将图层添加到 keras 预训练模型中

Question

我正在尝试构建一个 ResNet50 CNN，用于将图像分类为 5 个不同的类。我首先导入这个模型：

ResNet = ResNet50(
    include_top= None, weights='imagenet', input_tensor=None, input_shape=([128, 217, 3]),
    pooling=None, classes=5)

然后我尝试为分类添加一些最终层：

ResNet.add(Flatten())
ResNet.add(Dense(units=512, activation='relu'))
ResNet.add(Dropout(0.5))
ResNet.add(Dense(units=256, activation='relu'))
ResNet.add(Dropout(0.5))
ResNet.add(Dense(units=5, activation='softmax'))

但是当我尝试这个时，我收到一条错误消息：

AttributeError: 'Functional' object has no attribute 'add'

有人知道如何解决这个问题吗？或者我应该如何尝试添加图层？

Answer 1

您应该执行以下操作：

from tensorflow.keras import models

ResNet = ResNet50(
    include_top= None, weights='imagenet', input_tensor=None, input_shape=([128, 217, 3]),
    pooling=None, classes=5)
model = models.Sequential()
model.add(ResNet)
model.add(Flatten())
model.add(Dense(units=512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=256, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(units=5, activation='softmax'))

您使用的语法是 Sequential() 模块的一部分。

Answer 2

这是因为您的模型不是顺序的。

您必须执行以下任一操作：

last_layer = ResNet.output

x = Flatten()(last_layer)
x = Dense(units=512, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(units=256, activation='relu')(x)
x = (Dropout(0.5)(x)
x = Dense(units=5, activation='softmax')(x)

# prevent the weights from being updated during training
ResNet.trainable = False

model = Model(inputs=ResNet.input, outputs=x)

要么使用顺序模式：

ResNet_load = ResNet50(
    include_top= None, weights='imagenet', input_tensor=None, input_shape=([128, 217, 3]),
    pooling=None, classes=5)

Resnet = Sequential()
ResNet.add(ResNet_load)
ResNet.add(Flatten())
ResNet.add(Dense(units=512, activation='relu'))
ResNet.add(Dropout(0.5))
ResNet.add(Dense(units=256, activation='relu'))
ResNet.add(Dropout(0.5))
ResNet.add(Dense(units=5, activation='softmax'))

Sequential model

Functional API

将图层添加到 keras 预训练模型中

add layers into keras pretrained model

python

neural-network

conv-neural-network

keras

resnet