Tensorflow实现GoogleInceptionNet_V3（小白入门）

Tensorflow实现GoogleInceptionNet_V3

引言:Google Inception Net首次出现在ILSVRC 2014的比赛中，并且有非常好的性能——top-5的错误率6.67%，后来又逐渐提出了Inception V2、Inception V3和Incepiton V4。本文实现的网络结构为Inception V3

---

• 相关阅读：
  Tensorflow实现VGGNet
  Tensorflow实现AlexNet

• 模块化方法：
  本文将通过模块化加注释的方法，来实现GoogleInceptionNet，这样可以帮助初学者快速读懂其结果以及实现方法

1. 定义网络中经常用的函数的默认参数：inception_v3_arg_scope()
2. 定义Inception v3网络结构（前面部分）：inception_v3_base()
3. 定义Inception V3网络的最后一部分：incepiton_v3()
4. 定义耗时：time_tensorflow_run()

• GoogleInceptionNet结构：

• GoogleInceptionNet创新之处：

1. 去除了最后的全连接层，用全局平均池化层（即将图片尺寸变为1x1）来取代他
2. 根据Hebbian原理，多次使用1x1卷积核（即将相关性很高、在同一个空间位置但是不同通道的特征连接起来）
3. 使用了Inception Module提高了参数利用效率（上截图Fig5-Fig7）
4. 提出了著名的Batch Normalization（BN）方法，该方法可以让大型卷积网络的训练速度加快很多倍，同时收敛后的分类内部准确率也可以得到大幅提高
5. 引入Factorization into small convolutions的思想，将一个较大的2维卷积拆成两个较小的一维卷积，如把3x3卷积拆成1x3和3x1两个小卷积，不仅可以加速运算还可以减轻过拟合

---

导入包：

from datetime import datetime
import math
import time
import tensorflow as tf
slim = tf.contrib.slim
trunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)  # 产生截断的正态分布

生成网络中经常用的函数的默认参数：

# 生成网络中经常用的函数的默认参数
def inception_v3_arg_scope(weight_decay = 4e-5, 
                           stddev = 0.1, 
                           batch_norm_var_collection = 'moving_vars'):
    '''
    weight_decay: L2正则weight_decay
    stddev: 标准差
    batch_norm_var_collection: 默认设置为moving_vars
    '''
    # 定义batch normalization的参数字典
    batch_norm_params = {
      'decay': 0.9997,
      'epsilon': 0.001,
      'updates_collections': tf.GraphKeys.UPDATE_OPS,
      'variables_collections': {
          'beta': None,
          'gamma': None,
          'moving_mean': [batch_norm_var_collection],
          'moving_variance': [batch_norm_var_collection],
      }
    }
    # 使用slim.arg_scope对参数进行自动复赋值
    with slim.arg_scope([slim.conv2d, slim.fully_connected],
                      weights_regularizer=slim.l2_regularizer(weight_decay)):
        with slim.arg_scope(
            [slim.conv2d],
            weights_initializer=trunc_normal(stddev),
            activation_fn=tf.nn.relu,
            normalizer_fn=slim.batch_norm,
            normalizer_params=batch_norm_params) as sc:
            return sc

定义Inception v3网络结构（前面部分）：

# 定义Inception v3网络结构
def inception_v3_base(inputs, scope = None):
    end_points = {}
    with tf.variable_scope(scope, 'InceptionV3', [inputs]):
        '''
        前面的卷积层结构 类似于VGGNet
        input:299x299x3
        output:35x35x192
        ''' 
        with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d], 
                            stride = 1, padding  = 'VALID'):
            net = slim.conv2d(inputs, 32, [3, 3], stride = 2, scope = 'Conv2d_1a_3x3')
            net = slim.conv2d(net, 32, [3, 3], scope = 'Conv2d_2a_3x3')
            net = slim.conv2d(net, 64, [3, 3], padding = 'SAME', scope = 'Conv2d_2b_3x3')
            net = slim.max_pool2d(net, [3, 3], stride = 2, scope = 'MaxPool_3a_3x3')
            net = slim.conv2d(net, 80, [1, 1], scope = 'Conv2d_3b_1x1')
            net = slim.conv2d(net, 192, [3, 3], scope = 'Conv2d_4a_3x3')
            net = slim.max_pool2d(net, [3, 3], stride = 2, scope = 'MaxPool_5a_3x3')

