of that in an update to the paper. Thank you.