'TensorVariable' object has no attribute '_keras_shape'

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

'TensorVariable' object has no attribute '_keras_shape' #12

Closed Perseus14 closed 7 years ago

Perseus14 commented 7 years ago

While running the code _make_keras_charnetmodel.py, I encountered the error "'TensorVariable' object has no attribute '_keras_shape'".

I had followed the instructions given in the README doc on overwriting convolution.py and _theanobackend.py

Apparently, the error in encountered in line 43

model.add(Convolution2D(nb_filters[layer1], nb_conv[layer1], nb_conv[layer1],weights=weights,input_shape=(1, img_rows,img_cols),activation='relu',border_mode='same'))

Image is attached below

keras

Perseus14 commented 7 years ago

My keras and theano versions were different. Once I installed the required version. It worked

mathDR commented 7 years ago

Thanks for closing. This is what I was going to suggest.