search

mathDR / reading-text-in-the-wild

A Keras/Theano implementation of "Reading Text in the Wild with Convolutional Neural Networks" by M Jaderberg et.al.
GNU General Public License v3.0
116 stars 30 forks source link

Files compatible with the recent versions of Keras and Theano/Tensorflow #15

Open The-Gupta opened 7 years ago

The-Gupta commented 7 years ago

Has anyone modified make_keras_dictnet_model.py and make_keras_charnet_model.py to make it compatible with the recent versions of Keras and Theano/Tensorflow?

mathDR commented 7 years ago

Not to my knowledge. This would be a great PR if you want to tackle this?

The-Gupta commented 7 years ago

Hello @mathDR, I need your help. Could you write to me at vishal.gupta@nitdelhi.ac.in?

mathDR commented 7 years ago

Hey @The-Gupta what is the issue? Can you elaborate?

The-Gupta commented 7 years ago

@mathDR , I see you've posted only the text recognition code but not region proposal implementaion - how the bounding boxes are found and their numbers are reduced. If you've not, do you know who has implemented these parts as well of Section 4 of Max's paper - https://arxiv.org/pdf/1412.1842.pdf, preferably in Python?

mathDR commented 7 years ago

@The-Gupta That was always the plan, but I never got to it. Furthermore, I don't know of the existence of python for doing this pipeline. I do recall, however, that most, if not all, of the functionality exists in python (numpy, scipy, scikit-learn, etc.) and you would just have to link it all together.

If you take an initial attempt at building it, then I could help out (some) in adding functionality.

archenroot commented 5 years ago

@The-Gupta - have you found out solution?