【转载】读书笔记_Inception_V3_下

原文链接：https://www.cnblogs.com/hellcat/p/8058335.html

极为庞大的网络结构，不过下一节的ResNet也不小

线性的组成，结构大体如下：

常规卷积部分->Inception模块组1->Inception模块组2->Inception模块组3->池化->1*1卷积(实现个线性变换)->分类器

                                                                                |_>辅助分类器


代码如下，

# Author : Hellcat

# Time   : 2017/12/12

# refer  : https://github.com/tensorflow/models/

#          blob/master/research/inception/inception/slim/inception_model.py


import time

import math

import tensorflow as tf

from datetime import datetime


slim = tf.contrib.slim

# 截断误差初始化生成器

trunc_normal = lambda stddev:tf.truncated_normal_initializer(0.0,stddev)


def inception_v3_arg_scope(weight_decay=0.00004,

                           stddv=0.1,

                           batch_norm_var_collection='moving_vars'):

    '''

    网络常用函数默认参数生成

    :param weight_decay: L2正则化decay

    :param stddv: 标准差

    :param batch_norm_var_collection:

    :return:

    '''

    batch_norm_params = {

        'decay':0.9997, # 衰减系数

        'epsilon':0.001,

        'updates_collections':{

            'bate':None,

            'gamma':None,

            'moving_mean':[batch_norm_var_collection], # 批次均值

            'moving_variance':[batch_norm_var_collection] # 批次方差

        }

    }

    # 外层环境

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

                        # 权重正则化函数

                        weights_regularizer=slim.l2_regularizer(weight_decay)):

        # 内层环境

        with slim.arg_scope([slim.conv2d],

                            # 权重初始化函数

                            weights_initializer=tf.truncated_normal_initializer(stddev=stddv),

                            # 激活函数，默认为nn.relu

                            activation_fn=tf.nn.relu,

                            # 正则化函数，默认为None

                            normalizer_fn=slim.batch_norm,

                            # 正则化函数参数，字典形式

                            normalizer_params=batch_norm_params) as sc:

            return sc


def inception_v3_base(inputs,scope=None):

    # 保存关键节点

    end_points = {}

    # 重载作用域的名称，创建新的作用域名称(前面是None时使用)，输入tensor

    with tf.variable_scope(scope,'Inception_v3',[inputs]):

        with slim.arg_scope([slim.conv2d,slim.max_pool2d,slim.avg_pool2d],

                            stride=1,padding='VALID'):

            # 299*299*3


            net = slim.conv2d(inputs,32,[3,3],stride=2,scope='Conv2d_1a_3x3') # 149*149*32

            net = slim.conv2d(net,32,[3,3],scope='Conv2d_2a_3x3') # 147*147*32

            net = slim.conv2d(net,64,[3,3],padding='SAME',scope='Conv2d_2b_3x3') # 147*147*64

            net = slim.max_pool2d(net,[3,3],stride=2,scope='MaxPool_3a_3x3') # 73*73*64

            net = slim.conv2d(net,80,[1,1],scope='Conv2d_3b_1x1') # 73*73*80

            net = slim.conv2d(net,192,[1,1],scope='Conv2d_4a_3x3') # 71*71*192

            net = slim.max_pool2d(net,[3,3],stride=2,scope='MaxPool_5a_3x3') # 35*35*192


        with slim.arg_scope([slim.conv2d,slim.max_pool2d,slim.avg_pool2d],

                            stride=1,padding='SAME'):

            '''Inception 第一模组块'''

            # Inception_Module_1

            with tf.variable_scope('Mixed_5b'): # 35*35*256

                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_3,32,[1,1],scope='Conv2d_0b_1x1')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            # Inception_Module_2

            with tf.variable_scope('Mixed_5c'): # 35*35*288

                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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            # Inception_Module_3

            with tf.variable_scope('Mixed_5d'): # 35*35*288

                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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            '''Inception 第二模组块'''

            # Inception_Module_1

            with tf.variable_scope('Mixed_6a'): # 17*17*768

                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_3x3')

                with tf.variable_scope('Branch_2'):

                    branch_2 = slim.max_pool2d(net,[3,3],stride=2,padding='VALID',

                                               scope='Max_Pool_1a_3x3')

                net = tf.concat([branch_0,branch_1,branch_2],axis=3)


