使用笔记:TF辅助工具--tensorflow slim(TF-Slim)

　　如果抛开Keras，TensorLayer，tfLearn，tensroflow 能否写出简介的代码？ 可以！slim这个模块是在16年新推出的，其主要目的是来做所谓的“代码瘦身”

一.简介

　　slim被放在tensorflow.contrib这个库下面，导入的方法如下：

　　import tensorflow.contrib.slim as slim

　　众所周知 tensorflow.contrib这个库，tensorflow官方对它的描述是：此目录中的任何代码未经官方支持，可能会随时更改或删除。每个目录下都有指定的所有者。它旨在包含额外功能和贡献，最终会合并到核心TensorFlow中，但其接口可能仍然会发生变化，或者需要进行一些测试，看是否可以获得更广泛的接受。所以slim依然不属于原生tensorflow。

　　slim是一个使构建，训练，评估神经网络变得简单的库。它可以消除原生tensorflow里面很多重复的模板性的代码，让代码更紧凑，更具备可读性。另外slim提供了很多计算机视觉方面的著名模型（VGG, AlexNet等），我们不仅可以直接使用，甚至能以各种方式进行扩展。

　　slim的子模块及功能介绍：

　　arg_scope: provides a new scope named arg_scope that allows a user to define default arguments for specific operations within that scope.

　　除了基本的namescope，variabelscope外，又加了argscope，它是用来控制每一层的默认超参数的。（后面会详细说）

　　data: contains TF-slim's dataset definition, data providers, parallel_reader, and decoding utilities.

　　貌似slim里面还有一套自己的数据定义，这个跳过，我们用的不多。

　　evaluation: contains routines for evaluating models.

　　评估模型的一些方法，用的也不多

　　layers: contains high level layers for building models using tensorflow.

　　这个比较重要，slim的核心和精髓，一些复杂层的定义

　　learning: contains routines for training models.

　　一些训练规则

　　losses: contains commonly used loss functions.

　　一些loss

　　metrics: contains popular evaluation metrics.

　　评估模型的度量标准

　　nets: contains popular network definitions such as VGG and AlexNet models.

　　包含一些经典网络，VGG等，用的也比较多

　　queues: provides a context manager for easily and safely starting and closing QueueRunners.

　　文本队列管理，比较有用。

　　regularizers: contains weight regularizers.

　　包含一些正则规则

　　variables: provides convenience wrappers for variable creation and manipulation.

　　slim管理变量的机制

二.slim定义模型

slim中定义一个变量的示例：

　　# Model Variables

weights = slim.model_variable('weights',

                              shape=[10, 10, 3 , 3],

                              initializer=tf.truncated_normal_initializer(stddev=0.1),

                              regularizer=slim.l2_regularizer(0.05),

                              device='/CPU:0')

model_variables = slim.get_model_variables()

 

# Regular variables

my_var = slim.variable('my_var',

                       shape=[20, 1],

                       initializer=tf.zeros_initializer())

regular_variables_and_model_variables = slim.get_variables()

 

　　如上，变量分为两类：模型变量和局部变量。局部变量是不作为模型参数保存的，而模型变量会再save的时候保存下来。这个玩过tensorflow的人都会明白，诸如global_step之类的就是局部变量。slim中可以写明变量存放的设备，正则和初始化规则。还有获取变量的函数也需要注意一下，get_variables是返回所有的变量。

　　slim中实现一个层：

　　首先让我们看看tensorflow怎么实现一个层，例如卷积层：

input = ...

with tf.name_scope('conv1_1') as scope:

kernel = tf.Variable(tf.truncated_normal([3, 3, 64, 128], dtype=tf.float32,

                                           stddev=1e-1), name='weights'

conv = tf.nn.conv2d(input, kernel, [1, 1, 1, 1], padding='SAME')

biases = tf.Variable(tf.constant(0.0, shape=[128], dtype=tf.float32),

                       trainable=True, name='biases')

bias = tf.nn.bias_add(conv, biases)

conv1 = tf.nn.relu(bias, name=scope)

然后slim的实现：

input = ...

