第J8周：Inception v1算法实战与解析

最新推荐文章于 2024-07-19 15:45:46 发布

weixin_46559155

最新推荐文章于 2024-07-19 15:45:46 发布

阅读量631

点赞数 19

文章标签：算法

本文链接：https://blog.csdn.net/weixin_46559155/article/details/136952195

版权

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍖 原作者：K同学啊|接辅导、项目定制

Inception v1 论文
Going deeper with convolutions.pdf

一、理论知识

GoogLeNet首次出现在2014年ILSVRC 比赛中获得冠军。这次的版本通常称其为Inception V1。Inception V1有22层深，参数量为5M。同一时期的VGGNet性能和Inception V1差不多，但是参数量也是远大于Inception V1。

Inception Module是Inception V1的核心组成单元，提出了卷积层的并行结构，实现了在同一层就可以提取不同的特征，如下图。

按照这样的结构来增加网络的深度，虽然可以提升性能，但是还面临计算量大（参数多）的问题。为改善这种现象，Inception Module借鉴Network-in-Network的思想，使用1x1的卷积核实现降维操作(也间接增加了网络的深度)，以此来减小网络的参数量与计算量，如上图b所示。

备注举例：假如前一层的输出为100x100x128，经过具有256个5x5卷积核的卷积层之后(stride=1，pad=2)，输出数据为100x100x256。其中，卷积层的参数为5x5x128x256+256。假如上一层输出先经过具有32个1x1卷积核的卷积层（1x1卷积降低了通道数，且特征图尺寸不变），再经过具有256个5x5卷积核的卷积层，最终的输出数据仍为为100x100x256，但卷积参数量已经减少为128x1x1x32+32 + 32x5x5x256+256，参数数量减少为原来的约4分之一

1x1卷积核的作用： 1x1卷积核的最大作用是降低输入特征图的通道数，减小网络的参数量与计算量。

最后Inception Module基本由11卷积，33卷积，55卷积，33最大池化四个基本单元组成，对四个基本单元运算结果进行通道上组合，不同大小的卷积核赋予不同大小的感受野，从而提取到图像不同尺度的信息，进行融合，得到图像更好的表征，就是Inception Module的核心思想。

二、算法结构

实现的Inception V1网络结构图如下：

注：另外增加了两个辅助分支，作用有两点，一是为了避免梯度消失，用于向前传导梯度。反向传播时如果有一层求导为0，链式求导结果则为0。二是将中间某一层输出用作分类，起到模型融合作用，实际测试时，这两个辅助softmax分支会被去掉，在后续模型的发展中，该方法被采用较少，可以直接绕过，重点学习卷积层的并行结构与1x1卷积核部分的内容即可。

详细网络结构图如下：

三、Pytorch复现

下面的代码是参考上方表格中的网络框架和参数实现的

Inception Module的代码实现

''' InceptionV1 '''
class InceptionV1(nn.Module):
    def __init__(self,
                 include_top=True,  # 是否包含位于网络顶部的全链接层
                 preact=False,      # 是否使用预激活
                 use_bias=True,     # 是否对卷积层使用偏置
                 input_shape=[3, 224, 224],
                 classes=1000,
                 pooling=None):     # 用于分类图像的可选类数
        super(InceptionV1, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(64, 64, kernel_size=1, stride=1, padding=0),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 192, kernel_size=3, stride=1, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        )
        self.conv3 = nn.Sequential(
            InceptionBlock(192, 64, [96,128], [16,32], 32),
            InceptionBlock(256, 128, [128,192], [32,96], 64),
            nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        )
        self.conv4 = nn.Sequential(
            InceptionBlock(480, 192, [96,208], [16,48], 64),
            InceptionBlock(512, 160, [112,224], [24,64], 64),
            InceptionBlock(512, 128, [128,256], [24,64], 64),
            InceptionBlock(512, 112, [144,288], [32,64], 64),
            InceptionBlock(528, 256, [160,320], [32,128], 128),
            nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        )
        self.conv5 = nn.Sequential(
            InceptionBlock(832, 256, [160,320], [32,128], 128),
            InceptionBlock(832, 384, [192,384], [48,128], 128),
            nn.AvgPool2d(kernel_size=7,stride=1,padding=0),
            nn.Dropout(0.4)
        )
        self.fc = nn.Sequential(
            nn.Flatten(),
            nn.Linear(in_features=1024, out_features=1024),
            nn.ReLU(inplace=True),
            nn.Linear(in_features=1024, out_features=classes),
            nn.Softmax(dim=1)
        )
    
