GoogleLeNet(Inception-V1)论文及代码解析
目录
-
GoogleLeNet论文
-
tensorflow代码分析
-
小结
GoogleLeNet论文
GoogleLeNet是2014年ImageNet中ILSVRC14竞赛的冠军，和VGG网络是同一届，VGG网络是当年竞赛的亚军。但是实际上两个网络的TOP-5错误率相差并不多，GoogleLeNet的网络结构相对VGG复杂一些，是一个22层的网络，并且提出了一种Inception的结构，是一个很大的进步。
论文地址 论文的标题是“Going deeper with convolutions”，跟VGG一样也是遵循需要将网络设计的更深的思想。网络名字取名为GoogleLeNet是为了向CNN网络的开山鼻祖LeNet致敬。
1.前言
论文中作者花了大量的篇幅来描述当前分类和检测网络取得的成绩，以及遇到的问题，然后引出作者对现状改进的思路。
众所周知，改进深度神经网络最直接的方式是增加网络的大小，包括增加深度和宽度。增加深度就是增加神经网络的层数，增加宽度就是增加每一层的units个数。但是增加了深度和宽度会带来两个问题：1.因为参数变多而使得网络更加容易overfitting，反而降低了精度。2.会大大增加计算量，而且可能训练出来部分参数最终为0，这样在计算资源有限的情况下浪费了计算资源。所以我们最终的目标是：减少参数，减少计算量，增加深度，增加宽度。
解决方法就是将全连接替换为稀疏连接结构，但是现有的计算方式对非均匀的稀疏计算效率非常低，所以作者提出了自己的改进方式。
2.Inception模型
上图是作者提出的两种Inception的模型，b模型是a模型的改进版本。
先说a模型，论文中说是以一种稠密组件去逼近和替代一个最优的局部稀疏结构。我的理解是这种连接模型既会有四种连接方式带来的信息汇总，而且相对直接3x3的Conv进行连接的方式参数反而有所减少。为何参数会有减少呢？举个下图的例子(只是举个例子，并不是真实使用情况)：
接着作者又发现，虽然参数量是有减少，但是减少的并不是太多，而且如果输入的数据channel值比较大，参数量仍然会比较大，所以作者提出了b模型。b模型相对a模型而言，在使用3x3或者5x5的Conv连接之前用了1x1的Conv进行降维，这样会使参数量再一次减少，并且在1x1Conv后也使用非线性**就增加了网络的非线性。参数量变化如下图(只是举个例子，并不是真实使用情况)：
总之，Inception模型的好处是既能增加网络的深度和宽度，又不会增加计算量，而且稀疏连接的方式还能有助于减少过拟合。
3.网络结构
论文中的网络结构如下图：
- Input为224x224x3的RGB图片，同样减去每个颜色通道的均值(同vgg网络一样)
- #3x3 reduce表示在3x3Conv之前1x1Conv连接的channel数量，同样#5x5 reduce表示在5x5Conv之前1x1Conv链接的channel数量.
- pool proj表示inception中max pooling后的1x1Conv连接的channel数量
- 只计算有参数的层数是22层，所有的卷基层都用Relu**，包括inception内部的卷积
- 作者发现去掉最后的FC，改为avg pool，精确度有提升，但是还要保留后面的dropout
另外作者担心这么深的网络越往后信息的传播能力会受损，但是作者发现中间层的网络还带有比较多的信息，因此在inception(4a)和inception(4d)后面增加了一个比较小的网络，做了softmax输出，和最后的softmax输出相加，但是权重设置为0.3.也就是最后的output是softmax2+0.3*softmax1+0.3*softmax0
inception(4a)处的softmax
inception(4d)处的softmax
最后output的softmax
对softmax0和softmax1处的小网络的说明：
- average pool是5x5 filter size，stride是3，inception(4a)会得到4x4x512的output，inception(4d)会得到4x4x528的output
- Conv 1x1是128个filter，后面跟着Relu进行**
- FC是1024 units，后面跟着Relu**
- 70%units被dropout
- softmax是1000类，但是预测的时候移除
训练时使用随机梯度下降，momentum为0.9，8个epoch学习率下降4%
tensorflow代码分析
代码地址。下面我们对照代码实现和上图中的网络模型来更进一步理解网络。inception的结构主要是在inception_v1_base函数中实现的，函数的实现对照论文中的网络模型也是以一个一个的block为单位。
1. 下面是第一个block，对应论文网络结构表中的第一行。filter size是7x7，filter个数是64，stride是2，padding从slim.arg_scope中可以看出默认是same。
end_point = 'Conv2d_1a_7x7'
net = slim.conv2d(inputs, 64, [7, 7], stride=2, scope=end_point)当padding为SAME时，tf会自动计算需要padding的值使output的size为input size除以stride，所以这里output size会是112x112，如果这里的padding为VALID，那么最后的output size会是109x109，计算方法是(224-7)/2 + 1 = 109
2. 下面是第二个block，有a,b,c三部分，对应论文网络结构表的第二和第三行。因为第三行的depth是2，所以convolution是两层，一个是#3x3 reduce表示的3x3之前的1x1 convolution计算，一个是#3x3 filter的convolution计算。
end_point = 'MaxPool_2a_3x3'
net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)
...
