【图像分类】【深度学习】【Pytorch版本】Inception-ResNet模型算法详解

news/2024/6/23 18:47:06

【图像分类】【深度学习】【Pytorch版本】Inception-ResNet模型算法详解

文章目录

  • 【图像分类】【深度学习】【Pytorch版本】Inception-ResNet模型算法详解
  • 前言
  • Inception-ResNet讲解
    • Inception-ResNet-V1
    • Inception-ResNet-V2
    • 残差模块的缩放(Scaling of the Residuals)
    • Inception-ResNet的总体模型结构
  • GoogLeNet(Inception-ResNet) Pytorch代码
    • ## Inception-ResNet-V1
    • Inception-ResNet-V2
  • 完整代码
    • Inception-ResNet-V1
    • Inception-ResNet-V2
  • 总结


前言

GoogLeNet(Inception-ResNet)是由谷歌的Szegedy, Christian等人在《Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning【AAAI-2017】》【论文地址】一文中提出的改进模型,受启发于ResNet【参考】在深度网络上较好的表现影响,论文将残差连接加入到Inception结构中形成2个Inception-ResNet版本的网络,它将残差连接取代原本Inception块中池化层部分,并将拼接变成了求和相加,提升了Inception的训练速度。

因为InceptionV4、Inception-Resnet-v1和Inception-Resnet-v2同出自一篇论文,大部分读者对InceptionV4存在误解,认为它是Inception模块与残差学习的结合,其实InceptionV4没有使用残差学习的思想,它基本延续了Inception v2/v3的结构,只有Inception-Resnet-v1和Inception-Resnet-v2才是Inception模块与残差学习的结合产物。


Inception-ResNet讲解

Inception-ResNet的核心思想是将Inception模块和ResNet模块进行融合,以利用它们各自的优点。Inception模块通过并行多个不同大小的卷积核来捕捉多尺度的特征,而ResNet模块通过残差连接解决了深层网络中的梯度消失和梯度爆炸问题,有助于更好地训练深层模型。Inception-ResNet使用了与InceptionV4【参考】类似的Inception模块,并在其中引入了ResNet的残差连接。这样,网络中的每个Inception模块都包含了两个分支:一个是常规的Inception结构,另一个是包含残差连接的Inception结构。这种设计使得模型可以更好地学习特征表示,并且在训练过程中可以更有效地传播梯度。

Inception-ResNet-V1

Inception-ResNet-v1:一种和InceptionV3【参考】具有相同计算损耗的结构。

  1. Stem结构: Inception-ResNet-V1的Stem结构类似于此前的InceptionV3网络中Inception结构组之前的网络层。

    所有卷积中没有标记为V表示填充方式为"SAME Padding",输入和输出维度一致;标记为V表示填充方式为"VALID Padding",输出维度视具体情况而定。

  2. Inception-resnet-A结构: InceptionV4网络中Inception-A结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。

    Inception-resnet结构残差连接代替了Inception中的池化层,并用残差连接相加操作取代了原Inception块中的拼接操作。

  3. Inception-resnet-B结构: InceptionV4网络中Inception-B结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。

  4. Inception-resnet-C结构: InceptionV4网络中Inception-C结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。

  5. Redution-A结构: 与InceptionV4网络中Redution-A结构一致,区别在于卷积核的个数。

    k和l表示卷积个数,不同网络结构的redution-A结构k和l是不同的。

  6. Redution-B结构:
    .

Inception-ResNet-V2

Inception-ResNet-v2:这是一种和InceptionV4具有相同计算损耗的结构,但是训练速度要比纯Inception-v4要快
Inception-ResNet-v2的整体框架和Inception-ResNet-v1的一致,除了Inception-ResNet-v2的stem结构与Inception V4的相同,其他的的结构Inception-ResNet-v2与Inception-ResNet-v1的类似,只不过卷积的个数Inception-ResNet-v2数量更多。