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.fc(x)
        return x

模型打印

----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 112, 112] 9,472
ReLU-2 [-1, 64, 112, 112] 0
MaxPool2d-3 [-1, 64, 56, 56] 0
Conv2d-4 [-1, 64, 56, 56] 4,160
ReLU-5 [-1, 64, 56, 56] 0
Conv2d-6 [-1, 192, 56, 56] 110,784
ReLU-7 [-1, 192, 56, 56] 0
MaxPool2d-8 [-1, 192, 28, 28] 0
Conv2d-9 [-1, 64, 28, 28] 12,352
BatchNorm2d-10 [-1, 64, 28, 28] 128
ReLU-11 [-1, 64, 28, 28] 0
Conv2d-12 [-1, 96, 28, 28] 18,528
BatchNorm2d-13 [-1, 96, 28, 28] 192
ReLU-14 [-1, 96, 28, 28] 0
Conv2d-15 [-1, 128, 28, 28] 110,720
BatchNorm2d-16 [-1, 128, 28, 28] 256
ReLU-17 [-1, 128, 28, 28] 0
Conv2d-18 [-1, 16, 28, 28] 3,088
BatchNorm2d-19 [-1, 16, 28, 28] 32
ReLU-20 [-1, 16, 28, 28] 0
Conv2d-21 [-1, 32, 28, 28] 12,832
BatchNorm2d-22 [-1, 32, 28, 28] 64
ReLU-23 [-1, 32, 28, 28] 0
MaxPool2d-24 [-1, 192, 28, 28] 0
Conv2d-25 [-1, 32, 28, 28] 6,176
BatchNorm2d-26 [-1, 32, 28, 28] 64
ReLU-27 [-1, 32, 28, 28] 0
InceptionBlock-28 [-1, 256, 28, 28] 0
Conv2d-29 [-1, 128, 28, 28] 32,896
BatchNorm2d-30 [-1, 128, 28, 28] 256
ReLU-31 [-1, 128, 28, 28] 0
Conv2d-32 [-1, 128, 28, 28] 32,896
BatchNorm2d-33 [-1, 128, 28, 28] 256
ReLU-34 [-1, 128, 28, 28] 0
Conv2d-35 [-1, 192, 28, 28] 221,376
BatchNorm2d-36 [-1, 192, 28, 28] 384
ReLU-37 [-1, 192, 28, 28] 0
Conv2d-38 [-1, 32, 28, 28] 8,224
BatchNorm2d-39 [-1, 32, 28, 28] 64
ReLU-40 [-1, 32, 28, 28] 0
Conv2d-41 [-1, 96, 28, 28] 76,896
BatchNorm2d-42 [-1, 96, 28, 28] 192
ReLU-43 [-1, 96, 28, 28] 0
MaxPool2d-44 [-1, 256, 28, 28] 0
Conv2d-45 [-1, 64, 28, 28] 16,448
BatchNorm2d-46 [-1, 64, 28, 28] 128
ReLU-47 [-1, 64, 28, 28] 0
InceptionBlock-48 [-1, 480, 28, 28] 0
MaxPool2d-49 [-1, 480, 14, 14] 0
Conv2d-50 [-1, 192, 14, 14] 92,352
BatchNorm2d-51 [-1, 192, 14, 14] 384
ReLU-52 [-1, 192, 14, 14] 0
Conv2d-53 [-1, 96, 14, 14] 46,176
BatchNorm2d-54 [-1, 96, 14, 14] 192
ReLU-55 [-1, 96, 14, 14] 0
Conv2d-56 [-1, 208, 14, 14] 179,920
BatchNorm2d-57 [-1, 208, 14, 14] 416
ReLU-58 [-1, 208, 14, 14] 0
Conv2d-59 [-1, 16, 14, 14] 7,696
BatchNorm2d-60 [-1, 16, 14, 14] 32
ReLU-61 [-1, 16, 14, 14] 0
Conv2d-62 [-1, 48, 14, 14] 19,248
BatchNorm2d-63 [-1, 48, 14, 14] 96
ReLU-64 [-1, 48, 14, 14] 0
MaxPool2d-65 [-1, 480, 14, 14] 0
Conv2d-66 [-1, 64, 14, 14] 30,784
BatchNorm2d-67 [-1, 64, 14, 14] 128