net = slim.conv2d(input, 128, [3, 3], scope='conv1_1')

但这个不是重要的，因为tenorflow目前也有大部分层的简单实现，这里比较吸引人的是slim中的repeat和stack操作：

net = ...

net = slim.conv2d(net, 256, [3, 3], scope='conv3_1')

net = slim.conv2d(net, 256, [3, 3], scope='conv3_2')

net = slim.conv2d(net, 256, [3, 3], scope='conv3_3')

net = slim.max_pool2d(net, [2, 2], scope='pool2')

 

在slim中的repeat操作可以减少代码量：

net = slim.repeat(net, 3, slim.conv2d, 256, [3, 3], scope='conv3')

net = slim.max_pool2d(net, [2, 2], scope='pool2')

 

而stack是处理卷积核或者输出不一样的情况：

假设定义三层FC：

# Verbose way:

x = slim.fully_connected(x, 32, scope='fc/fc_1')

x = slim.fully_connected(x, 64, scope='fc/fc_2')

x = slim.fully_connected(x, 128, scope='fc/fc_3')

使用stack操作：

slim.stack(x, slim.fully_connected, [32, 64, 128], scope='fc')

同理卷积层也一样：

# 普通方法:

x = slim.conv2d(x, 32, [3, 3], scope='core/core_1')

x = slim.conv2d(x, 32, [1, 1], scope='core/core_2')

x = slim.conv2d(x, 64, [3, 3], scope='core/core_3')

x = slim.conv2d(x, 64, [1, 1], scope='core/core_4')

 

# 简便方法:

slim.stack(x, slim.conv2d, [(32, [3, 3]), (32, [1, 1]), (64, [3, 3]), (64, [1, 1])], scope='core')

slim中的argscope：

如果你的网络有大量相同的参数，如下：

net = slim.conv2d(inputs, 64, [11, 11], 4, padding='SAME',

                  weights_initializer=tf.truncated_normal_initializer(stddev=0.01),

                  weights_regularizer=slim.l2_regularizer(0.0005), scope='conv1')

net = slim.conv2d(net, 128, [11, 11], padding='VALID',

                  weights_initializer=tf.truncated_normal_initializer(stddev=0.01),

                  weights_regularizer=slim.l2_regularizer(0.0005), scope='conv2')

net = slim.conv2d(net, 256, [11, 11], padding='SAME',

                  weights_initializer=tf.truncated_normal_initializer(stddev=0.01),

                  weights_regularizer=slim.l2_regularizer(0.0005), scope='conv3')

 

然后我们用arg_scope处理一下：

with slim.arg_scope([slim.conv2d], padding='SAME',

                      weights_initializer=tf.truncated_normal_initializer(stddev=0.01)

                      weights_regularizer=slim.l2_regularizer(0.0005)):

net = slim.conv2d(inputs, 64, [11, 11], scope='conv1')

net = slim.conv2d(net, 128, [11, 11], padding='VALID', scope='conv2')

net = slim.conv2d(net, 256, [11, 11], scope='conv3')

这里额外说明一点，arg_scope的作用范围内，是定义了指定层的默认参数，若想特别指定某些层的参数，可以重新赋值（相当于重写），如上倒数第二行代码。

那如果除了卷积层还有其他层呢？那就要如下定义：

with slim.arg_scope([slim.conv2d, slim.fully_connected],

                      activation_fn=tf.nn.relu,

                      weights_initializer=tf.truncated_normal_initializer(stddev=0.01),

                      weights_regularizer=slim.l2_regularizer(0.0005)):

　　with slim.arg_scope([slim.conv2d], stride=1, padding='SAME'):

　　　　net = slim.conv2d(inputs, 64, [11, 11], 4, padding='VALID', scope='conv1')

　　　　net = slim.conv2d(net, 256, [5, 5],

          　　　　            weights_initializer=tf.truncated_normal_initializer(stddev=0.03),

          　　　　            scope='conv2')

　　　　net = slim.fully_connected(net, 1000, activation_fn=None, scope='fc')

