question about evaluation

[vanche]

Hi, thank you for sharing your nice work!
I'd like to ask the processing characterset when evaluating your code.
Did you use 'label = re.sub('[^0-9a-zA-Z]+', '', label)' at evaluation mode ?
I wonder how did you process ground truth in the results reported on your paper.
If you use 32-charset, did you ignore out of vocab character(^0-9a-z) for gt?
for example, if gt is 'hello?' and pred is 'hello', do you check the result is correct?