tensorflow 2.0 深度学习 (第三部分卷积神经网络 part1)

拼搏现实的明天。 2023-07-02 05:21 53阅读 0赞

# ![20191009191333910.png][] #

# [日萌社][Link 1] #

[人工智能AI：Keras PyTorch MXNet TensorFlow PaddlePaddle 深度学习实战（不定时更新）][AI_Keras PyTorch MXNet TensorFlow PaddlePaddle]

--------------------

[tensorflow 2.0 深度学习（第一部分 part1）][tensorflow 2.0 _ part1]

[tensorflow 2.0 深度学习（第一部分 part2）][tensorflow 2.0 _ part2]

[tensorflow 2.0 深度学习（第一部分 part3）][tensorflow 2.0 _ part3]

[tensorflow 2.0 深度学习（第二部分 part1）][tensorflow 2.0 _ part1 1]

[tensorflow 2.0 深度学习（第二部分 part2）][tensorflow 2.0 _ part2 1]

[tensorflow 2.0 深度学习（第二部分 part3）][tensorflow 2.0 _ part3 1]

[tensorflow 2.0 深度学习 (第三部分 卷积神经网络 part1)][tensorflow 2.0 _ _ _ part1]

[tensorflow 2.0 深度学习 (第三部分 卷积神经网络 part2)][tensorflow 2.0 _ _ _ part2]

[tensorflow 2.0 深度学习（第四部分 循环神经网络）][tensorflow 2.0 _]

[tensorflow 2.0 深度学习（第五部分 GAN生成神经网络 part1）][tensorflow 2.0 _ GAN_ part1]

[tensorflow 2.0 深度学习（第五部分 GAN生成神经网络 part2）][tensorflow 2.0 _ GAN_ part2]

[tensorflow 2.0 深度学习（第六部分 强化学习）][tensorflow 2.0 _ 1]

--------------------

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70][]

Model: "sequential"
    _________________________________________________________________
    Layer (type)                 Output Shape              Param #
    =================================================================
    dense (Dense)                multiple                  200960
    _________________________________________________________________
    dense_1 (Dense)              multiple                  65792
    _________________________________________________________________
    dense_2 (Dense)              multiple                  65792
    _________________________________________________________________
    dense_3 (Dense)              multiple                  2570
    =================================================================
    Total params: 335,114
    Trainable params: 335,114
    Non-trainable params: 0
    _________________________________________________________________

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 1][]

import tensorflow as tf
    from tensorflow import keras
    # 获取所有 GPU 设备列表
    gpus = tf.config.experimental.list_physical_devices('GPU')
    #gpus 打印为 [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
    if gpus:
        try:
            # 设置 GPU 显存占用为按需分配
            for gpu in gpus:
                tf.config.experimental.set_memory_growth(gpu, True)
            logical_gpus = tf.config.experimental.list_logical_devices('GPU')
            #1 Physical GPUs, 1 Logical GPUs
            print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
        except RuntimeError as e:
            # 异常处理
            print(e)

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 2][]

--------------------

# **局部相关性**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 3][] #

--------------------

# **权值共享** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 4][]

--------------------

# **卷积运算原理** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 5][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 6][]

--------------------

# **表示学习** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 7][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 8][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 9][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 10][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 11][]

--------------------

# **梯度传播**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 12][] #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 13][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 14][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 15][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 16][]

--------------------

# **卷积神经网络 基础组成部分** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 17][]

--------------------

# **单通道输入 单卷积核** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 18][]

--------------------

# **多通道输入 单卷积核** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 19][]

--------------------

# **多通道输入 多卷积核**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 20][] #

--------------------

# **步长** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 21][]

--------------------

# **填充**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 22][] #

#  #

--------------------

# **自定义权值/偏置：tf.nn.conv2d卷积运算函数** #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 23][]

--------------------

# **卷积层类 layers.Conv2D **![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 24][] #

import  os
    os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
    import  tensorflow as tf
    from    tensorflow import keras
    from    tensorflow.keras import layers, optimizers, datasets, Sequential
    
    # 模拟输入，3通道，高宽为5
    x = tf.random.normal([2,5,5,3]) 
    # 需要根据[k,k,cin,cout]格式创建，4个卷积核
    w = tf.random.normal([3,3,3,4]) 
    # 步长为1, padding为0，即padding=[[0,0],[0,0],[0,0],[0,0]]
    out = tf.nn.conv2d(x,w, strides=1, padding=[[0,0],[0,0],[0,0],[0,0]])
    
