Keras: Calling Keras Convolutional Layer on TensorFlow Tensor Error

[ritchieng]

Calling Keras Convolutional Layer on TensorFlow Tensor Error.

('x_train shape:', (50000, 32, 32, 3))

 # Basic info
self.batch_num = 50
self.img_row = 32
self.img_col = 32
self.img_channels = 3
self.nb_classes = 10

img = tf.placeholder(tf.float32, shape=(self.batch_num, self.img_col, self.img_row, self.img_channels))
labels = tf.placeholder(tf.float32, shape=(self.batch_num, self.nb_classes))

x = Convolution2D(16, 3, 3, border_mode='same')(img)
x = BatchNormalization(axis=3)(x)
x = Activation('relu')(x)
x = AveragePooling2D(pool_size=(8, 8), strides=None, border_mode='valid')(x)
x = Flatten()(x)

preds = Dense(self.nb_classes, activation='softmax')(x)

I tried using the following and it still gives the same error. It seems we can't call a Convolution2D layer on an input? @fchollet

If I do not use a placeholder and use inputs = Input(shape=(self.img_rows, self.img_cols, self.img_channels), it works.

I noticed a similar error persisting with https://github.com/fchollet/keras/issues/3450

img = K.placeholder(ndim=4)

I'm having the following error:

Traceback (most recent call last):
  File "cnn.py", line 176, in <module>
    a.step()
  File "cnn.py.py", line 156, in step
    preds = Dense(self.nb_classes, activation='softmax')(x)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 487, in __call__
    self.build(input_shapes[0])
  File "/usr/local/lib/python2.7/dist-packages/keras/layers/core.py", line 695, in build
    name='{}_W'.format(self.name))
  File "/usr/local/lib/python2.7/dist-packages/keras/initializations.py", line 58, in glorot_uniform
    s = np.sqrt(6. / (fan_in + fan_out))
TypeError: unsupported operand type(s) for +: 'NoneType' and 'int'

I'm using this with TensorFlow due to the flexibility I need. But I broke it down to a simple example and I can't figure out why I'm getting an error for such a simple problem.

[ritchieng]

So I managed to solve this with 2 steps:

1. K.set_learning_phase(0) before everything.
2. Instead of Flatten, change to tf.reshape(x, [-1, np.prod(x.get_shape()[1:].as_list())]).

I gathered this fix from studying a few issues:
https://github.com/fchollet/keras/pull/3253
https://github.com/fchollet/keras/issues/3450
https://github.com/fchollet/keras/pull/3253

I personally think this deserves to be updated in Keras :)

[Hippogriff]

Is it finally updated? What is the current status?

[mjdietzx]

also experiencing this issue. where i have a conv layer => flatten => dense layer.

gives error:
File "/Users/mjdietzx/Documents/GitHub/keras/keras/initializations.py", line 65, in glorot_uniform s = np.sqrt(6. / (fan_in + fan_out)) TypeError: unsupported operand type(s) for +: 'NoneType' and 'int'

using your workaround fixes

[rfarouni]

I am also encountering the same problem

[stale[bot]]

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