“模型”對象沒有屬性“ _name”

Question

我正在Keras制作CNN。 但是我在制作Keras模型時遇到問題。 這是我的代碼：

x = Input(shape=(256,256,1))
for i in range(16):
    u = int(16 * 2 ** (i//4))
    x = BatchNormalization()(x)
    x1 = Conv2D(u, kernel_size=(1,1), strides=(1,1), activation='relu')(x)
    x1 = MaxPooling2D(pool_size=(3,3), strides=(1,1))(x1)
    x2 = Conv2D(u, kernel_size=(2,2), strides=(1,1), activation='relu')(x)
    x2 = MaxPooling2D(pool_size=(2,2), strides=(1,1))(x2)
    x3 = Conv2D(u, kernel_size=(3,3), strides=(1,1), activation='relu')(x)
    x3 = MaxPooling2D(pool_size=(1,1), strides=(1,1))(x3)
    x = multiply([x1, x2, x3])
    #x = Dropout(0.45)(x)
    x = MaxPooling2D(pool_size=(3,3), strides=(1,1))(x)
out = BatchNormalization()(x)
model = tf.keras.models.Model(inputs=x, outputs=out)

我收到以下錯誤：

AttributeError                            Traceback (most recent call last)
<ipython-input-99-630b3ef0b15f> in <module>()
     13     x = MaxPooling2D(pool_size=(3,3), strides=(1,1))(x)
     14 out = BatchNormalization()(x)
---> 15 model = tf.keras.models.Model(inputs=x, outputs=out)
...

AttributeError: 'Model' object has no attribute '_name'

Answer 1

問題在於，在將x定義為輸入張量之后，您正在將其他張量分配給x 。 因此，它不能用作模型的輸入，即inputs=x 。 要用最少的修改來解決這個問題，只需在將x定義為輸入張量后將x存儲在另一個變量中即可：

x = Input(shape=(256,256,1))
inp = x

# the rest is the same...

model = tf.keras.models.Model(inputs=inp, outputs=out) # pass `inp` as inputs