Keras: Problem using tensor for model output

[kovacspeter]

Hey guys I am having problem using keras interchangeably with tensorflow API can you please point me to where I am not understanding correctly model construction.

        # IMAGENET PRETRAINED VGG16
        base_model = VGG16(weights='imagenet', include_top=True)
        # KERAS LAYER DEPENDING ON PARAMS
        vgg_out = base_model.get_layer(name='fc2') if include_fc else base_model.get_layer(name='block5_pool')

        # GET FLATTENED TENSORFLOW TENSOR
        flatten = Flatten()(vgg_out.output)

        # DO WE WANT TO BACKPROP GRADS TO PRETRAINED LAYERS?
        out = flatten if finetune_extractor else tf.stop_gradient(flatten, 'stop_grad')

        # HASHING LAYERS
        hash_layer = Dense(hash_size, activation='relu')(out)

        # CREATE KERAS MODEL
        return Model(inputs=base_model.input, outputs=hash_layer, name="hash_output")

The error I am having is

Traceback (most recent call last): File "/home/*/*/*/main.py", line 24, in <module> model = UTH(64, finetune_extractor=False, include_fc=False, input_shape=(None, 32, 32, 3), margin=1) File "/home/*/*/*/UnsupervisedTripleHashing/model.py", line 11, in __init__ hash_net = self.inference(hash_size, finetune_extractor, include_fc) File "/home/*/*/*/UnsupervisedTripleHashing/model.py", line 64, in inference return Model(inputs=base_model.input, outputs=hash_layer, name="hash_output") File "/home/kovacs/tf-venv3/lib/python3.5/site-packages/tensorflow/contrib/keras/python/keras/engine/topology.py", line 1643, in __init__ build_map_of_graph(x, seen_nodes, depth=0) File "/home/kovacs/tf-venv3/lib/python3.5/site-packages/tensorflow/contrib/keras/python/keras/engine/topology.py", line 1634, in build_map_of_graph next_node = layer.inbound_nodes[node_index] AttributeError: 'NoneType' object has no attribute 'inbound_nodes' Exception ignored in: <bound method BaseSession.__del__ of <tensorflow.python.client.session.Session object at 0x7f49d8526518>> Traceback (most recent call last): File "/home/kovacs/tf-venv3/lib/python3.5/site-packages/tensorflow/python/client/session.py", line 587, in __del__ AttributeError: 'NoneType' object has no attribute 'TF_NewStatus'

[Dref360]

wrap tf.stop_gradient in a lambda layer

[akshaychawla]

I fixed the Exception ignored in: > , I do this by adding :

import keras.backend as K
# get data, train model
model.fit_generator(...)
# clear the session manually
K.clear_session()

This fixed the error in my case. Apparently model.fit did not end the session cleanly.
Source for fix:
http://qiita.com/TomokIshii/items/178938b6db1edc16b94e

setup:
tensorflow - 1.1.0
keras - 2.0.4

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.