end_point = 'Conv2d_2b_1x1'
net = slim.conv2d(net, 64, [1, 1], scope=end_point)
...
end_point = 'Conv2d_2c_3x3'
net = slim.conv2d(net, 192, [3, 3], scope=end_point)代码实现中似乎略过了LocalRespNorm
3. 下面是第三个block，也是a，b，c三个部分，对应论文网络结构表中的第四，第五和第六行。
end_point = 'MaxPool_3a_3x3'
net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_3b'
with tf.variable_scope(end_point):
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, 96, [1, 1], scope='Conv2d_0a_1x1')
branch_1 = slim.conv2d(branch_1, 128, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 16, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 32, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 32, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_3c'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 128, [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, 192, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points4. 下面是第四个block，有a，b，c，d，e，f这六个部分，对应论文网络结构表的7-12行。
end_point = 'MaxPool_4a_3x3'
net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_4b'
with tf.variable_scope(end_point):
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, 96, [1, 1], scope='Conv2d_0a_1x1')
branch_1 = slim.conv2d(branch_1, 208, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 16, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 48, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_4c'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_0a_1x1')
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 112, [1, 1], scope='Conv2d_0a_1x1')
branch_1 = slim.conv2d(branch_1, 224, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 24, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_4d'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 128, [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, 256, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 24, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_4e'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 112, [1, 1], scope='Conv2d_0a_1x1')
with tf.variable_scope('Branch_1'):
branch_1 = slim.conv2d(net, 144, [1, 1], scope='Conv2d_0a_1x1')
branch_1 = slim.conv2d(branch_1, 288, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_4f'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 256, [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, 320, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points5. 下面是第五个block，有a，b，c三个部分。对应论文结构表中的13-15行
end_point = 'MaxPool_5a_2x2'
net = slim.max_pool2d(net, [2, 2], stride=2, scope=end_point)
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_5b'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 256, [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, 320, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0a_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
end_point = 'Mixed_5c'
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = slim.conv2d(net, 384, [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, 384, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_2'):
branch_2 = slim.conv2d(net, 48, [1, 1], scope='Conv2d_0a_1x1')
branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0b_3x3')
with tf.variable_scope('Branch_3'):
branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')
branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')
net = tf.concat(
axis=3, values=[branch_0, branch_1, branch_2, branch_3])
end_points[end_point] = net
if final_endpoint == end_point: return net, end_points
主要的网络实现后，在inception_v1中实现了最后的softmax，对应论文网络结构中的最后四行，avg_pool，dropout，linear(用Convolution运算来替代)和softmax(prediction_fn是函数传入的参数，默认为slim.softmax)
# Pooling with a fixed kernel size.
net = slim.avg_pool2d(net, [7, 7], stride=1, scope='AvgPool_0a_7x7')
end_points['AvgPool_0a_7x7'] = net
...
net = slim.dropout(net, dropout_keep_prob, scope='Dropout_0b')
logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,
normalizer_fn=None, scope='Conv2d_0c_1x1')
...
end_points['Predictions'] = prediction_fn(logits, scope='Predictions')代码中所有的中间参数都放在了end_points中，便于用户做fine tuning。
小结
GoogleLeNet也叫做inception V1提出了inception block的结构，在不增加网络参数的情况下让网络变的越来越宽，越来越深。用1x1的Conv来做降维，用average pooling来替代全链接层，都是作者提出的改进点，为后续的v2,v3,v4打下了基础。
上一篇： PHP面试题之小杂鱼