ReLU-68 [-1, 64, 14, 14] 0
InceptionBlock-69 [-1, 512, 14, 14] 0
Conv2d-70 [-1, 160, 14, 14] 82,080
BatchNorm2d-71 [-1, 160, 14, 14] 320
ReLU-72 [-1, 160, 14, 14] 0
Conv2d-73 [-1, 112, 14, 14] 57,456
BatchNorm2d-74 [-1, 112, 14, 14] 224
ReLU-75 [-1, 112, 14, 14] 0
Conv2d-76 [-1, 224, 14, 14] 226,016
BatchNorm2d-77 [-1, 224, 14, 14] 448
ReLU-78 [-1, 224, 14, 14] 0
Conv2d-79 [-1, 24, 14, 14] 12,312
BatchNorm2d-80 [-1, 24, 14, 14] 48
ReLU-81 [-1, 24, 14, 14] 0
Conv2d-82 [-1, 64, 14, 14] 38,464
BatchNorm2d-83 [-1, 64, 14, 14] 128
ReLU-84 [-1, 64, 14, 14] 0
MaxPool2d-85 [-1, 512, 14, 14] 0
Conv2d-86 [-1, 64, 14, 14] 32,832
BatchNorm2d-87 [-1, 64, 14, 14] 128
ReLU-88 [-1, 64, 14, 14] 0
InceptionBlock-89 [-1, 512, 14, 14] 0
Conv2d-90 [-1, 128, 14, 14] 65,664
BatchNorm2d-91 [-1, 128, 14, 14] 256
ReLU-92 [-1, 128, 14, 14] 0
Conv2d-93 [-1, 128, 14, 14] 65,664
BatchNorm2d-94 [-1, 128, 14, 14] 256
ReLU-95 [-1, 128, 14, 14] 0
Conv2d-96 [-1, 256, 14, 14] 295,168
BatchNorm2d-97 [-1, 256, 14, 14] 512
ReLU-98 [-1, 256, 14, 14] 0
Conv2d-99 [-1, 24, 14, 14] 12,312
BatchNorm2d-100 [-1, 24, 14, 14] 48
ReLU-101 [-1, 24, 14, 14] 0
Conv2d-102 [-1, 64, 14, 14] 38,464
BatchNorm2d-103 [-1, 64, 14, 14] 128
ReLU-104 [-1, 64, 14, 14] 0
MaxPool2d-105 [-1, 512, 14, 14] 0
Conv2d-106 [-1, 64, 14, 14] 32,832
BatchNorm2d-107 [-1, 64, 14, 14] 128
ReLU-108 [-1, 64, 14, 14] 0
InceptionBlock-109 [-1, 512, 14, 14] 0
Conv2d-110 [-1, 112, 14, 14] 57,456
BatchNorm2d-111 [-1, 112, 14, 14] 224
ReLU-112 [-1, 112, 14, 14] 0
Conv2d-113 [-1, 144, 14, 14] 73,872
BatchNorm2d-114 [-1, 144, 14, 14] 288
ReLU-115 [-1, 144, 14, 14] 0
Conv2d-116 [-1, 288, 14, 14] 373,536
BatchNorm2d-117 [-1, 288, 14, 14] 576
ReLU-118 [-1, 288, 14, 14] 0
Conv2d-119 [-1, 32, 14, 14] 16,416
BatchNorm2d-120 [-1, 32, 14, 14] 64
ReLU-121 [-1, 32, 14, 14] 0
Conv2d-122 [-1, 64, 14, 14] 51,264
BatchNorm2d-123 [-1, 64, 14, 14] 128
ReLU-124 [-1, 64, 14, 14] 0
MaxPool2d-125 [-1, 512, 14, 14] 0
Conv2d-126 [-1, 64, 14, 14] 32,832
BatchNorm2d-127 [-1, 64, 14, 14] 128
ReLU-128 [-1, 64, 14, 14] 0
InceptionBlock-129 [-1, 528, 14, 14] 0
Conv2d-130 [-1, 256, 14, 14] 135,424
BatchNorm2d-131 [-1, 256, 14, 14] 512
ReLU-132 [-1, 256, 14, 14] 0
Conv2d-133 [-1, 160, 14, 14] 84,640
BatchNorm2d-134 [-1, 160, 14, 14] 320
ReLU-135 [-1, 160, 14, 14] 0
