Keras: Exception: The shape of the input to "Flatten" is not fully defined

[hadikazemi]

The following code generates no error on a virtualbox Ubuntu 14.04 (CPU only). The code is an tutorial from www.pyimagesearch.com

    from sklearn.cross_validation import train_test_split
    from sklearn import datasets
    from keras.optimizers import SGD
    from keras.utils import np_utils
    import numpy as np
    import argparse
    import cv2
    from keras.models import Sequential
    from keras.layers.convolutional import Convolution2D
    from keras.layers.convolutional import MaxPooling2D
    from keras.layers.core import Activation
    from keras.layers.core import Flatten
    from keras.layers.core import Dense

    def LeNet_build(width, height, depth, classes, weightsPath=None):
        print width, height, depth
        model = Sequential()

        # first set of CONV => RELU => POOL
        model.add(Convolution2D(20, 5, 5, border_mode="same",
            input_shape=(depth, height, width)))
        model.add(Activation("relu"))
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))

        # second set of CONV => RELU => POOL
        model.add(Convolution2D(50, 5, 5, border_mode="same"))
        model.add(Activation("relu"))
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))

        # set of FC => RELU layers
        model.add(Flatten())
        model.add(Dense(500))
        model.add(Activation("relu"))

        # softmax classifier
        model.add(Dense(classes))
        model.add(Activation("softmax"))

        # if a weights path is supplied (inicating that the model was
        # pre-trained), then load the weights
        if weightsPath is not None:
            model.load_weights(weightsPath)

        # return the constructed network architecture
        return model

    dataset = datasets.fetch_mldata("MNIST Original")

    # reshape the MNIST dataset from a flat list of 784-dim vectors, to
    #28 x 28 pixel images, then scale the data to the range [0, 1.0]
    # and construct the training and testing splits
    data = dataset.data.reshape((dataset.data.shape[0], 28, 28))
    data = data[:, np.newaxis, :, :]
    (trainData, testData, trainLabels, testLabels) = train_test_split(
        data / 255.0, dataset.target.astype("int"), test_size=0.33)

    # transform the training and testing labels into vectors in the
    # range [0, classes] -- this generates a vector for each label,
    # where the index of the label is set to `1` and all other entries
    # to `0`; in the case of MNIST, there are 10 class labels
    trainLabels = np_utils.to_categorical(trainLabels, 10)
    testLabels = np_utils.to_categorical(testLabels, 10)

    # initialize the optimizer and model
    print("[INFO] compiling model...")
    opt = SGD(lr=0.01)
    model = LeNet_build(width=28, height=28, depth=1, classes=10, weightsPath=None)

However, it generates the following error inside a fresh installed Ubuntu 14.04 with Titan X Pascal and Cuda 8.0 (tested both on CPU and GPU - Theano backend) - the same error is reported before and remained unsolved (https://github.com/fchollet/keras/issues/1592):

 Exception                                 Traceback (most recent call last)
<ipython-input-3-c61646d5fa78> in <module>()
     23 print("[INFO] compiling model...")
     24 opt = SGD(lr=0.01)
---> 25 model = LeNet_build(width=28, height=28, depth=1, classes=10, weightsPath=None)

<ipython-input-2-a36c66483235> in LeNet_build(width, height, depth, classes, weightsPath)
     16 
     17         # set of FC => RELU layers
---> 18         model.add(Flatten())
     19         model.add(Dense(500))
     20         model.add(Activation("relu"))

/home/hadi/.virtualenvs/keras/local/lib/python2.7/site-packages/keras/models.pyc in add(self, layer)
    306                  output_shapes=[self.outputs[0]._keras_shape])
    307         else:
--> 308             output_tensor = layer(self.outputs[0])
    309             if type(output_tensor) is list:
    310                 raise Exception('All layers in a Sequential model '

/home/hadi/.virtualenvs/keras/local/lib/python2.7/site-packages/keras/engine/topology.pyc in __call__(self, x, mask)
    512         if inbound_layers:
    513             # this will call layer.build() if necessary
--> 514             self.add_inbound_node(inbound_layers, node_indices, tensor_indices)
    515             input_added = True
    516 

/home/hadi/.virtualenvs/keras/local/lib/python2.7/site-packages/keras/engine/topology.pyc in add_inbound_node(self, inbound_layers, node_indices, tensor_indices)
    570         # creating the node automatically updates self.inbound_nodes
    571         # as well as outbound_nodes on inbound layers.
--> 572         Node.create_node(self, inbound_layers, node_indices, tensor_indices)
    573 
    574     def get_output_shape_for(self, input_shape):

