How about the performance of attention masks which claim to localize the character #2
Description
Dear author, have you visualized the attention masks like the pics in the original paper? Can that really find the location of each character?
{{ message }}