Conv2d-136 [-1, 320, 14, 14] 461,120
BatchNorm2d-137 [-1, 320, 14, 14] 640
ReLU-138 [-1, 320, 14, 14] 0
Conv2d-139 [-1, 32, 14, 14] 16,928
BatchNorm2d-140 [-1, 32, 14, 14] 64
ReLU-141 [-1, 32, 14, 14] 0
Conv2d-142 [-1, 128, 14, 14] 102,528
BatchNorm2d-143 [-1, 128, 14, 14] 256
ReLU-144 [-1, 128, 14, 14] 0
MaxPool2d-145 [-1, 528, 14, 14] 0
Conv2d-146 [-1, 128, 14, 14] 67,712
BatchNorm2d-147 [-1, 128, 14, 14] 256
ReLU-148 [-1, 128, 14, 14] 0
InceptionBlock-149 [-1, 832, 14, 14] 0
MaxPool2d-150 [-1, 832, 7, 7] 0
Conv2d-151 [-1, 256, 7, 7] 213,248
BatchNorm2d-152 [-1, 256, 7, 7] 512
ReLU-153 [-1, 256, 7, 7] 0
Conv2d-154 [-1, 160, 7, 7] 133,280
BatchNorm2d-155 [-1, 160, 7, 7] 320
ReLU-156 [-1, 160, 7, 7] 0
Conv2d-157 [-1, 320, 7, 7] 461,120
BatchNorm2d-158 [-1, 320, 7, 7] 640
ReLU-159 [-1, 320, 7, 7] 0
Conv2d-160 [-1, 32, 7, 7] 26,656
BatchNorm2d-161 [-1, 32, 7, 7] 64
ReLU-162 [-1, 32, 7, 7] 0
Conv2d-163 [-1, 128, 7, 7] 102,528
BatchNorm2d-164 [-1, 128, 7, 7] 256
ReLU-165 [-1, 128, 7, 7] 0
MaxPool2d-166 [-1, 832, 7, 7] 0
Conv2d-167 [-1, 128, 7, 7] 106,624
BatchNorm2d-168 [-1, 128, 7, 7] 256
ReLU-169 [-1, 128, 7, 7] 0
InceptionBlock-170 [-1, 832, 7, 7] 0
Conv2d-171 [-1, 384, 7, 7] 319,872
BatchNorm2d-172 [-1, 384, 7, 7] 768
ReLU-173 [-1, 384, 7, 7] 0
Conv2d-174 [-1, 192, 7, 7] 159,936
BatchNorm2d-175 [-1, 192, 7, 7] 384
ReLU-176 [-1, 192, 7, 7] 0
Conv2d-177 [-1, 384, 7, 7] 663,936
BatchNorm2d-178 [-1, 384, 7, 7] 768
ReLU-179 [-1, 384, 7, 7] 0
Conv2d-180 [-1, 48, 7, 7] 39,984
BatchNorm2d-181 [-1, 48, 7, 7] 96
ReLU-182 [-1, 48, 7, 7] 0
Conv2d-183 [-1, 128, 7, 7] 153,728
BatchNorm2d-184 [-1, 128, 7, 7] 256
ReLU-185 [-1, 128, 7, 7] 0
MaxPool2d-186 [-1, 832, 7, 7] 0
Conv2d-187 [-1, 128, 7, 7] 106,624
BatchNorm2d-188 [-1, 128, 7, 7] 256
ReLU-189 [-1, 128, 7, 7] 0
InceptionBlock-190 [-1, 1024, 7, 7] 0
AvgPool2d-191 [-1, 1024, 1, 1] 0
Dropout-192 [-1, 1024, 1, 1] 0
Flatten-193 [-1, 1024] 0
Linear-194 [-1, 1024] 1,049,600
ReLU-195 [-1, 1024] 0
Linear-196 [-1, 2] 2,050
Softmax-197 [-1, 2] 0
================================================================
Total params: 7,039,122
Trainable params: 7,039,122
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 81.87
Params size (MB): 26.85
Estimated Total Size (MB): 109.30
----------------------------------------------------------------

