pytorch版GoogLeNet(V 1)

10449 ワード

GoogLeNet(V 1)ネットワークモデルを論文『Going deeper with convolutions』に示す.
pytorchプログラミング実装は主にボリューム、補助分類器(auxiliary classifiers)、inception構造、GoogLeNetネットワーク構造の4つのclassに分けられる.
1.ボリューム(conv)
Googlenetのボリューム操作には、一般的に、ボリューム(conv)+一括正規化(BN)+アクティブ化関数(relu)が含まれています.
class BasicConv2d(nn.Module):                                     #     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       ,  bn      。
        self.bn = nn.BatchNorm2d(out_channels, eps=0.001)      # eps     bn    0,   
    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        return F.relu(x, inplace=True)          # inplace          ,true         

注意:関数を定義するときによく遭遇します.  *Argsと  **kwargs ,
           *args              list,**kwargs        key value dictionary

2.補助分類器(auxiliary classifiers)
  著者らが2つの補助分類器を設計した理由は,勾配の消失(ネットワーク層数が多い場合にしばしば現れる)を避けるために,中間層出力で逆伝搬を増強するためである.二つ目は、比較的浅いネットワークが強力な性能を有している場合、ネットワーク中間層が生成する特徴がより区別されるべきであるという認識に基づいている.pytorchが公式に与えた補助分類器コードは以下の通りである.
class InceptionAux(nn.Module):        #     avepooling + conv + flatten + fc + relu + dropout + fc,            
    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

注意:訓練時には、これらの分類器の損失は重み付け(論文で設定した重み0.3)の形でネットワークの総損失を計算されるが、test/predictの場合、これらの補助ネットワークは廃棄される. 
3.inception構造
著者らはinception構造を多尺度の観点から設計し,1*1ボリュームbranch,3*3ボリュームbranch,5*5ボリュームbranch,maxpooling branchに分けた.このうち,3*3ボリューム,5*5ボリューム,maxpooling branchはいずれも1*1ボリュームを加えて次元を減らす.以下はpytorch公式に与えられたinception構造コードである.
class Inception(nn.Module):                                           # inception v1 :  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  
        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  
        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  
        )
        self.branch4 = nn.Sequential(
            nn.MaxPool2d(kernel_size=3, stride=1, padding=1, ceil_mode=True),    # ceil_mode true        ,         
            conv_block(in_channels, pool_proj, kernel_size=1)               # max pooling,1*1  
        )

    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          。(0     )

注意:(1)接尾辞がredの場合、例えば3*3 red、5*5 redであり、実際には1*1ボリュームであり、次元を減らすために使用される(redはreduceの略).
       (2)branch 3,すなわち5*5分岐路には,5*5ボリュームであるはずの既知のバグがあるが,与えられたコードは3*3ボリュームである.(2つの3*3ボリュームは1つの5*5ボリュームに等しいが、2つの3*3ボリュームパラメータは1つの5*5ボリュームパラメータの18/25しかない)
4.googlenetネットワーク構造
pytorch公式に与えられたコードは以下の通りです.
class GoogLeNet(nn.Module):                  # googlenet       
    __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()          (     )       (     )。
                import scipy.stats as stats
                X = stats.truncnorm(-2, 2, scale=0.01)    #       
                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()   @torch.no_grad()           ,         
                    m.weight.copy_(values)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)      #  1 0    

    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   #        1   
            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

注意:次のライブラリをインポートする必要があります.
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

最後に、以下のコードを入力してネットワークモデル全体を印刷することができます(汎用で、他のネットワークに適しています).
if __name__=="__main__":
    #  .py        ,if __name__ == '__main__'          ; .py           ,if __name__ == '__main__'          。
    model=GoogLeNet()
    print(model,(3,224,224))