论文地址：Going Deeper with Convolutions
GoogleLeNet最大的特点是控制了计算量和参数量的同时，获得了很好的分类效果--top-5错误率为6.67%。Inception V1控制算量和参数量的方法：1）模型层数更深，表达能力更强（V1 ： 22层， VGG-19 ：19层）；2）使用1x1卷积核（全局平均池化层）替代全连接层。全连接层几乎占据了VGG 90%的参数量，而且会过拟合，去除全连接层能加快训练速度且避免过拟合；3）采用NIN（netWork in netWork）思想，使用小网络反复堆叠构建更大网络， V1在此基础之上引入分支网络，能够产生更丰富的非线性特征。
基于Hebbian原理的Inception model：
(Hebbian原理简言之：神经反射的活动的持续与重复会导致神经元连接稳定性的持久提升，当两个神经元A和B距离很近，并且A参与了对B重复、持续的兴奋，某次代谢变化导致将A作为能使B兴奋的细胞，即：一起发射的神经元会连在一起)
GoogleLeNet inception V1结构：
GoogleLeNet神经网络模型：
GoogleLeNet系列特征改进：
Inception V2 ： 提出BN（Batch Normalization）算法,对每一个mini-batch数据的内部进行标准化处理，使输出范围到N(0, 1)分布，减少内部神经元分布的变化。基于BN算法的相应调整：增大学习速率并加快学习衰减速率以适应BN规范后的数据；去除Dropout并减轻L2正则；去除LRN，对数据更彻底shuffle；减少数据增强过程对数据的光学畸变；
Inception V3 ：引入Fraction into small convolutions思想，将一个较大的二维卷积拆成两个较小的一维卷积，节约大量参数，加速运算且避免过拟合，增加了一层非线性模型扩展能力；优化inception model结构，采用35x35, 17x17, 8x8三种结构。network in network in network。
Inception V4：结合微ResNet方法，后续学习中再总结。
基于Tensorflow的GoogleLeNet神经网络时间评估：
import tensorflow as tf
form datetime import datetime
import math
import time
#import the predefined googlenet network in slim
slim = tf.contrib.slim
#define the truncated normal distributation
trunc_normal = lambda stddev : tf.truncated_normal_initializer(0.0, stddev)
#define parameters dict
parameters = []
#define inception_v3_arg_scope function to predefined the inception unit parameters
def inception_v3_arg_scope(weight_decay = 0.00004, stddev = 0.1, batch_norm_var_collection = 'moving_vars'):
'''
description: to predefined the inception unit parameters
Args: weight_decay: the L2 regularation weight delay
stddev: stdandard devience
batch_norm_var_collection: the BP algorithm training parameters
Returns: scope: the
'''
#define the batch_norm_params
batch_norm_params = {'decay':0.9997, 'epsilon':0.001, 'updates_collections' : tf.GraphKeys.UPDATE_OPS,
'variables_collections' : {
'beta' : None, 'gramma' : None, 'moving_mean':[batch_norm_var_collection], 'moving_variance':[batch_norm_var_collection],
}
};
#initialze the slim.conv2d and slim full_connected
with slim.arg_scope([slim.conv2d, slim.full_connected], weights_regularizer = slim.l2_regularizer(weight_decay)):
with slim.arg_scope([slim.conv2d], weights_regularizer = tf.truncated_normal_initializer(stddev = stddev),
activation_fn = tf.nn.relu, normalizer_params = batch_norm_params) as scope:
return scope
#define inception_v3_base function to setup the (229 * 229) image as tensor
def inception_v3_base(x, scope = None):
'''
description: to setup the (229 * 299) image as tensor
Args: x : the input data
scope: the scope output form the last inception
Returns: net: the tensor branch sum
end_points: key points output
'''
#define the end_points nodes
end_points = {}
#define inceptionV3 network
with tf.variable_scope(scope, 'InceptionV3', [x]):
#set slim argments scope 1 padding is VALID
with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d], stride = 1, padding = 'VALID'):
#conv 229*229*3 image to 35 * 35 * 192 image with 3 3 * 3 kernal
net = slim.conv2d(x, 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')
#define max_pool2d
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')
#set slim argments scope 2 padding is SAME
with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d], stride = 1, padding = 'SAME'):
#the first inception model in inception 1
with tf.variable_scope('Mixed_5b'):
#channel 1 : 1 * 1 * 64
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
#channel 2 : 1 * 1 * 1 * 48 -> 1 * 5 * 5 * 64
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')
#channel 3 : 1 * 1 * 1 * 64 -> 2 * 3 * 3 * 96
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')
#channel 4 : 1 * 3 * 3 avg_pool2d -> 1 * 1 * 1 * 32
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')
#sum of channels: 64 + 64 + 96 + 32 = 256, 1 * 35 * 35 * 256
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the second inception model in inception 1
with tf.variable_scope('Mixed_5c'):
#channel 1 : 1 * 1 * 1 * 64
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
#channel 2 : 1 * 1 * 1 * 48 -> 1 * 5 * 5 * 64
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 = 'Conv2d_0b_5x5')
#channel 3 : 1 * 1 * 1 * 64 -> 2 * 3 * 3 * 96
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')
#channel 4 : 1 * 3 * 3 avg_pool2d -> 1 * 1 * 1 * 64
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')
#sum of channels: 64 + 64 + 96 + 64 = 288, 1 * 35 * 35 * 288
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the third inception in inception 1
with tf.variable_scope('Mixed_5d'):
#channel 1 : 1 * 1 * 1 * 64
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_0a_1x1')