        with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],
                            stride = 1, padding = 'SAME'):        
            '''
            第一个Inception模块组：第一个Module
            input:35x35x192
            output:35x35x256
            '''
            with tf.variable_scope('Mixed_5b'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 48, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 64, [5, 5], scope = 'Conv2d_0a_5x5')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope = 'Conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope = 'Conv2d_0c_3x3')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 32, [1, 1], scope = 'Conv2d_0b_3x3')
                # 将四个分支在channel上连起来 64 + 64 + 96 + 32 = 256
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第一个Inception模块组：第二个Module
            input:35x35x256
            output:35x35x288
            '''
            with tf.variable_scope('Mixed_5c'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 48, [1, 1], scope = 'Conv2d_0b_1x1')
                    branch_1 = slim.conv2d(branch_1, 64, [5, 5], scope = 'Conv_1_0c_5x5')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope = 'Conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope = 'Conv2d_0c_3x3')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope = 'Conv2d_0b_1x1')
                # 64 + 64 + 96 + 64 = 288
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第一个Inception模块组：第三个Module(与第二个结构相同)
            input:35x35x288
            output:35x35x288
            '''
            with tf.variable_scope('Mixed_5d'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 64, [1, 1], scope='Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 48, [1, 1], scope='Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 64, [5, 5], scope='Conv2d_0b_5x5')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 64, [1, 1], scope='Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0b_3x3')
                    branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0c_3x3')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
                # 64 + 64 + 96 + 64 = 288
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)   
                
            '''
            第二个Inception模块组：第一个Module
            input:35x35x288
            output:17x17x768
            '''
            with tf.variable_scope('Mixed_6a'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 384, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 96, [3, 3], scope = 'Conv2d_0b_3x3')
                    branch_1 = slim.conv2d(branch_1, 96, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_1x1')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.max_pool2d(net, [3, 3], stride = 2, padding = 'VALID', scope = 'MaxPool_1a_3x3')
                # 384 + 96 + 288 = 768
                net = tf.concat([branch_0, branch_1, branch_2], 3)
            '''
            第二个Inception模块组：第二个Module
            input:17x17x768
            output:17x17x768
            '''
            with tf.variable_scope('Mixed_6b'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    # Factorization into small convolutions思想，减少参数的同时也减轻了过拟合
                    branch_1 = slim.conv2d(net, 128, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 128, [1, 7], scope = 'Conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_0c_7x1')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 128, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 128, [7, 1], scope = 'Conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2, 128, [1, 7], scope = 'Conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2, 128, [7, 1], scope = 'Conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope = 'Conv2d_0e_1x7')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 192 + 192 + 192 + 192 = 768
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第二个Inception模块组：第三个Module
            input:17x17x768
            output:17x17x768
            '''
            with tf.variable_scope('Mixed_6c'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 160, [1, 7], scope = 'Conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_0c_7x1')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope = 'Conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2, 160, [1, 7], scope = 'Conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope = 'Conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope = 'Conv2d_0e_1x7')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 192 + 192 + 192 + 192 = 768
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第二个Inception模块组：第四个Module(与第三个结构相同)
            input:17x17x768
            output:17x17x768
            '''
            with tf.variable_scope('Mixed_6d'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 160, [1, 7], scope = 'Conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_0c_7x1')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope = 'Conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2, 160, [1, 7], scope = 'Conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope = 'Conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope = 'Conv2d_0e_1x7')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 192 + 192 + 192 + 192 = 768
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第二个Inception模块组：第四个Module(与第三个结构相同)
            input:17x17x768
            output:17x17x768
            '''
            with tf.variable_scope('Mixed_6e'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 192, [1, 7], scope = 'Conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_0c_7x1')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 192, [7, 1], scope = 'Conv2d_0b_7x1')
                    branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope = 'Conv2d_0c_1x7')
                    branch_2 = slim.conv2d(branch_2, 192, [7, 1], scope = 'Conv2d_0d_7x1')
                    branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope = 'Conv2d_0e_1x7')
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 192 + 192 + 192 + 192 = 768
                net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            end_points['Mixed_6e'] = net

