Inputs should have rank 3. Received input shape: input_shape

Package:

keras

52268

Exception Class:

ValueError

Raise code

    raise ValueError('Stride ' + str(self.strides) + ' must be '
                           'greater than output padding ' +
                           str(self.output_padding))

  def build(self, input_shape):
    input_shape = tf.TensorShape(input_shape)
    if len(input_shape) != 3:
      raise ValueError('Inputs should have rank 3. Received input shape: ' +
                       str(input_shape))
    channel_axis = self._get_channel_axis()
    if input_shape.dims[channel_axis].value is None:
      raise ValueError('The channel dimension of the inputs '
                       'should be defined. Found `None`.')
    input_dim = int(input_shape[channel_axis])
    self.input_spec = InputSpec(ndim=3, axes={channel_axis: input_dim})
    ke

Links to the raise (1)

https://github.com/keras-team/keras/blob/4a978914d2298db2c79baa4012af5ceff4a4e203/keras/layers/convolutional.py#L966

Ways to fix

When doing convolution using Conv1DTranspose the input tensor should have rank 3. Any 4D tensor given to the layer causes this error.

Reproducing the error:

pipenv install tensorflow

from tensorflow.keras.layers import Conv1DTranspose
import numpy as np
filters=16
kernel_size=7
conv_layer = Conv1DTranspose(filters, 
                             kernel_size, 
                             strides=1, 
                             padding='valid')
inp = np.arange(15).reshape((1,5, 3, 1)).astype(np.float32)
output = conv_layer(inp)
print(output.shape)

The error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-15-94d4abacd6b4> in <module>()
      8                              padding='valid')
      9 img = np.arange(15).reshape((1,5, 3, 1)).astype(np.float32)
---> 10 output = conv_layer(img)
     11 print(output.shape)

/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self, *args, **kwargs)
   1021       with ops.name_scope_v2(name_scope):
   1022         if not self.built:
-> 1023           self._maybe_build(inputs)
   1024 
   1025         if self._autocast:

/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/base_layer.py in _maybe_build(self, inputs)
   2623         # operations.
   2624         with tf_utils.maybe_init_scope(self):
-> 2625           self.build(input_shapes)  # pylint:disable=not-callable
   2626       # We must set also ensure that the layer is marked as built, and the build
   2627       # shape is stored since user defined build functions may not be calling

/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/layers/convolutional.py in build(self, input_shape)
    965     if len(input_shape) != 3:
    966       raise ValueError('Inputs should have rank 3. Received input shape: ' +
--> 967                        str(input_shape))
    968     channel_axis = self._get_channel_axis()
    969     if input_shape.dims[channel_axis].value is None:

ValueError: Inputs should have rank 3. Received input shape: (1, 5, 3, 1)

Fix:

The input tensor to the convolution layer Conv1DTranspose should be a 3D tensor in the form of (batch_size, steps, channels).

from tensorflow.keras.layers import Conv1DTranspose
import numpy as np
filters=16
kernel_size=7
conv_layer = Conv1DTranspose(filters, 
                             kernel_size, 
                             strides=1, 
                             padding='valid')
img = np.arange(15).reshape((5, 3, 1)).astype(np.float32)
output = conv_layer(img)
print(output.shape)