图断开连接：无法获取张量 KerasTensor() 迁移学习的值

Question

我正在尝试在我自己的模型上实施迁移学习，但失败了。我的实现遵循此处的指南

https://keras.io/guides/transfer_learning/

How to do transfer-learning on our own models?

https://github.com/anujshah1003/Transfer-Learning-in-keras---custom-data/blob/master/transfer_learning_resnet50_custom_data.py

tensoflow 2.4.1
Keras 2.4.3

旧模型（效果很好）：

model = Sequential()
inputShape = (256, 256, 3)
chanDim = -1
  
# CONV => RELU => POOL
model.add(Conv2D(32, (3, 3), padding="same", input_shape=inputShape))
model.add(Activation("relu"))
model.add(BatchNormalization(axis=chanDim))
model.add(MaxPooling2D(pool_size=(3, 3)))
model.add(Dropout(0.25))

# (CONV => RELU) * 2 => POOL
model.add(Conv2D(64, (3, 3), padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization(axis=chanDim))

model.add(Conv2D(64, (3, 3), padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization(axis=chanDim))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

# (CONV => RELU) * 2 => POOL
model.add(Conv2D(128, (3, 3), padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization(axis=chanDim))
model.add(Conv2D(128, (3, 3), padding="same"))
model.add(Activation("relu"))
model.add(BatchNormalization(axis=chanDim))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

# first (and only) set of FC => RELU layers
model.add(Flatten())
model.add(Dense(1024))
model.add(Activation("relu"))
model.add(BatchNormalization())
model.add(Dropout(0.5))

# softmax classifier
model.add(Dense(classes))
model.add(Activation("softmax"))

迁移学习：

old_model = load_model('old.model')

# removes top 2 activation layers
for i in range(2):
  old_model.pop()

# mark loaded layers as not trainable
for layer in old_model.layers:
    layer.trainable = False

# initialize the new model

in_puts = Input(shape=(256, 256, 3))
count = len(old_model.layers)
ll = old_model.layers[count - 1].output
classes = len(lb.classes_)
ll = Dense(classes)(ll)
ll = Activation("softmax", name="activation3_" + NODE)(ll)
model = Model(inputs=in_puts, outputs=ll) # ERROR

opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="categorical_crossentropy", optimizer=opt, metrics=["accuracy"])

# train the network
H = model.fit(x_train, y_train, validation_data=(x_test, y_test), steps_per_epoch=len(x_train) // BS, epochs=EPOCHS, verbose=1)


# save the model to disk
model.save("new.model")

错误

ValueError: Graph disconnected: cannot obtain value for tensor
KerasTensor(type_spec=TensorSpec(shape=(None, 256, 256, 3), dtype=tf.float32,
name='conv2d_input'), name='conv2d_input', description="created by layer 'conv2d_input'") at
layer "conv2d". The following previous layers were accessed without issue: []

Answer 1

这是使用模型操作迁移学习的简单方法

classes = 10
sub_old_model = Model(old_model.input, old_model.layers[-3].output)
sub_old_model.trainable = False

ll = Dense(classes)(sub_old_model.output)
ll = Activation("softmax")(ll)

model = Model(inputs=sub_old_model.input, outputs=ll) 

首先，创建一个子模型，其中包含要冻结的旧模型的层 (trainable = False)。在我们的示例中，我们采用除最后一个Dense 和Softmax 激活之外的所有层。

然后将子模型输出传递到新的可训练层。

此时，您只需要创建一个新模型实例来组装所有部件

Answer 2

在@Marco Cerliani 的帮助下，我能够通过更改为LabelEncoder 和sparse_categorical_crossentropy 来解决问题