pytorch版 GoogLeNet（V1）
GoogLeNet（V1）网络模型如论文《 Going deeper with convolutions 》所示。
pytorch编程实现主要分为卷积、辅助分类器（auxiliary classifiers）、inception结构、GoogLeNet网络结构 四个class。
1.卷积（conv）
googlenet的卷积操作一般包括卷积（conv）+批量归一化（BN）+激活函数（relu），以上归纳为一个基础卷积块：
class BasicConv2d(nn.Module): # 实现了 conv + bn + relu
def __init__(self, in_channels, out_channels, **kwargs):
super(BasicConv2d, self).__init__()
self.conv = nn.Conv2d(in_channels, out_channels, bias=False, **kwargs) # bn前卷积不加偏置，因为bn时会减去均值。
self.bn = nn.BatchNorm2d(out_channels, eps=0.001) # eps是为了防止bn时分母为0，故加上
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
return F.relu(x, inplace=True) # inplace代表是否进行覆盖运算，true时可以节省运算内存注：定义函数时，我们经常会遇见 *args 和 **kwargs ，
*args可以当作可容纳多个变量组成的list，**kwargs可以当作容纳多个key和value的dictionary
2.辅助分类器（ auxiliary classifiers ）
作者设计加入两个辅助分类器的原因有两个：一是为了避免梯度消失（网络层数较多时经常会出现），故在中间层输出以增强反向传播。二是基于这样一种认识：相对较浅的网络已经具有强大的性能，那么网络中间层产生的特征应该更有区分性。pytorch官方给出的辅助分类器代码如下。
class InceptionAux(nn.Module): # 实现了 avepooling + conv + flatten + fc + relu + dropout + fc，从中间层输出用以辅助分类
def __init__(self, in_channels, num_classes, conv_block=None):
super(InceptionAux, self).__init__()
if conv_block is None:
conv_block = BasicConv2d
self.conv = conv_block(in_channels, 128, kernel_size=1)
self.fc1 = nn.Linear(2048, 1024)
self.fc2 = nn.Linear(1024, num_classes)
def forward(self, x):
# aux1: N x 512 x 14 x 14, aux2: N x 528 x 14 x 14
x = F.adaptive_avg_pool2d(x, (4, 4))
# aux1: N x 512 x 4 x 4, aux2: N x 528 x 4 x 4
x = self.conv(x)
# N x 128 x 4 x 4
x = torch.flatten(x, 1)
# N x 2048
x = F.relu(self.fc1(x), inplace=True)
# N x 1024
x = F.dropout(x, 0.7, training=self.training)
# N x 1024
x = self.fc2(x)
# N x 1000 (num_classes)
return x注：在训练时，这些分类器的损失会被以加权（论文设置的权重为0.3）的形式计算网络的总损失当中，但是在 test/predict 时，这些辅助网络会被废弃不用。
3.inception结构
作者设计inception结构是从多尺度的角度考虑的，分为1*1卷积branch、3*3卷积branch、5*5卷积branch、maxpooling branch。其中，3*3卷积、5*5卷积、maxpooling branch都加入了1*1卷积用来减少维度。以下为pytorch官方给出的inception结构代码。
class Inception(nn.Module): # inception v1 ：四个branch
def __init__(self, in_channels, ch1x1, ch3x3red, ch3x3, ch5x5red, ch5x5, pool_proj,
conv_block=None):
super(Inception, self).__init__()
if conv_block is None:
conv_block = BasicConv2d
self.branch1 = conv_block(in_channels, ch1x1, kernel_size=1) # 1*1支路
self.branch2 = nn.Sequential(
conv_block(in_channels, ch3x3red, kernel_size=1),
conv_block(ch3x3red, ch3x3, kernel_size=3, padding=1)
