Erreur lors de la vérification de l'entrée du modèle : on s'attendait à ce que convolution2d_input_1 ait la forme (None, 3, 32, 32) mais on a obtenu un tableau avec la forme (50000, 32, 32, 3)

Question

Quelqu'un peut-il m'indiquer comment corriger cette erreur ? Je viens de commencer à travailler sur Keras :

 1 from keras.datasets import cifar10
  2 from matplotlib import pyplot
  3 from scipy.misc import toimage
  4 
  5 (x_train, y_train), (x_test, y_test) = cifar10.load_data()
  6 for i in range(0, 9):
  7     pyplot.subplot(330 + 1 + i)
  8     pyplot.imshow(toimage(x_train[i]))
  9 pyplot.show()
 10 
 11 import numpy
 12 from keras.models import Sequential
 13 from keras.layers import Dense
 14 from keras.layers import Dropout
 15 from keras.layers import Flatten
 16 from keras.constraints import maxnorm
 17 from keras.optimizers import SGD
 18 from keras.layers.convolutional import Convolution2D
 19 from keras.layers.convolutional import MaxPooling2D
 20 from keras.utils import np_utils
 21 from keras import backend
 22 backend.set_image_dim_ordering('th')
 23 
 24 seed = 7
 25 numpy.random.seed(seed)
 26 
 27 x_train = x_train.astype('float32')
 28 x_test = x_test.astype('float32')
 29 x_train = x_train / 255.0
 30 x_test = x_test / 255.0
 31 
 32 y_train = np_utils.to_categorical(y_train)
 33 y_test = np_utils.to_categorical(y_test)
 34 num_classes = y_test.shape[1]
 35 
 36 model = Sequential()
 37 model.add(Convolution2D(32, 3, 3, input_shape=(3, 32, 32), border_mode='same', activation='relu', W_constraint=maxnorm(3)))
 38 model.add(Dropout(0.2))
 39 model.add(Convolution2D(32, 3, 3, activation='relu', border_mode='same', W_constraint=maxnorm(3)))
 40 model.add(Flatten())
 41 model.add(Dense(512, activation='relu', W_constraint=maxnorm(3)))
 42 model.add(Dropout(0.5))
 43 model.add(Dense(num_classes, activation='softmax'))
 44 
 45 epochs = 25
 46 learning_rate = 0.01
 47 learning_rate_decay = learning_rate/epochs
 48 sgd = SGD(lr=learning_rate, momentum=0.9, decay=learning_rate_decay, nesterov=False)
 49 model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
 50 print(model.summary())
 51 
 52 model.fit(x_train, y_train, validation_data=(x_test, y_test), nb_epoch=epochs, batch_size=32)
 53 scores = model.evaluate(x_test, y_test, verbose=0)
 54 print("Accuracy: %.2f%%" % (scores[1]*100))

La sortie est :

mona@pascal:/data/wd1$ python test_keras.py 
Using TensorFlow backend.
I tensorflow/stream_executor/dso_loader.cc:111] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:111] successfully opened CUDA library libcudnn.so.5.0 locally
I tensorflow/stream_executor/dso_loader.cc:111] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:111] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:111] successfully opened CUDA library libcurand.so.8.0 locally
____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
convolution2d_1 (Convolution2D)  (None, 32, 32, 32)    896         convolution2d_input_1[0][0]      
____________________________________________________________________________________________________
dropout_1 (Dropout)              (None, 32, 32, 32)    0           convolution2d_1[0][0]            
____________________________________________________________________________________________________
convolution2d_2 (Convolution2D)  (None, 32, 32, 32)    9248        dropout_1[0][0]                  
____________________________________________________________________________________________________
flatten_1 (Flatten)              (None, 32768)         0           convolution2d_2[0][0]            
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 512)           16777728    flatten_1[0][0]                  
____________________________________________________________________________________________________
dropout_2 (Dropout)              (None, 512)           0           dense_1[0][0]                    
____________________________________________________________________________________________________
dense_2 (Dense)                  (None, 10)            5130        dropout_2[0][0]                  
====================================================================================================
Total params: 16,793,002
Trainable params: 16,793,002
Non-trainable params: 0
____________________________________________________________________________________________________
None
Traceback (most recent call last):
  File "test_keras.py", line 52, in <module>
    model.fit(x_train, y_train, validation_data=(x_test, y_test), nb_epoch=epochs, batch_size=32)
  File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 664, in fit
    sample_weight=sample_weight)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1068, in fit
    batch_size=batch_size)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 981, in _standardize_user_data
    exception_prefix='model input')
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 113, in standardize_input_data
    str(array.shape))
ValueError: Error when checking model input: expected convolution2d_input_1 to have shape (None, 3, 32, 32) but got array with shape (50000, 32, 32, 3)

Answer 1

Si vous imprimez x_train.shape vous verrez que la forme est (50000, 32, 32, 3) alors que vous avez donné input_shape=(3, 32, 32) dans la première couche. L'erreur indique simplement que la forme d'entrée attendue et les données fournies sont différentes.

Tout ce que vous devez faire, c'est donner input_shape=(32, 32, 3) . Aussi, si vous utilisez cette forme, vous devez utiliser tf comme commande d'images. backend.set_image_dim_ordering('tf') .

Sinon, vous pouvez permuter l'axe des données.

x_train = x_train.transpose(0,3,1,2)
x_test = x_test.transpose(0,3,1,2)
print x_train.shape