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
573 stars 147 forks source link

Recurrence code of SVHN original datasets #107

Open WangzekunY opened 2 years ago

WangzekunY commented 2 years ago

When I get the original SVHN dataset from here，I find train.tar.gz, test.tar.gz , extra.tar.gz. I want to reproduce the results with 95.2% accuracy obtained in the paper, I don't quite understand which dataset you use as a training set and a validation set

Bartzi commented 2 years ago

Sry, for the long time to answer. As far as I remember, We used the files train and test for training and testing. You can also split some parts of the train dataset to obtain a validation split. The extra file is not used.

WangzekunY commented 2 years ago

Thanks for your reply, could you provide training parameters with reproduction accuracy？

Bartzi commented 2 years ago

Hmm, :sweat_smile: since we published this many years ago. I don't think I can give you the exact training parameters, but I can provide you with the trained model but be aware that this model does not work with the code from this repo. It only works with the code from our earlier repo but this code is very very old. I'm not sure whether it is still easily possible to make it work...

Anyway, to train such a model on SVHN, you'll need to follow the steps for the training of text recognition models.