ValueError：尝试将具有不受支持的类型 (<class 'NoneType'>) 的值 (None) 转换为张量。展平层答案

【问题标题】：ValueError: Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor. Flatten LayerValueError：尝试将具有不受支持的类型 (<class 'NoneType'>) 的值 (None) 转换为张量。展平层
【发布时间】：2021-05-26 19:15:16
【问题描述】：

我正在尝试使用 Keras 的 VGG16，我标记了 include_top=false
但我遇到了ValueError: Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor.
这是代码：

input_shape = (150,150,3)
model_1 = VGG16(weights='imagenet',include_top=False,input_shape=input_shape)
Last_layer=model_1.layers[-1].output
print(Last_layer)
print(type(Last_layer))
Model_Vgg=keras.layers.Flatten()(Last_layer) #<---- error rised here
 
#Model_Vgg=keras.Model(model.input,layer_output)

Model_Vgg = layers.Dropout(0.5)(Model_Vgg)


Model_Vgg = layers.Dense(units=3, activation='softmax') (Model_Vgg)

model = keras.Model(inputs =model_1.input,outputs = Model_Vgg )
model.compile(loss='categorical_crossentropy',optimizer=optimizers.SGD(lr=0.005708),metrics=['accuracy'])

monitor = EarlyStopping(monitor='accuracy',patience=50, mode='auto', restore_best_weights=True)
model.fit(X_Train,Y_Train,callbacks=[monitor],epochs=280,verbose=0)
(loss, accuracy) = model.evaluate(X_Test, Y_Test, batch_size=32, verbose=50)
print("[INFO] loss={:.4f}, accuracy: {:.4f}%".format(loss,accuracy * 100))

它显示print(type(Last_layer)) = <class 'keras.engine.keras_tensor.KerasTensor'>
我不知道为什么该行引用无类型对象

【问题讨论】：

标签： python tensorflow machine-learning keras deep-learning

【解决方案1】：

我能够复制您的问题，如下所示

import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras.applications.vgg16 import VGG16

input_shape = (150,150,3)
model_1 = VGG16(weights='imagenet',include_top=False,input_shape=input_shape)
Last_layer=model_1.layers[-1].output
#print(Last_layer)
#print(type(Last_layer))
Model_Vgg=keras.layers.Flatten()(Last_layer)
Model_Vgg = layers.Dropout(0.5)(Model_Vgg)
Model_Vgg = layers.Dense(units=3, activation='softmax') (Model_Vgg)

model = keras.Model(inputs =model_1.input,outputs = Model_Vgg )
model.compile(loss='categorical_crossentropy',optimizer=tf.keras.optimizers.SGD(learning_ratek=0.005708),metrics=['accuracy'])

输出：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-20-3d087167b224> in <module>()
      8 #print(Last_layer)
      9 #print(type(Last_layer))
---> 10 Model_Vgg=keras.layers.Flatten()(Last_layer)
     11 Model_Vgg = layers.Dropout(0.5)(Model_Vgg)
     12 Model_Vgg = layers.Dense(units=3, activation='softmax') (Model_Vgg)

5 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
     96       dtype = dtypes.as_dtype(dtype).as_datatype_enum
     97   ctx.ensure_initialized()
---> 98   return ops.EagerTensor(value, ctx.device_name, dtype)
     99 
    100 

ValueError: Attempt to convert a value (None) with an unsupported type (<class 'NoneType'>) to a Tensor.

固定代码：

将keras.layers.Flatten() 替换为layers.Flatten() 即可解决您的问题。

工作代码如下图

import tensorflow as tf
from tensorflow.keras import layers, Model
from tensorflow.keras.applications.vgg16 import VGG16

input_shape = (150,150,3)
model_1 = VGG16(weights='imagenet',include_top=False,input_shape=input_shape)
Last_layer=model_1.layers[-1].output
print(Last_layer)
print(type(Last_layer))
Model_Vgg=layers.Flatten()(Last_layer)
Model_Vgg = layers.Dropout(0.5)(Model_Vgg)
Model_Vgg = layers.Dense(units=3, activation='softmax') (Model_Vgg)

model = Model(inputs =model_1.input,outputs = Model_Vgg )
model.compile(loss='categorical_crossentropy',optimizer=tf.keras.optimizers.SGD(learning_rate=0.005708),metrics=['accuracy'])

输出：

KerasTensor(type_spec=TensorSpec(shape=(None, 4, 4, 512), dtype=tf.float32, name=None), name='block5_pool/MaxPool:0', description="created by layer 'block5_pool'")
<class 'tensorflow.python.keras.engine.keras_tensor.KerasTensor'>

注意：切勿混合使用 keras 和 tensorflow。

【讨论】：

【解决方案2】：

我的代码也有同样的问题：

from keras.layers import Dense, Flatten
x = vgg.output(Flatten())

然后我改成

from tensorflow.keras import layers
x = layers.Flatten()(vgg.output)

它成功了。

【讨论】：

我找到了一些解决方案：

【解决方案3】：

我找到了这个解决方案，它对我有用

def Create_Model():
  #input_shape = (150,150,3)
  model_1 = VGG16(weights='imagenet',include_top=False)
  input = keras.layers.Input(shape=(150,150,3))
  Last_layer=model_1(input)

  Model_Vgg=keras.layers.Flatten()(Last_layer)   
  #Model_Vgg=keras.Model(model.input,layer_output)
  Model_Vgg = layers.Dropout(0.5)(Model_Vgg)
  Model_Vgg = layers.Dense(units=3, activation='softmax') (Model_Vgg)
  model = keras.Model(inputs =input,outputs = Model_Vgg )
  return model

【讨论】：