深度学习《Transfer Learning》

基础网络列表：

一、简单迁移代码如下：

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics

baseModel = keras.applications.VGG16(weights="imagenet",include_top=False,input_shape = (512, 512, 3), pooling="None")
baseModel.trainable = False
baseModel.summary()
model = Sequential()
model.add( baseModel )
model.add( layers.Flatten() )
model.add( layers.Dense( units = 512, activation = "relu" ) )
model.add( layers.Dense( units = 2, activation = "softmax" ) )
model.summary()

二、截取任意层作为输出

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics

basemodel = tf.keras.applications.Xception(weights="imagenet",include_top=False,input_shape = (128, 128, 3),pooling=None)
# basemodel.summary()
model = tf.keras.models.Model(inputs=pre_model.input, outputs=pre_model.get_layer('block14_sepconv1_act').output)
model.summary()

Xception网络summary()如下：

组合之后网络summary()如下：

三、多层输出迁移学习代码如下：

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, optimizers, Sequential, metrics

baseModel = keras.applications.VGG16(weights="imagenet",include_top=False,input_shape = (512, 512, 3), pooling=None)
baseModel.trainable = False
# baseModel.summary()
out1 = baseModel.get_layer('block5_pool').output
out2 = baseModel.get_layer('block4_pool').output
out3 = baseModel.get_layer('block3_pool').output
model = keras.models.Model(inputs=baseModel.input, outputs=[out1,out2,out3])
model.summary()

t1 = tf.random.normal([6, 512, 512, 3], mean = 1 , stddev = 1)
pred = model(t1)
print(pred[0].shape, pred[1].shape, pred[2].shape)