如何在 Keras 中获取图层的输出形状？

Question

I have the following code in Keras (Basically I am modifying this code for my use) and I get this error:

'ValueError: Error when checking target: expected conv3d_3 to have 5 dimensions, but got array with shape (10, 4096)'

Code:

from keras.models import Sequential
from keras.layers.convolutional import Conv3D
from keras.layers.convolutional_recurrent import ConvLSTM2D
from keras.layers.normalization import BatchNormalization
import numpy as np
import pylab as plt
from keras import layers

# We create a layer which take as input movies of shape
# (n_frames, width, height, channels) and returns a movie
# of identical shape.

model = Sequential()
model.add(ConvLSTM2D(filters=40, kernel_size=(3, 3),
                   input_shape=(None, 64, 64, 1),
                   padding='same', return_sequences=True))
model.add(BatchNormalization())

model.add(ConvLSTM2D(filters=40, kernel_size=(3, 3),
                   padding='same', return_sequences=True))
model.add(BatchNormalization())

model.add(ConvLSTM2D(filters=40, kernel_size=(3, 3),
                   padding='same', return_sequences=True))
model.add(BatchNormalization())

model.add(ConvLSTM2D(filters=40, kernel_size=(3, 3),
                   padding='same', return_sequences=True))
model.add(BatchNormalization())

model.add(Conv3D(filters=1, kernel_size=(3, 3, 3),
               activation='sigmoid',
               padding='same', data_format='channels_last'))
model.compile(loss='binary_crossentropy', optimizer='adadelta')

the data I feed is in the following format: [1, 10, 64, 64, 1]. So I would like to know where I am wrong and also how to see the output_shape of each layer.

Answer 1

You can get the output shape of a layer by layer.output_shape .

for layer in model.layers:
    print(layer.output_shape)

Gives you:

(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 40)
(None, None, 64, 64, 1)

Alternatively you can pretty print the model using model.summary :

model.summary()

Gives you the details about the number of parameters and output shapes of each layer and an overall model structure in a pretty format:

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv_lst_m2d_1 (ConvLSTM2D)  (None, None, 64, 64, 40)  59200     
_________________________________________________________________
batch_normalization_1 (Batch (None, None, 64, 64, 40)  160       
_________________________________________________________________
conv_lst_m2d_2 (ConvLSTM2D)  (None, None, 64, 64, 40)  115360    
_________________________________________________________________
batch_normalization_2 (Batch (None, None, 64, 64, 40)  160       
_________________________________________________________________
conv_lst_m2d_3 (ConvLSTM2D)  (None, None, 64, 64, 40)  115360    
_________________________________________________________________
batch_normalization_3 (Batch (None, None, 64, 64, 40)  160       
_________________________________________________________________
conv_lst_m2d_4 (ConvLSTM2D)  (None, None, 64, 64, 40)  115360    
_________________________________________________________________
batch_normalization_4 (Batch (None, None, 64, 64, 40)  160       
_________________________________________________________________
conv3d_1 (Conv3D)            (None, None, 64, 64, 1)   1081      
=================================================================
Total params: 407,001
Trainable params: 406,681
Non-trainable params: 320
_________________________________________________________________

If you want to access information about a specific layer only, you can use name argument when constructing that layer and then call like this:

...
model.add(ConvLSTM2D(..., name='conv3d_0'))
...

model.get_layer('conv3d_0')

---

EDIT: For reference sake it will always be same as layer.output_shape and please don't actually use Lambda or custom layers for this. But you can use Lambda layer to echo the shape of a passing tensor.

...
def print_tensor_shape(x):
    print(x.shape)
    return x
model.add(Lambda(print_tensor_shape))
...

Or write a custom layer and print the shape of the tensor on call() .

class echo_layer(Layer):
...
    def call(self, x):
        print(x.shape)
        return x
...

model.add(echo_layer())

Answer 2

You can get the output shape with 'output.shape[1:]' command. It will get the shape of output layer and can be used for other purposes.