            # Inception_Module_2

            with tf.variable_scope('Mixed_6b'): # 17*17*768

                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,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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            # Inception_Module_3

            with tf.variable_scope('Mixed_6c'): # 17*17*768

                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='Conv2d_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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            # Inception_Module_4

            with tf.variable_scope('Mixed_6d'): # 17*17*768

                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='Conv2d_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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)


            # Inception_Module_5

            with tf.variable_scope('Mixed_6e'): # 17*17*768

                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')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],axis=3)

            end_points['Mixed_6e'] = net


            '''Inception 第三模组块'''

            # Inception_Module_1

            with tf.variable_scope('Mixed_7a'): # 8*8*1280

                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_1a_3x3')

                net = tf.concat([branch_0,branch_1,branch_2],3)


            # Inception_Module_2

            with tf.variable_scope('Mixed_7b'): # 8*8*2048

                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')],axis=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_0d_3x1')],axis=3)

                with tf.variable_scope('Branch_3'):

                    branch_3 = slim.max_pool2d(net,[3,3],scope='AvgPool_0a_3x3')

                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='Conv2d_0b_1x1')

                net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)


            # Inception_Module_3

            with tf.variable_scope('Mixed_7c'): # 8*8*2048

                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')],axis=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_0d_3x1')],axis=3)

                with tf.variable_scope('Branch_3'):

                    branch_3 = slim.max_pool2d(net,[3,3],scope='AvgPool_0a_3x3')

                    branch_3 = slim.conv2d(branch_3,192,[1,1],scope='Conv2d_0b_1x1')

            net = tf.concat([branch_0,branch_1,branch_2,branch_3],3)


            return net,end_points


def inception_v3(inputs,

                 num_classes=1000,

                 is_training=True,

                 dropout_keep_prob=0.8,

                 prediction_fn=slim.softmax,

                 spatial_squeeze=True,

                 reuse=None,

                 scope='Inception_v3'):

    with tf.variable_scope(scope,'Inception_v3',[inputs,num_classes],reuse=reuse) as scope:

        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'):

                # 17*17*768

                aux_logits = end_points['Mixed_6e']

                with tf.variable_scope('AuxLogits'):

                    aux_logits = slim.avg_pool2d(aux_logits,[5,5],stride=3,padding='VALID',scope='AvgPool_1a_5x5')

                    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_classes,[1,1],activation_fn=None,

                                             normalizer_fn=None,weights_initializer=trunc_normal(0.001),

                                             scope='Conv2d_2b_1x1')

                    if spatial_squeeze:

                        aux_logits = tf.squeeze(aux_logits,[1,2],

                                                name='SpatialSqueeze')

                    end_points['AuxLogits'] = aux_logits

                with tf.variable_scope('Logits'):

                    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_classes,[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

                    end_points['Predictions'] = prediction_fn(logits,scope='Predictions')

                return logits, end_points


def time_tensorflow_run(session, target, info_string):

    '''

    网路运行时间测试函数

    :param session: 会话对象

    :param target: 运行目标节点

    :param info_string:提示字符

    :return: None

    '''

    num_steps_burn_in = 10 # 预热轮数

    total_duration = 0.0 # 总时间

    total_duration_squared = 0.0 # 总时间平方和

    for i in range(num_steps_burn_in + num_batches):

        start_time = time.time()

        _ = session.run(target)

        duration = time.time() - start_time # 本轮时间

        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_squared += duration**2


    mn = total_duration/num_batches # 平均耗时

    vr = total_duration_squared/num_batches - mn**2

    sd = math.sqrt(vr)

    print('%s:%s across %d steps, %.3f +/- %.3f sec / batch' %

          (datetime.now(), info_string, num_batches, mn, sd))


if __name__ == '__main__':

    batch_size=32

    height,width = 299,299

    inputs = tf.random_uniform((batch_size,height,width,3))

    with slim.arg_scope(inception_v3_arg_scope()):

        logits,end_points = inception_v3(inputs,is_training=False)

    init = tf.global_variables_initializer()

    sess = tf.Session()

    sess.run(init)

    num_batches = 100

    time_tensorflow_run(sess,logits,'Forward')

