如何修复“预计会看到 2 个数组，但得到以下 1 个数组列表”

Question

我正在训练一个多类 cnn 模型。 model.fit 方法可以正常工作，但是当我使用 fit_generator 方法时，会出现标题中的错误。

y_train_age = utils.to_categorical(y_train_age, 117)
y_test_age = utils.to_categorical(y_test_age, 117)

y_train_gender = utils.to_categorical(y_train_gender, 2)
y_test_gender = utils.to_categorical(y_test_gender, 2)

y_train = np.concatenate((y_train_age, y_train_gender), axis=1)
y_test = np.concatenate((y_test_age, y_test_gender), axis=1)

print(x_train.shape)
print(x_test.shape)
print(y_train.shape)
print(y_test.shape)

(15000, 100, 100, 3),(8708, 100, 100, 3),(15000, 119),(8708, 119)

型号：

from keras import layers
from keras.models import Model
from keras.layers import Input, Dense, Activation
from keras.layers import AveragePooling2D, MaxPooling2D, Flatten, Conv2D, ZeroPadding2D

x_input = Input((100,100,3))

x = Conv2D(64, (3,3))(x_input)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)

x = Conv2D(64, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)

x = Conv2D(128, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)

x = Conv2D(256, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)

x = Flatten()(x)
x = Dense(64, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(128, activation='relu')(x)

y1 = Dense(117, activation='softmax', name="Age")(x)
y2 = Dense(2, activation='softmax', name="Gender")(x)

model = Model(inputs=x_input, outputs=[y1, y2])
model.compile(loss=['categorical_crossentropy', 'categorical_crossentropy'], optimizer='adam', metrics=['accuracy'])
model.summary()

还有问题：

from keras.preprocessing.image import ImageDataGenerator

model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1, 
                    horizontal_flip=True).flow(x_train, y_train, 32),
                    steps_per_epoch=len(x_train) / 32,
                    epochs=5, verbose=1,
                    validation_data=(x_test, y_test))

错误：

ValueError: Error when checking model target: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 2 array(s), but instead got the following list of 1 arrays: [array([[0., 0., 0., ..., 0., 1., 0.],
       [0., 0., 0., ..., 0., 0., 1.],
       [0., 0., 0., ..., 0., 0., 1.],
       ...,
       [0., 0., 0., ..., 0., 0., 1.],
       [0., 0., 0., ..., 0., 0., 1....

请帮帮我，谢谢。

答案

generator = ImageDataGenerator(...)

def generate_data_generator(generator, X, Y1, Y2):
    genX1 = generator.flow(X, Y1, seed=7)
    genX2 = generator.flow(X, Y2, seed=7)
    while True:
        X1i = genX1.next()
        X2i = genX2.next()
        yield X1i[0], [X1i[1], X2i[1]]

history = model.fit_generator(generate_data_generator(generator, x_train, y_train_age, y_train_gender),
                    steps_per_epoch=len(x_train) / 32,
                    epochs=5, 
                    verbose=1, 
                    callbacks = callbacks,
                    validation_data=(x_test, [y_test_age, y_test_gender]))

Answer 1

您定义了一个具有 2 个输出的模型：[y1, y2]
因此，它需要配备两个不同的标签数组，一个大小为 (, 119)，另一个大小为 (,2)，对应于您的 2 个 Dense 输出层。

使用 fit 函数，它看起来像这样：

model.fit( x = X_train,
           y = [y_train, y_train_gender],
           validation_data=(X_test, [y_test, y_test_gender]),
           batch_size = batch_size,
           epochs = num_epochs,
           verbose = 1)

我不太习惯 ImageDataGenerator，但试试这样的：

from keras.preprocessing.image import ImageDataGenerator

model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1, 
                    horizontal_flip=True).flow(x_train, [y_train, y_train_gender], 32),
                    steps_per_epoch=len(x_train) / 32,
                    epochs=5, verbose=1,
                    validation_data=(x_test, [y_test, y_test_gender]))

编辑

试试这个帖子的小改编：
Keras: How to use fit_generator with multiple outputs of different type

generator = ImageDataGenerator(shear_range=0.3,
                                zoom_range=0.1, 
                                horizontal_flip=True) 


def generate_data_generator(generator, X, Y1, Y2):
    genX1 = generator.flow(X, Y1, seed=7)
    genX2 = generator.flow(X, Y2, seed=7)
    while True:
        X1i = genX1.next()
        X2i = genX2 .next()
        yield X1i[0], [X1i[1], X2i[1]]


model.fit_generator(generate_data_generator(generator, x_train, y_train, y_train_gender),
                    steps_per_epoch=len(x_train) / 32,
                    epochs=5, 
                    verbose=1)

Answer 2

对现有答案稍作修改，您使用的是y_train，它是Age 和Gender 的串联向量，但它应该只包含Age，因为您已经拥有y_train_gender，它编码Gender，我已经对代码的几个部分进行了更改以适应这个

y1 = Dense(117, activation='softmax', name="Age")(x)    # and not 119
y2 = Dense(2, activation='softmax', name="Gender")(x)

只需在您的.fit() 和.fit_generator() 方法中将y_train 替换为y_train_age。这样我们就可以使用y1 作为Age 的输出和y2 作为Gender 的输出。

希望这会有所帮助！