            '''
            第三个Inception模块组：第一个Module
            input:17x17x768
            output:8x8x1280
            '''
            with tf.variable_scope('Mixed_7a'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_0 = slim.conv2d(branch_0, 320, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_3x3')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = slim.conv2d(branch_1, 192, [1, 7], scope = 'Conv2d_0b_1x7')
                    branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_0c_7x1')
                    branch_1 = slim.conv2d(branch_1, 192, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_3x3')
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.max_pool2d(net, [3, 3], stride = 2, padding = 'VALID', scope = 'MaxPool_la_3x3')
                # 320 + 192 + 768 = 1280
                net = tf.concat([branch_0, branch_1, branch_2], 3)
            '''
            第三个Inception模块组：第二个Module
            input:8x8x1280
            output:8x8x2048
            '''
            with tf.variable_scope('Mixed_7b'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 320, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 384, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = tf.concat([
                        slim.conv2d(branch_1, 384, [1, 3], scope = 'Conv2d_0b_1x3'),
                        slim.conv2d(branch_1, 384, [3, 1], scope = 'Conv2d_0b_3x1')], 3)
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 448, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 384, [3, 3], scope = 'Conv2d_0b_3x3')
                    branch_2 = tf.concat([
                        slim.conv2d(branch_2, 384, [1, 3], scope = 'Conv2d_0c_1x3'),
                        slim.conv2d(branch_2, 384, [3, 1], scope = 'Conv2d_0c_3x1')], 3)
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 320 + 384*2 + 384*2 + 192 = 2048
                tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            '''
            第三个Inception模块组：第三个Module（与第二个完全一样）
            input:8x8x2048
            output:8x8x2048
            '''
            with tf.variable_scope('Mixed_7c'):
                with tf.variable_scope('Branch_0'):
                    branch_0 = slim.conv2d(net, 320, [1, 1], scope = 'Conv2d_0a_1x1')
                with tf.variable_scope('Branch_1'):
                    branch_1 = slim.conv2d(net, 384, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_1 = tf.concat([
                        slim.conv2d(branch_1, 384, [1, 3], scope = 'Conv2d_0b_1x3'),
                        slim.conv2d(branch_1, 384, [3, 1], scope = 'Conv2d_0b_3x1')], 3)
                with tf.variable_scope('Branch_2'):
                    branch_2 = slim.conv2d(net, 448, [1, 1], scope = 'Conv2d_0a_1x1')
                    branch_2 = slim.conv2d(branch_2, 384, [3, 3], scope = 'Conv2d_0b_3x3')
                    branch_2 = tf.concat([
                        slim.conv2d(branch_2, 384, [1, 3], scope = 'Conv2d_0c_1x3'),
                        slim.conv2d(branch_2, 384, [3, 1], scope = 'Conv2d_0c_3x1')], 3)
                with tf.variable_scope('Branch_3'):
                    branch_3 = slim.avg_pool2d(net, [3, 3], scope = 'AvgPool_0a_3x3')
                    branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope = 'Conv2d_0b_1x1')
                # 320 + 384*2 + 384*2 + 192 = 2048
                tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
            return net, end_points

