Keras: ошибка в модели. Оценить с помощью входных фигур

Question

Этот код корректно обучает сеть, но я не понимаю, почему в следующем коде, который реализует сеть Fractional Max Pooling:

import numpy
import keras
from keras.datasets import cifar10
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import Flatten
from keras.constraints import maxnorm
from keras.optimizers import SGD
from keras.utils import np_utils
from keras import backend as K
from keras.layers import Conv2D,Conv1D,LSTM
from keras.layers.core import Reshape
from keras.layers.pooling import MaxPooling1D
from Fractional_MAXPOOL import FractionalPooling2D
from keras.callbacks import ModelCheckpoint
from keras.layers.advanced_activations import LeakyReLU
# K.set_image_dim_ordering('tf')

# fix random seed for reproducibility
seed = 7
numpy.random.seed(seed)

# load data
(X_train, y_train), (X_test, y_test) = cifar10.load_data()
print("X_train.shape: " +str(X_train.shape))
X_valid = X_train[40000:49984]
y_valid = y_train[40000:49984]
print("X_valid.shape: " +str(X_valid.shape))
X_train = X_train[0:40000]
y_train = y_train[0:40000]
X_test = X_test[0:9984]
y_test = y_test[0:9984]

# normalize inputs from 0-255 to 0.0-1.0
X_train = X_train.astype('float32')
X_valid = X_valid.astype('float32')
X_test = X_test.astype('float32')
X_train = X_train / 255.0
X_valid = X_valid / 255.0
X_test = X_test / 255.0

# one hot encode outputs
y_train = np_utils.to_categorical(y_train)
y_valid = np_utils.to_categorical(y_valid)
y_test = np_utils.to_categorical(y_test)
num_classes = y_test.shape[1]

# Create the model
model = Sequential()
# Block 1
model.add(Conv2D(64, (3, 3), batch_input_shape=(64, 32, 32, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(64, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.6, 1.6, 1),pseudo_random = True,overlap=True))
# Block 2
model.add(Conv2D(128, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(128, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.25, 1.25, 1),pseudo_random = True,overlap=True))
# Block 3
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.6, 1.6, 1),pseudo_random = True,overlap=True))
# Block 4
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(256, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.25, 1.25, 1),pseudo_random = True,overlap=True))
# Block 5
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.6, 1.6, 1),pseudo_random = True,overlap=True))
# Block 6
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(Conv2D(512, (3, 3), padding='same'))
model.add(LeakyReLU(alpha = 0.3))
model.add(FractionalPooling2D(pool_ratio=(1, 1.25, 1.25, 1),pseudo_random = True,overlap=True))
model.add(Reshape((16,512)))
# fc layer_1
model.add(Dense(4096, kernel_constraint=maxnorm(3)))
model.add(LeakyReLU(alpha = 0.3))
# fc_layer_2
model.add(Dense(4096, kernel_constraint=maxnorm(3)))
model.add(LeakyReLU(alpha = 0.3))
model.add(Flatten())
model.add(Dense(num_classes, activation='softmax'))

opt = keras.optimizers.Adadelta(0.1,decay=1e-4)

model.compile(loss='categorical_crossentropy', optimizer=opt,metrics=['accuracy'])
print(model.summary())

checkpoint = ModelCheckpoint('Model.hdf5', monitor='val_loss', save_best_only = True, verbose=1, mode='min')

callbacks_list = [checkpoint]
#model.load_weights('Model.hdf5')
epochs = 30
model.fit(X_train, y_train, validation_data = [X_valid,y_valid], nb_epoch=epochs, batch_size=64, callbacks=callbacks_list)

#X_valid = X_valid.reshape(9984, 3072).astype('float32')
#X_test = X_test.reshape(9984, 3072).astype('float32')

# evaluation of the model on the validation set
print("X_valid.shape: " +str(X_valid.shape))
print("y_valid.shape: " +str(y_valid.shape))

scores = model.evaluate(X_valid, y_valid, verbose=0)
print("Validation Accuracy: %.2f%%" % (scores[1]*100))

# Final evaluation of the model
scores = model.evaluate(X_test, y_test, verbose=0)
print("Test Accuracy: %.2f%%" % (scores[1]*100))

Traceback (most recent call last):
  File "cifar_test.py", line 124, in <module>
    scores = model.evaluate(X_valid, y_valid, verbose=0)
  File "/home/filippo/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 1361, ...
tensorflow.python.framework.errors_impl.InvalidArgumentError:  Input to reshape is a tensor with 2097152 values, but the requested shape has 4194304

Я получаю ошибку в строке 124 на model.evaluate (): Вход для изменения формы - это тензор со значениями 2097152, но запрошенная форма имеет 4194304. Я не понимаю, почему форма ввода неправильная.