  1. Stem结构: Inception-ResNet-v2的stem结构与Inception V4的相同。
  2. Inception-resnet-A结构: InceptionV4网络中Inception-A结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。
  3. Inception-resnet-B结构: InceptionV4网络中Inception-B结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。
  4. Inception-resnet-C结构: InceptionV4网络中Inception-C结构的变体,1×1卷积的目的是为了保持主分支与shortcut分支的特征图形状保持完全一致。
  5. Redution-A结构: 与InceptionV4网络中Redution-A结构一致,区别在于卷积核的个数。

    k和l表示卷积个数,不同网络结构的redution-A结构k和l是不同的。

    1. Redution-B结构:

残差模块的缩放(Scaling of the Residuals)

如果单个网络层卷积核数量过多(超过1000),残差网络开始出现不稳定,网络会在训练过程早期便会开始失效—经过几万次训练后,平均池化层之前的层开始只输出0。降低学习率、增加额外的BN层都无法避免这种状况。因此在将shortcut分支加到当前残差块的输出之前,对残差块的输出进行放缩能够稳定训练

通常,将残差放缩因子定在0.1到0.3之间去缩放残差块输出。即使缩放并不是完全必须的,它似乎并不会影响最终的准确率,但是放缩能有益于训练的稳定性。

Inception-ResNet的总体模型结构

下图是原论文给出的关于 Inception-ResNet-V1模型结构的详细示意图:

下图是原论文给出的关于 Inception-ResNet-V2模型结构的详细示意图:

读者注意了,原始论文标注的 Inception-ResNet-V2通道数有一部分是错的,写代码时候对应不上。

两个版本的总体结构相同,具体的Stem、Inception块、Redution块则稍微不同。
Inception-ResNet-V1和 Inception-ResNet-V2在图像分类中分为两部分:backbone部分: 主要由 Inception-resnet模块、Stem模块和池化层(汇聚层)组成,分类器部分:由全连接层组成。


GoogLeNet(Inception-ResNet) Pytorch代码

## Inception-ResNet-V1

卷积层组: 卷积层+BN层+激活函数

# 卷积组: Conv2d+BN+ReLU
class BasicConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0):super(BasicConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding)self.bn = nn.BatchNorm2d(out_channels)self.relu = nn.ReLU(inplace=True)def forward(self, x):x = self.conv(x)x = self.bn(x)x = self.relu(x)return x

Stem模块: 卷积层组+池化层

# Stem:BasicConv2d+MaxPool2d
class Stem(nn.Module):def __init__(self, in_channels):super(Stem, self).__init__()# conv3x3(32 stride2 valid)self.conv1 = BasicConv2d(in_channels, 32, kernel_size=3, stride=2)# conv3*3(32 valid)self.conv2 = BasicConv2d(32, 32, kernel_size=3)# conv3*3(64)self.conv3 = BasicConv2d(32, 64, kernel_size=3, padding=1)# maxpool3*3(stride2 valid)self.maxpool4 = nn.MaxPool2d(kernel_size=3, stride=2)# conv1*1(80)self.conv5 = BasicConv2d(64, 80, kernel_size=1)# conv3*3(192 valid)self.conv6 = BasicConv2d(80, 192, kernel_size=1)# conv3*3(256 stride2 valid)self.conv7 = BasicConv2d(192, 256, kernel_size=3, stride=2)def forward(self, x):x = self.maxpool4(self.conv3(self.conv2(self.conv1(x))))x = self.conv7(self.conv6(self.conv5(x)))return x

Inception_ResNet-A模块: 卷积层组+池化层

# Inception_ResNet_A:BasicConv2d+MaxPool2d
class Inception_ResNet_A(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3red, ch3x3, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext, scale=1.0):super(Inception_ResNet_A, self).__init__()# 缩减指数self.scale = scale# conv1*1(32)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(32)+conv3*3(32)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3red, 1),BasicConv2d(ch3x3red, ch3x3, 3, stride=1, padding=1))# conv1*1(32)+conv3*3(32)+conv3*3(32)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, 3, stride=1, padding=1),BasicConv2d(ch3x3X2_1, ch3x3X2_2, 3, stride=1, padding=1))# conv1*1(256)self.conv = BasicConv2d(ch1x1+ch3x3+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)# 拼接x_res = torch.cat((x0, x1, x2), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)

