The number of downloaded images can not match with the number pointed out in the paper #4
Description
Thank you for your excellent work providing a benchmark for domain adaptation methods for text recognition. We downloaded the datasets via Baidu Netdisk, however the number of downloaded images for the three datasets （hand, street, document） could not match the number presented in the paper.
The differences are as follows
- hand:1897021 (paper),1198876 (download)
- street:199346(paper),199343 (download)
- document:1710885(paper),1794300 (download)
The toal number of images download is 4,511,067 while the reported number from the paper is 5,209,215.