From e776e900580e339a9d888b1968d996e53862bf16 Mon Sep 17 00:00:00 2001 From: xiaoting <31891223+tink2123@users.noreply.github.com> Date: Fri, 3 Jan 2020 14:21:11 +0800 Subject: [PATCH] Replace conv pool (#839) * fix dead link for book * replace simple_img_conv_pool to conv_pool --- 02.recognize_digits/README.cn.md | 35 +++++++++++++++++++++++++++++-- 02.recognize_digits/README.md | 34 ++++++++++++++++++++++++++++-- 02.recognize_digits/index.cn.html | 35 +++++++++++++++++++++++++++++-- 02.recognize_digits/index.html | 34 ++++++++++++++++++++++++++++-- 4 files changed, 130 insertions(+), 8 deletions(-) diff --git a/02.recognize_digits/README.cn.md b/02.recognize_digits/README.cn.md index 33b528f..42c7bf7 100644 --- a/02.recognize_digits/README.cn.md +++ b/02.recognize_digits/README.cn.md @@ -239,6 +239,37 @@ def multilayer_perceptron(): return prediction ``` +- 卷积池化层：在LeNet-5中会出现多个卷积-池化的操作，为避免代码重复书写，将串联的卷积-池化写成conv_pool函数。 + +```python +def conv_pool(input, num_filters, filter_size, pool_size, pool_stride, act="relu"): + """ + 定义卷积池化层： + 含有一个卷积层和一个池化层 + Args： + input —— 网络输入 + num_filters —— 卷积核的个数 + filter_size —— 卷积核的大小 + pool_size —— 池化核的大小 + pool_stride —— 池化的步长 + act —— 卷积层的激活函数 + + Return: + out -- 经过卷积池化后的特征图 + """ + conv_out = fluid.layers.conv2d( + input=input, + num_filters=num_filters, + filter_size=filter_size, + act=act) + out = fluid.layers.pool2d( + input=conv_out, + pool_size=pool_size, + pool_stride=pool_stride) + return out +``` + + - 卷积神经网络LeNet-5: 输入的二维图像，首先经过两次卷积层到池化层，再经过全连接层，最后使用以softmax为激活函数的全连接层作为输出层。 ```python @@ -254,7 +285,7 @@ def convolutional_neural_network(): img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') # 第一个卷积-池化层 # 使用20个5*5的滤波器，池化大小为2，池化步长为2，激活函数为Relu - conv_pool_1 = fluid.nets.simple_img_conv_pool( + conv_pool_1 = conv_pool( input=img, filter_size=5, num_filters=20, @@ -264,7 +295,7 @@ def convolutional_neural_network(): conv_pool_1 = fluid.layers.batch_norm(conv_pool_1) # 第二个卷积-池化层 # 使用50个5*5的滤波器，池化大小为2，池化步长为2，激活函数为Relu - conv_pool_2 = fluid.nets.simple_img_conv_pool( + conv_pool_2 = conv_pool( input=conv_pool_1, filter_size=5, num_filters=50, diff --git a/02.recognize_digits/README.md b/02.recognize_digits/README.md index b0832f5..a03b6d4 100644 --- a/02.recognize_digits/README.md +++ b/02.recognize_digits/README.md @@ -218,6 +218,36 @@ def multilayer_perceptron(): return prediction ``` +-Conv_pool layer: LeNet-5 has multiple convolution-pooling operations. In order to avoid repeated code writing, the convolution-pooling in series is written as conv_pool function. + +```python +def conv_pool(input, num_filters, filter_size, pool_size, pool_stride, act="relu"): + """ + Define convolution-pooling layer： + Conv_pool layer has a convolutional layer and a pooling layer + Args： + input —— Input + num_filters —— The number of filter + filter_size —— The filter size + pool_size —— The pool kernel size + pool_stride —— The pool stride size + act —— Activation type + + Return: + out -- output + """ + conv_out = fluid.layers.conv2d( + input=input, + num_filters=num_filters, + filter_size=filter_size, + act=act) + out = fluid.layers.pool2d( + input=conv_out, + pool_size=pool_size, + pool_stride=pool_stride) + return out + +``` -Convolutional neural network LeNet-5: The input two-dimensional image first passes through two convolutional layers to the pooling layer, then passes through the fully connected layer, and finally fully connection layer with softmax as activation function is used as output layer. ```python @@ -233,7 +263,7 @@ def convolutional_neural_network(): img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') # the first convolution-pooling layer # Use 20 5*5 filters, the pooling size is 2, the pooling step is 2, and the activation function is Relu. - conv_pool_1 = fluid.nets.simple_img_conv_pool( + conv_pool_1 = conv_pool( input=img, filter_size=5, num_filters=20, @@ -243,7 +273,7 @@ def convolutional_neural_network(): conv_pool_1 = fluid.layers.batch_norm(conv_pool_1) # the second convolution-pooling layer # Use 20 5*5 filters, the pooling size is 2, the pooling step is 2, and the activation function is Relu. - conv_pool_2 = fluid.nets.simple_img_conv_pool( + conv_pool_2 = conv_pool( input=conv_pool_1, filter_size=5, num_filters=50, diff --git a/02.recognize_digits/index.cn.html b/02.recognize_digits/index.cn.html index 5733f2d..655d7ce 100644 --- a/02.recognize_digits/index.cn.html +++ b/02.recognize_digits/index.cn.html @@ -281,6 +281,37 @@ def multilayer_perceptron(): return prediction ``` +- 卷积池化层：在LeNet-5中会出现多个卷积-池化的操作，为避免代码重复书写，将串联的卷积-池化写成conv_pool函数。 + +```python +def conv_pool(input, num_filters, filter_size, pool_size, pool_stride, act="relu"): + """ + 定义卷积池化层： + 含有一个卷积层和一个池化层 + Args： + input —— 网络输入 + num_filters —— 卷积核的个数 + filter_size —— 卷积核的大小 + pool_size —— 池化核的大小 + pool_stride —— 池化的步长 + act —— 卷积层的激活函数 + + Return: + out -- 经过卷积池化后的特征图 + """ + conv_out = fluid.layers.conv2d( + input=input, + num_filters=num_filters, + filter_size=filter_size, + act=act) + out = fluid.layers.pool2d( + input=conv_out, + pool_size=pool_size, + pool_stride=pool_stride) + return out +``` + + - 卷积神经网络LeNet-5: 输入的二维图像，首先经过两次卷积层到池化层，再经过全连接层，最后使用以softmax为激活函数的全连接层作为输出层。 ```python @@ -296,7 +327,7 @@ def convolutional_neural_network(): img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') # 第一个卷积-池化层 # 使用20个5*5的滤波器，池化大小为2，池化步长为2，激活函数为Relu - conv_pool_1 = fluid.nets.simple_img_conv_pool( + conv_pool_1 = conv_pool( input=img, filter_size=5, num_filters=20, @@ -306,7 +337,7 @@ def convolutional_neural_network(): conv_pool_1 = fluid.layers.batch_norm(conv_pool_1) # 第二个卷积-池化层 # 使用50个5*5的滤波器，池化大小为2，池化步长为2，激活函数为Relu - conv_pool_2 = fluid.nets.simple_img_conv_pool( + conv_pool_2 = conv_pool( input=conv_pool_1, filter_size=5, num_filters=50, diff --git a/02.recognize_digits/index.html b/02.recognize_digits/index.html index 44c80a1..72f9382 100644 --- a/02.recognize_digits/index.html +++ b/02.recognize_digits/index.html @@ -260,6 +260,36 @@ def multilayer_perceptron(): return prediction ``` +-Conv_pool layer: LeNet-5 has multiple convolution-pooling operations. In order to avoid repeated code writing, the convolution-pooling in series is written as conv_pool function. + +```python +def conv_pool(input, num_filters, filter_size, pool_size, pool_stride, act="relu"): + """ + Define convolution-pooling layer： + Conv_pool layer has a convolutional layer and a pooling layer + Args： + input —— Input + num_filters —— The number of filter + filter_size —— The filter size + pool_size —— The pool kernel size + pool_stride —— The pool stride size + act —— Activation type + + Return: + out -- output + """ + conv_out = fluid.layers.conv2d( + input=input, + num_filters=num_filters, + filter_size=filter_size, + act=act) + out = fluid.layers.pool2d( + input=conv_out, + pool_size=pool_size, + pool_stride=pool_stride) + return out + +``` -Convolutional neural network LeNet-5: The input two-dimensional image first passes through two convolutional layers to the pooling layer, then passes through the fully connected layer, and finally fully connection layer with softmax as activation function is used as output layer. ```python @@ -275,7 +305,7 @@ def convolutional_neural_network(): img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') # the first convolution-pooling layer # Use 20 5*5 filters, the pooling size is 2, the pooling step is 2, and the activation function is Relu. - conv_pool_1 = fluid.nets.simple_img_conv_pool( + conv_pool_1 = conv_pool( input=img, filter_size=5, num_filters=20, @@ -285,7 +315,7 @@ def convolutional_neural_network(): conv_pool_1 = fluid.layers.batch_norm(conv_pool_1) # the second convolution-pooling layer # Use 20 5*5 filters, the pooling size is 2, the pooling step is 2, and the activation function is Relu. - conv_pool_2 = fluid.nets.simple_img_conv_pool( + conv_pool_2 = conv_pool( input=conv_pool_1, filter_size=5, num_filters=50, -- GitLab