/home/hadi/.virtualenvs/keras/local/lib/python2.7/site-packages/keras/engine/topology.pyc in create_node(cls, outbound_layer, inbound_layers, node_indices, tensor_indices)
    150             output_masks = to_list(outbound_layer.compute_mask(input_tensors[0], input_masks[0]))
    151             # TODO: try to auto-infer shape if exception is raised by get_output_shape_for
--> 152             output_shapes = to_list(outbound_layer.get_output_shape_for(input_shapes[0]))
    153         else:
    154             output_tensors = to_list(outbound_layer.call(input_tensors, mask=input_masks))

/home/hadi/.virtualenvs/keras/local/lib/python2.7/site-packages/keras/layers/core.pyc in get_output_shape_for(self, input_shape)
    400             raise Exception('The shape of the input to "Flatten" '
    401                             'is not fully defined '
--> 402                             '(got ' + str(input_shape[1:]) + '. '
    403                             'Make sure to pass a complete "input_shape" '
    404                             'or "batch_input_shape" argument to the first '

Exception: The shape of the input to "Flatten" is not fully defined (got (0, 7, 50). Make sure to pass a complete "input_shape" or "batch_input_shape" argument to the first layer in your model.

Update: I have checked all of my previous Keras codes which were working in the virtualbox and all of them give the same error:

Exception: The shape of the input to "Flatten" is not fully defined (got (0, 7, 512). Make sure to pass a complete "input_shape" or "batch_input_shape" argument to the first layer in your model.

[ExpectationMax]

Had a similar issue. In my case, the one machine was set up to use tensorflow image_dim_ordering and the other using the theano ordering. Somehow the install routines selected different defaults.

Compare the file ~/.keras/keras.json between your workstations. If they differ set image_dim_ordering to "th" (your tutorial seems to be using the theano backend) and try again.

[hadikazemi]

You saved my day... image_dim_ordering was 'tf'

[MaryamAkhavan]

I had the same problem and I checked all of comments, but my problem did not solved!!

[hadikazemi]

@MaryamAkhavan have you checked your ~/.keras/keras.json file? Are you sure if both backend and image_dim_ordering are set to the same library? they both should be Tensorflow , TF or Theano, TH

[MaryamAkhavan]

Yes, I checked. Backend is theano and image_dim_ordering is (chanels, dim1, dim2)

[ExpectationMax]

Are you running the same example? If not, please post the model construction and the error message.

[MaryamAkhavan]

I ran both the example of Simple Convolutional Neural Network for MNIST located at this link : http://machinelearningmastery.com/handwritten-digit-recognition-using-convolutional-neural-networks-python-keras/
and my code. Both examples have the same error.

[hadikazemi]

Please post the error here...

[0bserver07]

I think this issue can closed, it's due to not having either:

• Theano as backend
  or;
• including this option image_dim_ordering in the ~/.keras/keras.json file.

'{
    "image_dim_ordering": "th",
    "epsilon": 1e-07,
    "floatx": "float32",
    "backend": "theano"
}' 

relevant issue #1592

[anujshah1003]

This two lines solved my issue
from keras import backend as K
K.set_image_dim_ordering('th')

[Moondra]

I'm having a similar problem, but I don't see K.set_image_dim_ordering. I see a image_data_format": "channels_last" however.

[anujshah1003]

yes thats because in keras 2 it is image_data_format instead of
image_dim_ordering

On Tue, Apr 25, 2017 at 8:19 PM, Moondra [e-mail] wrote:

I'm having a similar problem, but I don't see K.set_image_dim_ordering. I
see a image_data_format": "channels_last" however.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/fchollet/keras/issues/3850#issuecomment-297054848,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AT2bYa6YHA1rrfvLiC-yM7FP1IfeyEeeks5rzggIgaJpZM4KEQM-
.

--
Best Regards,

Anuj Shah
([email protected])

[Moondra]

Thanks. Yeah, I had to manually delete the json file before reinstalling Keras 1.2.

[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.

[DSLituiev]

I have tensorflow backend both in my json configuration file and it is what keras outputs on import. My model is super simple, and i still get this error:

    model = Sequential()
    model.add(InputLayer(input_shape=[512,512,1]))
    model.add(Flatten())
    model.add(Dense(n_classes, activation='relu', ))

What is the problem? What to do if I need to use TF, not Theano?

[0bserver07]

@DSLituiev Maybe it's the channel related issue, did you try including this on top of the model:

from keras import backend as K
K.set_image_dim_ordering('tf')

check here if that's the issue (https://keras.io/backend/)

[DSLituiev]

@0bserver07: thank you! That was not the issue.
My bad! I fed tensors from TFRecords pipeline with undefined shapes.