VGG：

def vgg16(inputs):

  with slim.arg_scope([slim.conv2d, slim.fully_connected],

                      activation_fn=tf.nn.relu,

                      weights_initializer=tf.truncated_normal_initializer(0.0, 0.01),

                      weights_regularizer=slim.l2_regularizer(0.0005)):

    net = slim.repeat(inputs, 2, slim.conv2d, 64, [3, 3], scope='conv1')

    net = slim.max_pool2d(net, [2, 2], scope='pool1')

    net = slim.repeat(net, 2, slim.conv2d, 128, [3, 3], scope='conv2')

    net = slim.max_pool2d(net, [2, 2], scope='pool2')

    net = slim.repeat(net, 3, slim.conv2d, 256, [3, 3], scope='conv3')

    net = slim.max_pool2d(net, [2, 2], scope='pool3')

    net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv4')

    net = slim.max_pool2d(net, [2, 2], scope='pool4')

    net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv5')

    net = slim.max_pool2d(net, [2, 2], scope='pool5')

    net = slim.fully_connected(net, 4096, scope='fc6')

    net = slim.dropout(net, 0.5, scope='dropout6')

    net = slim.fully_connected(net, 4096, scope='fc7')

    net = slim.dropout(net, 0.5, scope='dropout7')

    net = slim.fully_connected(net, 1000, activation_fn=None, scope='fc8')

  return net

三.训练模型

import tensorflow as tf

vgg = tf.contrib.slim.nets.vgg

 

# Load the images and labels.

images, labels = ...

 

# Create the model.

predictions, _ = vgg.vgg_16(images)

 

# Define the loss functions and get the total loss.

loss = slim.losses.softmax_cross_entropy(predictions, labels)

　　

关于loss,要说一下定义自己的loss的方法，以及注意不要忘记加入到slim中让slim看到你的loss。

还有正则项也是需要手动添加进loss当中的，不然最后计算的时候就不优化正则目标了。

# Load the images and labels.

images, scene_labels, depth_labels, pose_labels = ...

 

# Create the model.

scene_predictions, depth_predictions, pose_predictions = CreateMultiTaskModel(images)

 

# Define the loss functions and get the total loss.

classification_loss = slim.losses.softmax_cross_entropy(scene_predictions, scene_labels)

sum_of_squares_loss = slim.losses.sum_of_squares(depth_predictions, depth_labels)

pose_loss = MyCustomLossFunction(pose_predictions, pose_labels)

slim.losses.add_loss(pose_loss) # Letting TF-Slim know about the additional loss.

 

# The following two ways to compute the total loss are equivalent:

regularization_loss = tf.add_n(slim.losses.get_regularization_losses())

total_loss1 = classification_loss + sum_of_squares_loss + pose_loss + regularization_loss

 

# (Regularization Loss is included in the total loss by default).

total_loss2 = slim.losses.get_total_loss()

四.读取保存模型变量

通过以下功能我们可以载入模型的部分变量：

# Create some variables.

v1 = slim.variable(name="v1", ...)

v2 = slim.variable(name="nested/v2", ...)

...

 

# Get list of variables to restore (which contains only 'v2').

variables_to_restore = slim.get_variables_by_name("v2")

 

# Create the saver which will be used to restore the variables.

restorer = tf.train.Saver(variables_to_restore)

 

with tf.Session() as sess:

  # Restore variables from disk.

  restorer.restore(sess, "/tmp/model.ckpt")

  print("Model restored.")

 

除了这种部分变量加载的方法外，我们甚至还能加载到不同名字的变量中。

假设我们定义的网络变量是conv1/weights，而从VGG加载的变量名为vgg16/conv1/weights，正常load肯定会报错（找不到变量名），但是可以这样：

def name_in_checkpoint(var):

  return 'vgg16/' + var.op.name

 

variables_to_restore = slim.get_model_variables()

variables_to_restore = {name_in_checkpoint(var):var for var in variables_to_restore}

restorer = tf.train.Saver(variables_to_restore)

 

with tf.Session() as sess:

  # Restore variables from disk.

  restorer.restore(sess, "/tmp/model.ckpt")

　　　　通过这种方式我们可以加载不同变量名的变量