Keras —— 迁移学习fine-tuning

该程序演示将一个预训练好的模型在新数据集上重新fine-tuning的过程。我们冻结卷积层，只调整全连接层。

在MNIST数据集上使用前五个数字[0…4]训练一个卷积网络。

在后五个数字[5…9]用卷积网络做分类，冻结卷积层并且微调全连接层

一、变量初始化

now = datetime.datetime.now

batch_size = 128
nb_classes = 5
nb_epoch = 5

# 输入图像的维度
img_rows, img_cols = 28, 28
# 使用卷积滤波器的数量
nb_filters = 32
# 用于max pooling的pooling面积的大小
pool_size = 2
# 卷积核的尺度
kernel_size = (3,3)
input_shape = (img_rows, img_cols, 1)

# 数据，在训练和测试数据集上混洗和拆分
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train_lt5 = X_train[y_train < 5]
y_train_lt5 = y_train[y_train < 5]
X_test_lt5 = X_test[y_test < 5]
y_test_lt5 = y_test[y_test < 5]

X_train_gte5 = X_train[y_train >= 5]
#使标签从0~4，故-5
y_train_gte5 = y_train[y_train >= 5] - 5
X_test_gte5 = X_test[y_test >= 5]
y_test_gte5 = y_test[y_test >= 5] - 5

二、模型的训练函数

def train_model(model, train, test, nb_classes):
#train[0]是图片，train[1]是标签
X_train = train[0].reshape((train[0].shape[0],) + input_shape)#1D+3D=4D
X_test = test[0].reshape((test[0].shape[0],) + input_shape)
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
X_train /= 255
X_test /= 255
print('X_train shape:', X_train.shape)
print(X_train.shape[0], 'train samples')
print(X_test.shape[0], 'test samples')
Y_train = np_utils.to_categorical(train[1], nb_classes)
Y_test = np_utils.to_categorical(test[1], nb_classes)
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy'])
t = now()
model.fit(X_train, Y_train,
batch_size=batch_size, nb_epoch=nb_epoch,
verbose=1,
validation_data=(X_test, Y_test))
print('Training time: %s' % (now() - t))
score = model.evaluate(X_test, Y_test, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])

三、建立模型，构建卷积层（特征层）和全连接层（分类层）

feature_layers = [
Convolution2D(nb_filters, kernel_size,
padding='valid',
input_shape=input_shape),
Activation('relu'),
Convolution2D(nb_filters, kernel_size),
Activation('relu'),
MaxPooling2D(pool_size=(pool_size, pool_size)),
Dropout(0.25),
Flatten(),
]
classification_layers = [
Dense(128),
Activation('relu'),
Dropout(0.5),
Dense(nb_classes),
Activation('softmax')
]
model = Sequential(feature_layers + classification_layers)

四、对模型进行预训练

train_model(model,
(X_train_lt5, y_train_lt5),
(X_test_lt5, y_test_lt5), nb_classes)

五、冻结预训练模型的特征层

for l in feature_layers:
l.trainable = False

六、fine_tuning分类层

train_model(model,
(X_train_gte5, y_train_gte5),
(X_test_gte5, y_test_gte5), nb_classes)

源码地址：

https://github.com/Zheng-Wenkai/Keras_Demo