定义Inception V3网络的最后一部分：

# Inception V3网络的最后一部分
def incepiton_v3(inputs,                              
                 num_class = 1000,                    
                 is_training = True,                  
                 dropout_keep_prob = 0.8,              
                 prediction_fn = slim.softmax,
                 spatial_squeeze = True,
                 reuse = None,
                 scope = 'InceptionV3'):
    '''
    inputs: 输入
    num_class: 最后需要分类的总数
    is_training: 标志是否是训练过程，对BN和Dropout有影响
    dropout_keep_prob: 保留节点的比例
    prediction_fn: 用来分类的函数
    spatial_squeeze: 是否要将型如5x3x1--->5x3
    reuse: 标志是否会对网络和Variable进行重复使用
    scope: 包含函数默认参数的环境
    '''
    with tf.variable_scope(scope, 'InceptionV3', [inputs, num_class], reuse = reuse) as scope:
        # 使用slim.arg_scope进行默认初始化设置
        with slim.arg_scope([slim.batch_norm, slim.dropout], is_training = is_training):
            net, end_points = inception_v3_base(inputs, scope = scope)
        with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d], stride = 1, padding = 'SAME'):
            aux_logits = end_points['Mixed_6e']  # 辅助分类结点
            with tf.variable_scope('AuxLogits'):
                # 进行池化
                aux_logits = slim.avg_pool2d(aux_logits, [5, 5], stride = 3, padding = 'VALID', scope = 'AvgPool_1a_5x5')
                # 进行卷积
                aux_logits = slim.conv2d(aux_logits, 128, [1, 1], scope = 'Conv2d_1b_1x1')
                aux_logits = slim.conv2d(aux_logits, 768, [5, 5], 
                                         weights_initializer = trunc_normal(0.01), 
                                         padding = 'VALID', scope = 'Conv2d_2a_5x5')
                aux_logits = slim.conv2d(aux_logits, num_class, [1, 1],
                                         activation_fn = None, 
                                         weights_initializer = trunc_normal(0.001),
                                         scope = 'Conv2d_2b_1x1')
                if spatial_squeeze:
                    # [1, 2]表示把第2,3维出现的1去掉（前提得2,3本就存在1）
                    aux_logits = tf.squeeze(aux_logits, [1, 2], name = 'SpatialSqueeze')
                end_points['AuxLogits'] = aux_logits
            with tf.variable_scope('Logits'):
                # 直接对Mixed_7e即最后一个卷积层的输出进行操作
                net = slim.avg_pool2d(net, [8, 8], padding = 'VALID', scope = 'AvgPool_1a_8x8')
                net = slim.dropout(net, keep_prob = dropout_keep_prob, scope = 'Dropout_1b')
                end_points['PreLogits'] = net
                logits = slim.conv2d(net, num_class, [1, 1], activation_fn = None,
                                     normalizer_fn = None, scope = 'Conv2d_1c_1x1')
                if spatial_squeeze:
                    logits = tf.squeeze(logits, [1, 2], name = 'SpatialSqueeze')
                end_points['Logits'] = logits
                # 使用softmax函数预测
                end_points['Predictions'] = prediction_fn(logits, scope = 'Predictions')
            return logits, end_points

定义耗时：

# 计算耗时
def time_tensorflow_run(session, target, info_string):
    num_steps_burn_in = 10  # 打印阈值
    total_duration = 0.0    # 每一轮所需要的迭代时间
    total_duration_aquared = 0.0  # 每一轮所需要的迭代时间的平方
    for i in range(num_batches + num_steps_burn_in):
        start_time = time.time()
        _ = session.run(target)
        duration = time.time() - start_time    # 计算耗时
        if i >= num_steps_burn_in:
            if not i % 10:
                print("%s : step %d, duration = %.3f" % (datetime.now(), i - num_steps_burn_in, duration))
            total_duration += duration
            total_duration_aquared += duration * duration
    mn = total_duration / num_batches   # 计算均值
    vr = total_duration_aquared / num_batches - mn * mn  # 计算方差
    sd = math.sqrt(vr) # 计算标准差
    print("%s : %s across %d steps, %.3f +/- %.3f sec/batch" % (datetime.now(), info_string, num_batches, mn, sd))

---

测试：
注：CNN的训练都是比较耗时的，所以这里也就是测试一下几张随机生成图的前向和反向传播过程

batch_size = 32
height, width = 299, 299
inputs = tf.random_uniform((batch_size, height, width, 3))
# 加载前面定义好的inception_v3_arg_scope()
with slim.arg_scope(inception_v3_arg_scope()):
    logits, end_points = incepiton_v3(inputs, is_training = False)

with tf.Session() as sess:
    init = tf.global_variables_initializer()
    sess.run(init)
    num_batches = 100
    time_tensorflow_run(sess, logits, "Forward")

运行效果：

2019-01-22 13:02:51.088969 : step 0, duration = 0.231
2019-01-22 13:02:53.406683 : step 10, duration = 0.231
2019-01-22 13:02:55.725737 : step 20, duration = 0.233
2019-01-22 13:02:58.050072 : step 30, duration = 0.231
2019-01-22 13:03:00.372407 : step 40, duration = 0.233
2019-01-22 13:03:02.694779 : step 50, duration = 0.234
2019-01-22 13:03:05.019621 : step 60, duration = 0.232
2019-01-22 13:03:07.343221 : step 70, duration = 0.234
2019-01-22 13:03:09.665730 : step 80, duration = 0.232
2019-01-22 13:03:11.990585 : step 90, duration = 0.233
2019-01-22 13:03:14.090070 : Forward across 100 steps, 0.232 +/- 0.001 sec/batch

---

[1] Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 2818-2826.
[2] Tensorflo实战.黄文坚，唐源

如果觉得我有地方讲的不好的或者有错误的欢迎给我留言，如果对您有帮助，帮我点个赞哦~，感谢大家阅读