Keras函数式API中Dense层后Reshape报错求助

问题代码 #

codings_size=10
decoder_inputs = tf.keras.layers.Input(shape=[codings_size])
# x=tf.keras.layers.Flatten(decoder_inputs)
x=tf.keras.layers.Dense(3 * 3 * 16)(decoder_inputs),
x=tf.keras.layers.Reshape((3, 3, 16))(x),

报错信息 #

AttributeError: Exception encountered when calling layer "reshape_28" (type Reshape).

'tuple' object has no attribute 'shape'

Call arguments received by layer "reshape_28" (type Reshape):
  • inputs=('tf.Tensor(shape=(None, 144), dtype=float32)',)

问题原因 #

代码中Dense层和Reshape层的赋值语句末尾多了逗号，导致变量x被定义为元组类型，而Reshape层需要接收张量（Tensor）对象作为输入，元组没有shape属性，因此触发错误。

修正后的代码 #

codings_size=10
decoder_inputs = tf.keras.layers.Input(shape=[codings_size])
# x=tf.keras.layers.Flatten(decoder_inputs)
x = tf.keras.layers.Dense(3 * 3 * 16)(decoder_inputs)
x = tf.keras.layers.Reshape((3, 3, 16))(x)

只需去掉两行赋值语句末尾的逗号，让x保持为张量类型即可解决问题。

内容的提问来源于stack exchange，提问作者Linh Chi Nguyen