ValueError: A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 523, 523, 32), etc

Question

I am trying to concatenate layers Using the following code using tensorflow , but getting an unexpected error. I am new to tensorflow

inp = Input(shape=(1050,1050,3))
x1= layers.Conv2D(16 ,(3,3), activation='relu')(inp)
x1= layers.Conv2D(32,(3,3), activation='relu')(x1)
x1= layers.MaxPooling2D(2,2)(x1)
x2= layers.Conv2D(32,(3,3), activation='relu')(x1)
x2= layers.Conv2D(64,(3,3), activation='relu')(x2)
x2= layers.MaxPooling2D(3,3)(x2)
x3= layers.Conv2D(64,(3,3), activation='relu')(x2)
x3= layers.Conv2D(64,(2,2), activation='relu')(x3)
x3= layers.Conv2D(64,(3,3), activation='relu')(x3)
x3= layers.Dropout(0.2)(x3)
x3= layers.MaxPooling2D(2,2)(x3)
x4= layers.Conv2D(64,(3,3), activation='relu')(x3)
x4= layers.MaxPooling2D(2,2)(x4)
x = layers.Dropout(0.2)(x4)
o = layers.Concatenate(axis=3)([x1, x2, x3, x4, x])
y = layers.Flatten()(o)
y = layers.Dense(1024, activation='relu')(y)
y = layers.Dense(5, activation='softmax')(y) 

model = Model(inp, y)
model.summary()
model.compile(loss='sparse_categorical_crossentropy',optimizer=RMSprop(lr=0.001),metrics=['accuracy'])

The main error can be seen in the heading But i have provided the traceback error for reference And the error is

ValueError                                Traceback (most recent call last)
<ipython-input-12-31a1fcec98a4> in <module>
     14 x4= layers.MaxPooling2D(2,2)(x4)
     15 x = layers.Dropout(0.2)(x4)
---> 16 o = layers.Concatenate(axis=3)([x1, x2, x3, x4, x])
     17 y = layers.Flatten()(o)
     18 y = layers.Dense(1024, activation='relu')(y)

/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self, inputs, *args, **kwargs)
    589           # Build layer if applicable (if the `build` method has been
    590           # overridden).
--> 591           self._maybe_build(inputs)
    592 
    593           # Wrapping `call` function in autograph to allow for dynamic control

/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py in _maybe_build(self, inputs)
   1879       # operations.
   1880       with tf_utils.maybe_init_scope(self):
-> 1881         self.build(input_shapes)
   1882     # We must set self.built since user defined build functions are not
   1883     # constrained to set self.built.

/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/utils/tf_utils.py in wrapper(instance, input_shape)
    293     if input_shape is not None:
    294       input_shape = convert_shapes(input_shape, to_tuples=True)
--> 295     output_shape = fn(instance, input_shape)
    296     # Return shapes from `fn` as TensorShapes.
    297     if output_shape is not None:

/opt/conda/lib/python3.6/site-packages/tensorflow/python/keras/layers/merge.py in build(self, input_shape)
    389                        'inputs with matching shapes '
    390                        'except for the concat axis. '
--> 391                        'Got inputs shapes: %s' % (input_shape))
    392 
    393   def _merge_function(self, inputs):

ValueError: A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 523, 523, 32), (None, 173, 173, 64), (None, 84, 84, 64), (None, 41, 41, 64), (None, 41, 41, 64)]

I have imported all necessary files required to run the code using tensorflow.keras

Answer 1

You cannot perform concatenate operation of input with different dimension i.e height and width. In your case you are trying to perform this operation layers.Concatenate(axis=3)([x1, x2, x3, x4, x]) where

x1 has dimension = (None, 523, 523, 32)
x2 has dimension = (None, 173, 173, 64)
x3 has dimension = (None, 84, 84, 64)
x4 has dimension = (None, 41, 41, 64)
and x has dimension = (None, 41, 41, 64)

The error occurred because, all of the input dimension i.e height and width to concatenate are different. To solve the error you will have to get all the input to same dimension i.e same height and width, this could be achieved by sampling the layers to a fixed dimension. Based on your use case you could either downsample or upsample to achieve the required dimension.

ValueError: A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 523, 523, 32), (None, 173, 173, 64), (None, 84, 84, 64), (None, 41, 41, 64), (None, 41, 41, 64)]

The error states layer requires inputs with matching shapes, this are nothing but the height and width of the input.