Keras 自动编码器输出错误的形状

Question

我正在尝试在 Keras 中构建一个深度卷积自动编码器，但它总是输出错误的形状。

法典：

def build_network(input_shape):
    input_input =  Input(shape=input_shape)
    #Encode
    x = Conv2D(16, (3, 3), activation='relu', padding = 'same')(input_input)
    x = MaxPooling2D((2, 2), padding='same')(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D((2, 2), padding='same')(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D((2, 2), padding='same')(x)
    #Decode
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = UpSampling2D((2, 2))(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = UpSampling2D((2, 2))(x)
    x = Conv2D(16, (3, 3), activation='relu', padding='same')(x) 
    x = UpSampling2D((2, 2))(x)
    decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)
    autoencoder = Model(input_input, decoded)
    return autoencoder

if __name__ == "__main__":
    print(build_network((1, 32, 32)).layers[-1].output)

我希望输出形状与输入形状相同，但它(8, 32, 1)用于(1, 32, 32)

Answer 1

尝试使用 print(build_network((32,32,1)).layers[-1].output) .或者，如果您想首先使用通道，则需要像这样更改模型，

def build_network(input_shape):
    input_input =  Input(shape=input_shape)
    #Encode
    x = Conv2D(16, (3, 3), activation='relu', padding = 'same')(input_input)
    x = MaxPooling2D((2, 2), padding='same')(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D((2, 2), padding='same')(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = MaxPooling2D((2, 2), padding='same')(x)
    #Decode
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = UpSampling2D(size=(2, 2),data_format="channels_first")(x)
    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)
    x = UpSampling2D(size=(2, 2),data_format="channels_first")(x)
    x = Conv2D(16, (3, 3), activation='relu', padding='same')(x) 
    decoded = UpSampling2D(size=(2, 2),data_format="channels_first")(x)
    # decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)
    autoencoder = Model(input_input, decoded)
    return autoencoder
if __name__ == "__main__":
    print(build_network((1, 32, 32)).layers[-1].output)

因为在UpSampling2D中，默认值为"channels_last"。