CODEWITHSUNDEEP
pythontensorflowkeras
Julien
Julien

Reputation: 781

How define an input layer for a transpose convolution

I'm using tensorflow and keras to build a neural network.

I want to use a transpose convolution with keras.layers.Conv2DTranspose (Definition in Keras documentation)

I used this tutorial and I defined my network as follow:

import tensorflow as tf

from keras.models import Sequential

from keras.layers import Conv2DTranspose

sess = tf.Session()

batch_size = 20

model = Sequential()

model.add(Conv2DTranspose(filters = (batch_size,1,2700, 1),kernel_size = (2700,1), activation = 'relu', input_shape = (1,1,1,1)))

There is the following error:

ValueError                                Traceback (most recent call last)
<ipython-input-3-01a3b17fa36f> in <module>()
     11 model = Sequential()
     12 
---> 13 model.add(Conv2DTranspose(filters = (batch_size,1,2700, 1),kernel_size = (2700,1), activation = 'relu', input_shape = (1,1,1,1)))

/usr/local/lib/python3.5/dist-packages/keras/models.py in add(self, layer)
    465                 # and create the node connecting the current layer
    466                 # to the input layer we just created.
--> 467                 layer(x)
    468 
    469             if len(layer._inbound_nodes[-1].output_tensors) != 1:

/usr/local/lib/python3.5/dist-packages/keras/engine/topology.py in __call__(self, inputs, **kwargs)
    573                 # Raise exceptions in case the input is not compatible
    574                 # with the input_spec specified in the layer constructor.
--> 575                 self.assert_input_compatibility(inputs)
    576 
    577                 # Collect input shapes to build layer.

/usr/local/lib/python3.5/dist-packages/keras/engine/topology.py in assert_input_compatibility(self, inputs)
    472                                      self.name + ': expected ndim=' +
    473                                      str(spec.ndim) + ', found ndim=' +
--> 474                                      str(K.ndim(x)))
    475             if spec.max_ndim is not None:
    476                 ndim = K.ndim(x)

ValueError: Input 0 is incompatible with layer conv2d_transpose_1: expected ndim=4, found ndim=5

Nevertheless, my input has a dimension 4 (input_shape = (1,1,1,1))

How can I define the input correctly then add some layers ?

Upvotes: 0

Views: 458

Answers (1)

mpariente
mpariente

Reputation: 660

From the documentation of Conv2DTranspose you can see that filters is a positional argument and is supposed to be an integer. This integer specifies the number of filters you want in your layer.
The next parameter is kernel_size (also positional) which specifies the shape of your filter. I think what you were looking for is :

model.add(Conv2DTranspose(n_filt, (2700, 1), activation='relu' input_shape = (1, 1, 1,)))

where n_filt is the number of transposed convolution filters in your layer.

Notes :

  • Don't provide the batch dimension in the input_shape argument, or use batch_input_shape instead.
    Edit to clarify : If your data batch has dimensions (batch_size, dim1, dim2, dim3), you should pass either input_shape = (dim1, dim2, dim3, ) (Notice the comma before the last whitespace) or batch_input_shape = (batch_size, dim1, dim2, dim3). I recommend using input_shape as you won't have any constraints on the batchsize you want to use.

  • Positional arguments cannot be specified with keywords i.e don't use filters = ... in the function call if filters is a positional argument.

Upvotes: 1

Related Questions