©著作权归作者所有,转载或内容合作请联系作者
  • 人面猴
    序言：七十年代末，一起剥皮案震惊了整个滨河市，随后出现的几起案子，更是在滨河造成了极大的恐慌，老刑警刘岩，带你破解...
    沈念sama阅读 203,324评论 5赞 476
  • 死咒
    序言：滨河连续发生了三起死亡事件，死亡现场离奇诡异，居然都是意外死亡，警方通过查阅死者的电脑和手机，发现死者居然都...
    沈念sama阅读 85,303评论 2赞 381
  • 救了他两次的神仙让他今天三更去死
    文/潘晓璐 我一进店门，熙熙楼的掌柜王于贵愁眉苦脸地迎上来，“玉大人，你说我怎么就摊上这事。” “怎么了？”我有些...
    开封第一讲书人阅读 150,192评论 0赞 337
  • 道士缉凶录：失踪的卖姜人
    文/不坏的土叔 我叫张陵，是天一观的道长。 经常有香客问我，道长，这世上最难降的妖魔是什么？ 我笑而不...
    开封第一讲书人阅读 54,555评论 1赞 273
  • 港岛之恋（遗憾婚礼）
    正文 为了忘掉前任，我火速办了婚礼，结果婚礼上，老公的妹妹穿的比我还像新娘。我一直安慰自己，他们只是感情好，可当我...
    茶点故事阅读 63,569评论 5赞 365
  • 恶毒庶女顶嫁案：这布局不是一般人想出来的
    文/花漫 我一把揭开白布。 她就那样静静地躺着，像睡着了一般。 火红的嫁衣衬着肌肤如雪。 梳的纹丝不乱的头发上，一...
    开封第一讲书人阅读 48,566评论 1赞 281
  • 城市分裂传说
    那天，我揣着相机与录音，去河边找鬼。 笑死，一个胖子当着我的面吹牛，可吹牛的内容都是我干的。 我是一名探鬼主播，决...
    沈念sama阅读 37,927评论 3赞 395
  • 双鸳鸯连环套：你想象不到人心有多黑
    文/苍兰香墨 我猛地睁开眼，长吁一口气：“原来是场噩梦啊……” “哼！你这毒妇竟也来了？” 一声冷哼从身侧响起，我...
    开封第一讲书人阅读 36,583评论 0赞 257
  • 万荣杀人案实录
    序言：老挝万荣一对情侣失踪，失踪者是张志新（化名）和其女友刘颖，没想到半个月后，有当地人在树林里发现了一具尸体，经...
    沈念sama阅读 40,827评论 1赞 297
  • 护林员之死
    正文 独居荒郊野岭守林人离奇死亡，尸身上长有42处带血的脓包…… 初始之章·张勋 以下内容为张勋视角 年9月15日...
    茶点故事阅读 35,590评论 2赞 320
  • 白月光启示录
    正文 我和宋清朗相恋三年，在试婚纱的时候发现自己被绿了。 大学时的朋友给我发了我未婚夫和他白月光在一起吃饭的照片。...
    茶点故事阅读 37,669评论 1赞 329
  • 活死人
    序言：一个原本活蹦乱跳的男人离奇死亡，死状恐怖，灵堂内的尸体忽然破棺而出，到底是诈尸还是另有隐情，我是刑警宁泽，带...
    沈念sama阅读 33,365评论 4赞 318
  • 日本核电站爆炸内幕
    正文 年R本政府宣布，位于F岛的核电站，受9级特大地震影响，放射性物质发生泄漏。R本人自食恶果不足惜，却给世界环境...
    茶点故事阅读 38,941评论 3赞 307
  • 男人毒药：我在死后第九天来索命
    文/蒙蒙 一、第九天 我趴在偏房一处隐蔽的房顶上张望。 院中可真热闹，春花似锦、人声如沸。这庄子的主人今日做“春日...
    开封第一讲书人阅读 29,928评论 0赞 19
  • 一桩弑父案，背后竟有这般阴谋
    文/苍兰香墨 我抬头看了看天上的太阳。三九已至，却和暖如春，着一层夹袄步出监牢的瞬间，已是汗流浃背。 一阵脚步声响...
    开封第一讲书人阅读 31,159评论 1赞 259
  • 情欲美人皮
    我被黑心中介骗来泰国打工， 没想到刚下飞机就差点儿被人妖公主榨干…… 1. 我叫王不留，地道东北人。 一个月前我还...
    沈念sama阅读 42,880评论 2赞 349
  • 代替公主和亲
    正文 我出身青楼，却偏偏与公主长得像，于是被迫代替她去往敌国和亲。 传闻我的和亲对象是个残疾皇子，可洞房花烛夜当晚...
    茶点故事阅读 42,399评论 2赞 342

推荐阅读更多精彩内容