Download the Omniglot dataset and unzip python/images_background.zip here. This folder should have the following structure:
omniglot
|-- Alphabet_of_the_Magi
|-- character01
|-- 0709_01.png
|-- 0709_02.png
|-- ...
|-- 0709_20.png
|-- character02
|-- ...
|-- character20
|-- Anglo-Saxon_Futhorc
|-- character01
|-- ...
|-- Arcadian
|-- character01
|-- ...
|-- ...