    # 模拟输入，3通道，高宽为5
    x = tf.random.normal([2,5,5,3]) 
    # 需要根据[k,k,cin,cout]格式创建，4个卷积核
    w = tf.random.normal([3,3,3,4])
    # 步长为1, padding为1，即padding=[[0,0],[1,1],[1,1],[0,0]]
    out = tf.nn.conv2d(x,w, strides=1, padding=[[0,0],[1,1],[1,1],[0,0]])
    
    # 模拟输入，3通道，高宽为5
    x = tf.random.normal([2,5,5,3])
    # 需要根据[k,k,cin,cout]格式创建，4个3x3大小的卷积核 
    w = tf.random.normal([3,3,3,4]) 
    # 步长为1，padding设置为输出、输入同大小，即padding='SAME'
    # 需要注意的是, padding='SAME'只有在strides=1时才是输出、输入同大小
    out = tf.nn.conv2d(x,w, strides=1, padding='SAME')
     
    # 模拟输入，3通道，高宽为5
    x = tf.random.normal([2,5,5,3])
    # 需要根据[k,k,cin,cout]格式创建，4个3x3大小的卷积核 
    w = tf.random.normal([3,3,3,4])
    # strides=3：高宽按3倍减少
    # padding设置为输出、输入同大小，即padding='SAME'。
    # padding='SAME'只有在strides=1时才是输出、输入同大小。
    out = tf.nn.conv2d(x,w, strides=3, padding='SAME')
    print(out.shape)
     
    # 根据[cout]格式创建偏置向量
    b = tf.zeros([4])
    # 在卷积输出上叠加偏置向量，它会自动broadcasting为[b,h',w',cout]
    out = out + b
     
    # 创建卷积层类
    # padding设置为输出、输入同大小，即padding='SAME'。
    # padding='SAME'只有在strides=1时才是输出、输入同大小。
    layer = layers.Conv2D(4,kernel_size=(3,4),strides=(2,1),padding='SAME')
    out = layer(x) # 前向计算
    out.shape
     
    layer.kernel, layer.bias
    # 返回所有待优化张量列表
    layer.trainable_variables
     
    from tensorflow.keras import Sequential
    # 网络容器
    network = Sequential([ 
        layers.Conv2D(6,kernel_size=3,strides=1), # 第一个卷积层, 6个3x3卷积核
        layers.MaxPooling2D(pool_size=2,strides=2), # 高宽各减半的池化层
        layers.ReLU(), # 激活函数
        layers.Conv2D(16,kernel_size=3,strides=1), # 第二个卷积层, 16个3x3卷积核
        layers.MaxPooling2D(pool_size=2,strides=2), # 高宽各减半的池化层
        layers.ReLU(), # 激活函数
        layers.Flatten(), # 打平层，方便全连接层处理
        layers.Dense(120, activation='relu'), # 全连接层，120个节点
        layers.Dense(84, activation='relu'), # 全连接层，84节点
        layers.Dense(10) # 全连接层，10个节点
                        ])
    # build一次网络模型，给输入X的形状，其中4为随意给的batchsz
    network.build(input_shape=(4, 28, 28, 1))
    # 统计网络信息
    network.summary()
     
    # 导入误差计算，优化器模块
    from tensorflow.keras import losses, optimizers
    # 创建损失函数的类，在实际计算时直接调用类实例即可
    criteon = losses.CategoricalCrossentropy(from_logits=True)
     
    # 构建梯度记录环境
    with tf.GradientTape() as tape: 
    	# 插入通道维度，=>[b,28,28,1]
    	x = tf.expand_dims(x,axis=3)
    	# 前向计算，获得10类别的预测分布，[b, 784] => [b, 10]
    	out = network(x)
    	# 真实标签one-hot编码，[b] => [b, 10]
    	y_onehot = tf.one_hot(y, depth=10)
    	# 计算交叉熵损失函数，标量
    	loss = criteon(y_onehot, out)
    # 自动计算梯度
    grads = tape.gradient(loss, network.trainable_variables)
    # 自动更新参数
    optimizer.apply_gradients(zip(grads, network.trainable_variables))
    
    # 记录预测正确的数量，总样本数量
    correct, total = 0,0
    for x,y in db_test: # 遍历所有训练集样本
    	# 插入通道维度，=>[b,28,28,1]
    	x = tf.expand_dims(x,axis=3)
    	# 前向计算，获得10类别的预测分布，[b, 784] => [b, 10]
    	out = network(x)
    	# 真实的流程时先经过softmax，再argmax
    	# 但是由于softmax不改变元素的大小相对关系，故省去
    	pred = tf.argmax(out, axis=-1)  
    	y = tf.cast(y, tf.int64)
    	# 统计预测正确数量
    	correct += float(tf.reduce_sum(tf.cast(tf.equal(pred, y),tf.float32)))
    	# 统计预测样本总数
    	total += x.shape[0]
    # 计算准确率
    print('test acc:', correct/total)