InceptionV1(
(conv1): Sequential(
(0): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3))
(1): ReLU(inplace=True)
(2): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
)
(conv2): Sequential(
(0): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
)
(conv3): Sequential(
(0): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(192, 96, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(96, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(192, 16, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(192, 32, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(1): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(128, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(256, 32, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(32, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(2): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
)
(conv4): Sequential(
(0): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(480, 192, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(480, 96, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(96, 208, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(208, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(480, 16, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(16, 48, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(480, 64, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(1): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(512, 160, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(512, 112, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(112, 224, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(224, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(512, 24, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(24, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(2): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(512, 24, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(24, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(3): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(512, 112, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(512, 144, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(144, 288, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(288, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(512, 32, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(512, 64, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(4): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(528, 256, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(528, 160, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(160, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(528, 32, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(32, 128, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(528, 128, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(5): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
)
(conv5): Sequential(
(0): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(832, 256, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(832, 160, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(160, 320, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(832, 32, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(32, 128, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(832, 128, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(1): InceptionBlock(
(conv1): Sequential(
(0): Conv2d(832, 384, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
)
(conv2): Sequential(
(0): Conv2d(832, 192, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(192, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv3): Sequential(
(0): Conv2d(832, 48, kernel_size=(1, 1), stride=(1, 1))
(1): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(48, 128, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
)
(conv4): Sequential(
(0): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(1): Conv2d(832, 128, kernel_size=(1, 1), stride=(1, 1))
(2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(3): ReLU(inplace=True)
)
)
(2): AvgPool2d(kernel_size=7, stride=1, padding=0)
(3): Dropout(p=0.4, inplace=False)
)
(fc): Sequential(
(0): Flatten(start_dim=1, end_dim=-1)
(1): Linear(in_features=1024, out_features=1024, bias=True)
(2): ReLU(inplace=True)
(3): Linear(in_features=1024, out_features=2, bias=True)
(4): Softmax(dim=1)
)
)

weixin_46559155

关注

19
点赞
踩
8

收藏

觉得还不错? 一键收藏
1
评论
第J8周：Inception v1算法实战与解析

二是将中间某一层输出用作分类，起到模型融合作用，实际测试时，这两个辅助softmax分支会被去掉，在后续模型的发展中，该方法被采用较少，可以直接绕过，重点学习卷积层的并行结构与1x1卷积核部分的内容即可。最后Inception Module基本由11卷积，33卷积，55卷积，33最大池化四个基本单元组成，对四个基本单元运算结果进行通道上组合，不同大小的卷积核赋予不同大小的感受野，从而提取到图像不同尺度的信息，进行融合，得到图像更好的表征，就是Inception Module的核心思想。
复制链接

扫一扫