AttributeError: 'Tensor' object has no attribute 'assign'

[SergeKrier]

Hi Felix, thanks for providing the resnet152 model in keras + weights.
However I have the following error when trying to run it :

File "resnet_152.py", line 252, in
model = resnet152_model(img_rows, img_cols, channel, num_class)
File "resnet_152.py", line 222, in resnet152_model
model.load_weights(weights_path)
File "/usr/local/lib/python3.4/dist-packages/keras/engine/topology.py", line 2500, in load_weights
load_weights_from_hdf5_group(f, self.layers)
File "/usr/local/lib/python3.4/dist-packages/keras/engine/topology.py", line 2913, in load_weights_from_hdf5_group
K.batch_set_value(weight_value_tuples)
File "/usr/local/lib/python3.4/dist-packages/keras/backend/tensorflow_backend.py", line 2022, in batch_set_value
assign_op = x.assign(assign_placeholder)
AttributeError: 'Tensor' object has no attribute 'assign'

tensorflow: 1.0.1
keras: 2.0.3

thanks

[flyyufelix]

Right now ResNet-152 supports Keras 1.2.2 and Tensorflow 0.12.0. The error you are seeing should be related to Tensorflow versioning. I will try to get the models work for the latest versions as soon as possible.

[SergeKrier]

Hi Felix, I see it closed don't see any recent commit for this.thanks

[lef-fan]

Hello, same issue here for Keras 2.0.3 and Tensorflow 1.1.0
Couldn't wait for it so I gave it a try.

At scale_layer.py change:

self.gamma = self.gamma_init(shape, name='{}_gamma'.format(self.name))
self.beta = self.beta_init(shape, name='{}_beta'.format(self.name))

to:

self.gamma = K.variable(self.gamma_init(shape), name='{}_gamma'.format(self.name))
self.beta = K.variable(self.beta_init(shape), name='{}_beta'.format(self.name))

"edited"
Check my repo here for an update to Keras 2.0.3 and Tensorflow 1.1.0, without the Cifar10 example.

[flyyufelix]

Thanks @lef-fan for the suggested patch, and apologize for the delay @SergeKrier . I've already made the necessary changes to get the code to work with the latest versions of Keras (2.0.5) and TensorFlow (1.2.1)