) # 1*1，3*3支路
self.branch3 = nn.Sequential(
conv_block(in_channels, ch5x5red, kernel_size=1),
# Here, kernel_size=3 instead of kernel_size=5 is a known bug.
# Please see https://github.com/pytorch/vision/issues/906 for details.
conv_block(ch5x5red, ch5x5, kernel_size=3, padding=1) # 1*1，5*5支路
)
self.branch4 = nn.Sequential(
nn.MaxPool2d(kernel_size=3, stride=1, padding=1, ceil_mode=True), # ceil_mode为true时结果会向上取整，代替默认的向下取整
conv_block(in_channels, pool_proj, kernel_size=1) # max pooling，1*1支路
)
def _forward(self, x):
branch1 = self.branch1(x)
branch2 = self.branch2(x)
branch3 = self.branch3(x)
branch4 = self.branch4(x)
outputs = [branch1, branch2, branch3, branch4]
return outputs
def forward(self, x):
outputs = self._forward(x)
return torch.cat(outputs, 1) # 1 为将输出横着拼起来。（0 为竖着拼）注：（1）后缀为red的，例如3*3red、5*5red，实际为1*1卷积，用来减少维度的（red为reduce的缩写）。
（2）在branch3中，即5*5支路中，有一个已知的bug，原本该为5*5卷积的，但给的代码为一个3*3卷积。（两个3*3卷积等价于一个5*5卷积，但两个3*3卷积参数只有一个5*5卷积参数的18/25）
4. googlenet网络结构
pytorch官方给的代码如下。
class GoogLeNet(nn.Module): # googlenet 整个网络模型
__constants__ = ['aux_logits', 'transform_input']
def __init__(self, num_classes=1000, aux_logits=True, transform_input=False, init_weights=True,
blocks=None):
super(GoogLeNet, self).__init__()
if blocks is None:
blocks = [BasicConv2d, Inception, InceptionAux]
if init_weights is None:
warnings.warn('The default weight initialization of GoogleNet will be changed in future releases of '
'torchvision. If you wish to keep the old behavior (which leads to long initialization times'
' due to scipy/scipy#11299), please set init_weights=True.', FutureWarning)
init_weights = True
assert len(blocks) == 3
conv_block = blocks[0]
inception_block = blocks[1]
inception_aux_block = blocks[2]
self.aux_logits = aux_logits
self.transform_input = transform_input
self.conv1 = conv_block(3, 64, kernel_size=7, stride=2, padding=3)
self.maxpool1 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.conv2 = conv_block(64, 64, kernel_size=1)
self.conv3 = conv_block(64, 192, kernel_size=3, padding=1)
self.maxpool2 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception3a = inception_block(192, 64, 96, 128, 16, 32, 32)
self.inception3b = inception_block(256, 128, 128, 192, 32, 96, 64)
self.maxpool3 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception4a = inception_block(480, 192, 96, 208, 16, 48, 64)
self.inception4b = inception_block(512, 160, 112, 224, 24, 64, 64)
self.inception4c = inception_block(512, 128, 128, 256, 24, 64, 64)
self.inception4d = inception_block(512, 112, 144, 288, 32, 64, 64)
self.inception4e = inception_block(528, 256, 160, 320, 32, 128, 128)
self.maxpool4 = nn.MaxPool2d(2, stride=2, ceil_mode=True)
self.inception5a = inception_block(832, 256, 160, 320, 32, 128, 128)
self.inception5b = inception_block(832, 384, 192, 384, 48, 128, 128)
if aux_logits:
self.aux1 = inception_aux_block(512, num_classes)
self.aux2 = inception_aux_block(528, num_classes)
else:
self.aux1 = None
self.aux2 = None
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.dropout = nn.Dropout(0.2)
self.fc = nn.Linear(1024, num_classes)
if init_weights:
self._initialize_weights()
def _initialize_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear): # isinstance() 函数来判断一个对象（第一个参数）是否是一个类型（第二个参数）。
import scipy.stats as stats
X = stats.truncnorm(-2, 2, scale=0.01) # 截断正态分布
values = torch.as_tensor(X.rvs(m.weight.numel()), dtype=m.weight.dtype)
values = values.view(m.weight.size())
with torch.no_grad(): # with torch.no_grad()或者 @torch.no_grad()中的数据不需要计算梯度，也不会进行反向传播
m.weight.copy_(values)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0) # 用1或0填充向量
def _transform_input(self, x):
# type: (Tensor) -> Tensor
if self.transform_input:
x_ch0 = torch.unsqueeze(x[:, 0], 1) * (0.229 / 0.5) + (0.485 - 0.5) / 0.5
x_ch1 = torch.unsqueeze(x[:, 1], 1) * (0.224 / 0.5) + (0.456 - 0.5) / 0.5
x_ch2 = torch.unsqueeze(x[:, 2], 1) * (0.225 / 0.5) + (0.406 - 0.5) / 0.5 # 加上一个维数为1的维度
x = torch.cat((x_ch0, x_ch1, x_ch2), 1)
return x
def _forward(self, x):
# type: (Tensor) -> Tuple[Tensor, Optional[Tensor], Optional[Tensor]]
# N x 3 x 224 x 224
x = self.conv1(x)
# N x 64 x 112 x 112
x = self.maxpool1(x)
# N x 64 x 56 x 56
x = self.conv2(x)
# N x 64 x 56 x 56
x = self.conv3(x)
# N x 192 x 56 x 56
x = self.maxpool2(x)
# N x 192 x 28 x 28
x = self.inception3a(x)
# N x 256 x 28 x 28
x = self.inception3b(x)
# N x 480 x 28 x 28
x = self.maxpool3(x)
# N x 480 x 14 x 14
x = self.inception4a(x)
# N x 512 x 14 x 14
aux1 = torch.jit.annotate(Optional[Tensor], None)
if self.aux1 is not None:
if self.training:
aux1 = self.aux1(x)
x = self.inception4b(x)
# N x 512 x 14 x 14
x = self.inception4c(x)
# N x 512 x 14 x 14
x = self.inception4d(x)
# N x 528 x 14 x 14
aux2 = torch.jit.annotate(Optional[Tensor], None)
if self.aux2 is not None:
if self.training:
aux2 = self.aux2(x)
x = self.inception4e(x)
# N x 832 x 14 x 14
x = self.maxpool4(x)
# N x 832 x 7 x 7
x = self.inception5a(x)
# N x 832 x 7 x 7
x = self.inception5b(x)
# N x 1024 x 7 x 7
x = self.avgpool(x)
# N x 1024 x 1 x 1
x = torch.flatten(x, 1)
# N x 1024
x = self.dropout(x)
x = self.fc(x)
# N x 1000 (num_classes)
return x, aux2, aux1注：需导入以下库。
import warnings
from collections import namedtuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.jit.annotations import Optional, Tuple
from torch import Tensor最后，可输入以下代码打印出整个网络模型（通用，可适用于其他网络）。
if __name__=="__main__":
# 当.py文件被直接运行时，if __name__ == '__main__'之下的代码块将被运行；当.py文件以模块形式被导入时，if __name__ == '__main__'之下的代码块不被运行。
model=GoogLeNet()
print(model,(3,224,224))