![20200126165900343.png][]

--------------------

# **池化层**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 25][] #

--------------------

# **BatchNorm 层**![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 26][] #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 27][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 28][]

# 构造输入
    x=tf.random.normal([100,32,32,3])
    # 将其他维度合并，仅保留通道维度
    x=tf.reshape(x,[-1,3])
    # 计算其他维度的均值，不计算通道维度
    ub=tf.reduce_mean(x,axis=0)
    ub
    
    # 创建BN层
    layer=layers.BatchNormalization()
    # 网络容器
    network = Sequential([ 
        layers.Conv2D(6,kernel_size=3,strides=1),
        # 插入BN层
        layers.BatchNormalization(),
        layers.MaxPooling2D(pool_size=2,strides=2),
        layers.ReLU(),
        layers.Conv2D(16,kernel_size=3,strides=1),
        # 插入BN层
        layers.BatchNormalization(),
        layers.MaxPooling2D(pool_size=2,strides=2),
        layers.ReLU(),
        layers.Flatten(),
        layers.Dense(120, activation='relu'),
        # 此处也可以插入BN层
        layers.Dense(84, activation='relu'), 
        # 此处也可以插入BN层
        layers.Dense(10)
                        ])
    
    with tf.GradientTape() as tape: 
    	# 插入通道维度
    	x = tf.expand_dims(x,axis=3)
    	# 前向计算，设置计算模式，[b, 784] => [b, 10]
    	# BN需要设置training=True
    	out = network(x, training=True)
    	# 遍历测试集
    	for x,y in db_test: 
    		# 插入通道维度
    		x = tf.expand_dims(x,axis=3)
    		# 前向计算，测试模式。BN需要设置training=True。
    		out = network(x, training=False)

import tensorflow as tf
    from tensorflow import keras
    from tensorflow.keras import layers, optimizers
    
    # 2 images with 4x4 size, 3 channels
    # we explicitly enforce the mean and stddev to N(1, 0.5)
    x = tf.random.normal([2,4,4,3], mean=1.,stddev=0.5)
    x.shape #TensorShape([2, 4, 4, 3])
    
    net = layers.BatchNormalization(axis=-1, center=True, scale=True, trainable=True)
    out = net(x)
    print('forward in test mode:', net.variables)
    #forward in test mode: [
    #<tf.Variable 'batch_normalization/gamma:0' shape=(3,) dtype=float32, numpy=array([1., 1., 1.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/beta:0' shape=(3,) dtype=float32, numpy=array([0., 0., 0.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_mean:0' shape=(3,) dtype=float32, numpy=array([0., 0., 0.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_variance:0' shape=(3,) dtype=float32, numpy=array([1., 1., 1.], dtype=float32)>]
    
    out = net(x, training=True)
    print('forward in train mode(1 step):', net.variables)
    #forward in train mode(1 step): [
    #<tf.Variable 'batch_normalization/gamma:0' shape=(3,) dtype=float32, numpy=array([1., 1., 1.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/beta:0' shape=(3,) dtype=float32, numpy=array([0., 0., 0.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_mean:0' shape=(3,) dtype=float32, numpy=array([0.01059513, 0.01055362, 0.01071654], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_variance:0' shape=(3,) dtype=float32, numpy=array([0.99315214, 0.99234647, 0.9939004 ], dtype=float32)>]
    
    for i in range(100):
        out = net(x, training=True)
    print('forward in train mode(100 steps):', net.variables)
    #forward in train mode(100 steps): [
    #<tf.Variable 'batch_normalization/gamma:0' shape=(3,) dtype=float32, numpy=array([1., 1., 1.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/beta:0' shape=(3,) dtype=float32, numpy=array([0., 0., 0.], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_mean:0' shape=(3,) dtype=float32, numpy=array([0.6755756, 0.6729286, 0.6833168], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_variance:0' shape=(3,) dtype=float32, numpy=array([0.5633595 , 0.51199085, 0.6110718 ], dtype=float32)>]
    
