search

Bartzi / see

Code for the AAAI 2018 publication "SEE: Towards Semi-Supervised End-to-End Scene Text Recognition"
GNU General Public License v3.0
575 stars 147 forks source link

Text Recognition on new dataset #72

Open SiDinhThoi opened 5 years ago

SiDinhThoi commented 5 years ago

Hi @Bartzi,

I am wondering about do your algorithm can work on following images?

0000000_orig 0000001_orig 0000002_orig

Thank you and looking forward to hearing back from you!

Bartzi commented 5 years ago

Hmm, I don't think it would work well... you see one of the problems with the approach is that it is currently not really possible to work with images that contain lots of text and also text that is in very different positions... But actually you should not need to have such a complicated deep learning system for localizing and detecting such text...

SiDinhThoi commented 5 years ago

Hi @Bartzi,

Thank you for your response

I am curious about if I train the model from scratch with this new dataset, will it work? or do you have any idea to make the algorithm can work with this new dataset?

Sincerely,

Si

Bartzi commented 5 years ago

As I said, I don't think it will work out of the box, as the proposed system has problems with many textlines in large position variations. The guidance obtained by the localization network from the recognition network is not enough for such problems, but all in all I think you are still better off with a standard print OCR engine, for such data...

SiDinhThoi commented 5 years ago

Thank you for your advice @Bartzi