#channel 2 : 1 * 1 * 1 * 48 -> 1 * 5 * 5 * 64
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')
#channel 3 : 1 * 1 * 1 * 64 -> 2 * 3 * 3 * 96
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')
#channel 4 : 1 * 3 * 3 avg_pool2d -> 1 * 1 * 1 * 6
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')
#sum of channels: 64 + 64 + 95 + 64 = 288, 1 * 35 * 35 * 288
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the first inception in inception 2
with tf.variable_scope('Mixed_6a'):
#channel 1 : 1 * 3 * 3 * 384 stride = 2, padding = 'VALID'
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 384, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_1x1')
#channel 2 : 1 * 1 * 1 * 64 -> 2 * 3 * 3 * 96
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 64, [1, 1], scope = 'Conv2d_1a_1x1')
branch_1 = slim.conv2d(branch_1, 96, [3, 3], scope = 'Conv2d_1b_3x3')
branch_1 = slim.conv2d(branch_1, 96, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1c_1x1')
#channel 3 : 1 * 3 * 3 avg_pool2d
with tf.variable_scope('Branch_2'):
branch_2 = slim.avg_pool2d(net, [3, 3], stride = 2, padding = 'VALID', scope = 'Conv2d_1a_1x1')
#sum of channels: 1 * 35 * 35 * 256
net = tf.concat([branch_0, branch_1, branch_2], 3)
#the second inception in inception 2
with tf.variable_scope('Mixed_6b'):
#channel 1 : 1 * 1 * 1 * 192
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_1a_1x1')
#channel 2 : 1 * 1 * 1 * 128 -> 1 * 1 * 7 * 128 -> 1 * 7 * 1 * 192
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 128, [1, 1], scope = 'Conv2d_1a_1x1')
branch_1 = slim.conv2d(branch_1, 128, [1, 7], scope = 'Conv2d_1b_1x7')
branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_1c_7x1')
#channel 3 : 1 * 1 * 1 * 128 -> 1 * 1 * 7 * 128 -> 1 * 7 * 1 * 128 -> 1 * 1 * 7 * 192
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 128, [1, 1], scope='Conv2d_1a_1x1')
branch_2 = slim.conv2d(branch_2, 128, [7, 1], scope='Conv2d_1b_7x1')
branch_2 = slim.conv2d(branch_2, 128, [1, 7], scope='Conv2d_1c_1x7')
branch_2 = slim.conv2d(branch_2, 128, [7, 1], scope='Conv2d_1d_7x1')
branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope='Conv2d_1e_1x7')
#channel 4 : 1 * 3 * 3 * avg_pool2d -> 1 * 1 * 1 * 192
with tf.variable_scope('Branch_3'):
branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_1a_3x3')
branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope='Conv2d_1b_1x1')
#sum of channels: 192 * 4 = 768, 1 * 17 * 17 * 768
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the first inception in inception 3
with tf.variable_scope('Mixed_6c'):
#channel 1 : 1 * 1 * 1 * 192
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_1a_1x1')
#channel 2 : 1 * 1 * 1 * 160 -> 1 * 1 * 7 * 160 -> 1 * 7 * 1 * 192
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2d_1a_1x1')
branch_1 = slim.conv2d(branch_1, 160, [1, 7], scope = 'Conv2d_1b_1x7')
branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_1c_7x1')
#channel 3 : 1 * 1 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 1 * 7 * 192
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_1a_1x1')
branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope='Conv2d_1b_7x1')
branch_2 = slim.conv2d(branch_2, 160, [1, 7], scope='Conv2d_1c_1x7')
branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope='Conv2d_1d_7x1')
branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope='Conv2d_1e_1x7')
#channel 4 : 1 * 3 * 3 * avg_pool2d -> 1 * 1 * 1 * 192
with tf.variable_scope('Branch_3'):
branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_1a_3x3')
branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope='Conv2d_1b_1x1')
#sum of channels: 192 * 4 = 768, 1 * 17 * 17 * 768
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the first inception in inception 4
with tf.variable_scope('Mixed_6d'):
#channel 1 : 1 * 1 * 1 * 192
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_1a_1x1')
#channel 2 : 1 * 1 * 1 * 160 -> 1 * 1 * 7 * 160 -> 1 * 7 * 1 * 192
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 160, [1, 1], scope = 'Conv2d_1a_1x1')
branch_1 = slim.conv2d(branch_1, 160, [1, 7], scope = 'Conv2d_1b_1x7')
branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_1c_7x1')
#channel 3 : 1 * 1 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 7 * 1 * 160 -> 1 * 1 * 7 * 192
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_1a_1x1')
branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope='Conv2d_1b_7x1')
branch_2 = slim.conv2d(branch_2, 160, [1, 7], scope='Conv2d_1c_1x7')
branch_2 = slim.conv2d(branch_2, 160, [7, 1], scope='Conv2d_1d_7x1')
branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope='Conv2d_1e_1x7')