    #SGD随机梯度下降
    optimizer = optimizers.SGD(lr=1e-2)
    for i in range(10):
        with tf.GradientTape() as tape:
            out = net(x, training=True)
            loss = tf.reduce_mean(tf.pow(out,2)) - 1
        grads = tape.gradient(loss, net.trainable_variables)
        optimizer.apply_gradients(zip(grads, net.trainable_variables))
    print('backward(10 steps):', net.variables)
    #backward(10 steps): [
    #<tf.Variable 'batch_normalization/gamma:0' shape=(3,) dtype=float32, numpy=array([0.9355032 , 0.93557316, 0.935464  ], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/beta:0' shape=(3,) dtype=float32, numpy=array([ 7.4505802e-09, -4.2840842e-09,  5.2154064e-10], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_mean:0' shape=(3,) dtype=float32, numpy=array([0.71228695, 0.70949614, 0.7204489 ], dtype=float32)>, 
    #<tf.Variable 'batch_normalization/moving_variance:0' shape=(3,) dtype=float32, numpy=array([0.5396321 , 0.485472  , 0.58993715], dtype=float32)>]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 29][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 30][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 31][]

[20191009191333910.png]: https://img-blog.csdnimg.cn/20191009191333910.png
[Link 1]: http://www.rimengshe.com/
[AI_Keras PyTorch MXNet TensorFlow PaddlePaddle]: https://blog.csdn.net/zimiao552147572/article/details/88867161
[tensorflow 2.0 _ part1]: https://azusa.blog.csdn.net/article/details/104086445
[tensorflow 2.0 _ part2]: https://azusa.blog.csdn.net/article/details/104086557
[tensorflow 2.0 _ part3]: https://azusa.blog.csdn.net/article/details/104086663
[tensorflow 2.0 _ part1 1]: https://azusa.blog.csdn.net/article/details/104086741
[tensorflow 2.0 _ part2 1]: https://azusa.blog.csdn.net/article/details/104086773
[tensorflow 2.0 _ part3 1]: https://azusa.blog.csdn.net/article/details/104087912
[tensorflow 2.0 _ _ _ part1]: https://azusa.blog.csdn.net/article/details/104088133
[tensorflow 2.0 _ _ _ part2]: https://azusa.blog.csdn.net/article/details/104088230
[tensorflow 2.0 _]: https://azusa.blog.csdn.net/article/details/104088366
[tensorflow 2.0 _ GAN_ part1]: https://azusa.blog.csdn.net/article/details/104088512
[tensorflow 2.0 _ GAN_ part2]: https://azusa.blog.csdn.net/article/details/104088559
[tensorflow 2.0 _ 1]: https://azusa.blog.csdn.net/article/details/104089741
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/2020012616465930.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200126164720722.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200126164729349.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200126164742523.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20200126164816655.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20200126164846679.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/20200126164857784.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 7]: https://img-blog.csdnimg.cn/20200126164952545.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 8]: https://img-blog.csdnimg.cn/20200126165046287.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 9]: https://img-blog.csdnimg.cn/20200126165056889.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 10]: https://img-blog.csdnimg.cn/20200126165103146.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 11]: https://img-blog.csdnimg.cn/20200126165103470.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 12]: https://img-blog.csdnimg.cn/20200126165159658.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 13]: https://img-blog.csdnimg.cn/20200126165214541.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 14]: https://img-blog.csdnimg.cn/20200126165235322.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 15]: https://img-blog.csdnimg.cn/20200126165235315.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 16]: https://img-blog.csdnimg.cn/20200126165244938.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 17]: https://img-blog.csdnimg.cn/20200126165314300.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 18]: https://img-blog.csdnimg.cn/20200126165336506.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 19]: https://img-blog.csdnimg.cn/20200126165357275.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 20]: https://img-blog.csdnimg.cn/20200126165419647.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 21]: https://img-blog.csdnimg.cn/20200126165451740.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 22]: https://img-blog.csdnimg.cn/20200126165544265.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 23]: https://img-blog.csdnimg.cn/20200126165725238.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 24]: https://img-blog.csdnimg.cn/2020012616584340.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[20200126165900343.png]: https://img-blog.csdnimg.cn/20200126165900343.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 25]: https://img-blog.csdnimg.cn/20200126165931220.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 26]: https://img-blog.csdnimg.cn/20200126165953638.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 27]: https://img-blog.csdnimg.cn/20200126170003392.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 28]: https://img-blog.csdnimg.cn/20200126170009610.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 29]: https://img-blog.csdnimg.cn/20200126170035751.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 30]: https://img-blog.csdnimg.cn/20200126170044165.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0_size_16_color_FFFFFF_t_70 31]: https://img-blog.csdnimg.cn/20200126170050253.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9henVzYS5ibG9nLmNzZG4ubmV0,size_16,color_FFFFFF,t_70