#channel 4 : 1 * 3 * 3 * avg_pool2d -> 1 * 1 * 1 * 192
with tf.variable_scope('Branch_3'):
branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_1a_3x3')
branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope='Conv2d_1b_1x1')
#sum of channels: 192 * 4 = 768, 1 * 17 * 17 * 768
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#the first inception in inception 5
with tf.variable_scope('Mixed_6e'):
#channel 1 : 1 * 1 * 1 * 192
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_1a_1x1')
#channel 2 : 1 * 1 * 1 * 192 -> 1 * 1 * 7 * 192 -> 1 * 7 * 1 * 192
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 192, [1, 1], scope = 'Conv2d_1a_1x1')
branch_1 = slim.conv2d(branch_1, 192, [1, 7], scope = 'Conv2d_1b_1x7')
branch_1 = slim.conv2d(branch_1, 192, [7, 1], scope = 'Conv2d_1c_7x1')
#channel 3 : 1 * 1 * 1 * 192 -> 1 * 7 * 1 * 192 -> 1 * 7 * 1 * 192 -> 1 * 7 * 1 * 192 -> 1 * 1 * 7 * 192
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 192, [1, 1], scope='Conv2d_1a_1x1')
branch_2 = slim.conv2d(branch_2, 192, [7, 1], scope='Conv2d_1b_7x1')
branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope='Conv2d_1c_1x7')
branch_2 = slim.conv2d(branch_2, 192, [7, 1], scope='Conv2d_1d_7x1')
branch_2 = slim.conv2d(branch_2, 192, [1, 7], scope='Conv2d_1e_1x7')
#channel 4 : 1 * 3 * 3 * avg_pool2d -> 1 * 1 * 1 * 192
with tf.variable_scope('Branch_3'):
branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_1a_3x3')
branch_3 = slim.conv2d(branch_3, 192, [1, 1], scope='Conv2d_1b_1x1')
#sum of channels: 192 * 4 = 768, 1 * 17 * 17 * 768
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
#save the Mixed_6e into end_points
end_points['Mixed_6e'] = net
# Inception 3
with tf.variable_scope('Mixed_7c'):
# channel 1 : 1 * 320 * 1 * 1
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 320, [1, 1], scope='Conv2d_0a_1x1')
#channel 2 : 1 * 1 * 1 * 384(1 * 3 -> 3 * 1)
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_0c_3x1')], 3)
#channel 23 : 1 * 1 * 1 * 448, 1 * 3 * 3 * 384 (1 * 3 -> 3 * 1)
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')], 3)
#channel 4: 1 * 3 * 3 avg_pool2d, 1 * 1 * 1 * 192
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')
#sum of channels: 320 + 768 + 768 + 192 = 2048
net = tf.concat([branch_0, branch_1, branch_2, branch_3], 3)
return net, end_points
# total region avg_pool2d
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='InceptionV3'):
with tf.variable_scope(scope, 'InceptionV3', [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)
#set the slim argments scope parameters
with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d], stride=1, padding='SAME'):
aux_logits = end_points['Mixed_6e']
#AuxLogits nodes
with tf.variable_scope('AuxLogits'):
# 1 * 5 * 5 avg_pool2d -> 1 * 1 * 1 * 128 -> 1 * 5 * 5 * 768 -> 1 * 1 * 1 * class(1000)
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:
#squeeze operation
aux_logits = tf.squeeze(aux_logits, [1, 2], name='SpatialSqueeze')
end_points['AuxLogits'] = aux_logits
#define the class result scope
with tf.variable_scope('Logits'):
# 1 * 8 * 8 avg_pool2d
net = slim.avg_pool2d(net, [8, 8], padding='VALID', scope='AvgPool_1a_8x8')
# dropout layer
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:
#squeeze operation
logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')
end_points['Logits'] = logits
#softmax prediction
end_points['Predictions'] = prediction_fn(logits, scope='Predictions')
return logits, end_points
import math
from datetime import datetime
import time
#evaluate the time consuming
def time_tensorflow_run(session, target, info_string):
num_steps_burn_in = 10
total_duration = 0.0
total_duration_squared = 0.0
for i in range(num_batches + num_steps_burn_in):
tart_time = time.time()
_ = session.run(target)
duration = time.time()- start_time
if i >= num_steps_burn_in:
if not i % 10:
print ('%s: step %d, duration = %.3f' % (datetime.now().strftime('%X'), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
# calculate the aveage time consuming and standand variance
mn = total_duration / num_batches
vr = total_duration_squared / num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' % (datetime.now().strftime('%X'), info_string, num_batches, mn, sd))
#main process
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()):
#get logits and end_points result
logits, end_points = inception_v3(inputs, is_training=False)
#create the Session
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
num_batches = 100
time_tensorflow_run(sess, logits, 'Forward')practice makes perfect!
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