Inception_ResNet-B模块: 卷积层组+池化层

# Inception_ResNet_B:BasicConv2d+MaxPool2d
class Inception_ResNet_B(nn.Module):def __init__(self, in_channels, ch1x1, ch_red, ch_1, ch_2, ch1x1ext, scale=1.0):super(Inception_ResNet_B, self).__init__()# 缩减指数self.scale = scale# conv1*1(128)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(128)+conv1*7(128)+conv1*7(128)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch_red, 1),BasicConv2d(ch_red, ch_1, (1, 7), stride=1, padding=(0, 3)),BasicConv2d(ch_1, ch_2, (7, 1), stride=1, padding=(3, 0)))# conv1*1(896)self.conv = BasicConv2d(ch1x1+ch_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)

Inception_ResNet-C模块: 卷积层组+池化层

# Inception_ResNet_C:BasicConv2d+MaxPool2d
class Inception_ResNet_C(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext,  scale=1.0, activation=True):super(Inception_ResNet_C, self).__init__()# 缩减指数self.scale = scale# 是否激活self.activation = activation# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(192)+conv1*3(192)+conv3*1(192)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, (1, 3), stride=1, padding=(0, 1)),BasicConv2d(ch3x3X2_1, ch3x3X2_2, (3, 1), stride=1, padding=(1, 0)))# conv1*1(1792)self.conv = BasicConv2d(ch1x1+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)if self.activation:return self.relu(x + self.scale * x_res)return x + self.scale * x_res

redutionA模块: 卷积层组+池化层

# redutionA:BasicConv2d+MaxPool2d
class redutionA(nn.Module):def __init__(self, in_channels, k, l, m, n):super(redutionA, self).__init__()# conv3*3(n stride2 valid)self.branch1 = nn.Sequential(BasicConv2d(in_channels, n, kernel_size=3, stride=2),)# conv1*1(k)+conv3*3(l)+conv3*3(m stride2 valid)self.branch2 = nn.Sequential(BasicConv2d(in_channels, k, kernel_size=1),BasicConv2d(k, l, kernel_size=3, padding=1),BasicConv2d(l, m, kernel_size=3, stride=2))# maxpool3*3(stride2 valid)self.branch3 = nn.Sequential(nn.MaxPool2d(kernel_size=3, stride=2))def forward(self, x):branch1 = self.branch1(x)branch2 = self.branch2(x)branch3 = self.branch3(x)# 拼接outputs = [branch1, branch2, branch3]return torch.cat(outputs, 1)

redutionB模块: 卷积层组+池化层

# redutionB:BasicConv2d+MaxPool2d
class redutionB(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3_1, ch3x3_2, ch3x3_3, ch3x3_4):super(redutionB, self).__init__()# conv1*1(256)+conv3x3(384 stride2 valid)self.branch_0 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_1, 3, stride=2, padding=0))# conv1*1(256)+conv3x3(256 stride2 valid)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_2, 3, stride=2, padding=0),)# conv1*1(256)+conv3x3(256)+conv3x3(256 stride2 valid)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_3, 3, stride=1, padding=1),BasicConv2d(ch3x3_3, ch3x3_4, 3, stride=2, padding=0))# maxpool3*3(stride2 valid)self.branch_3 = nn.MaxPool2d(3, stride=2, padding=0)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)x3 = self.branch_3(x)return torch.cat((x0, x1, x2, x3), dim=1)

Inception-ResNet-V2

Inception-ResNet-V2除了Stem,其他模块在结构上与Inception-ResNet-V1一致。
卷积层组: 卷积层+BN层+激活函数

# 卷积组: Conv2d+BN+ReLU
class BasicConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0):super(BasicConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding)self.bn = nn.BatchNorm2d(out_channels)self.relu = nn.ReLU(inplace=True)def forward(self, x):x = self.conv(x)x = self.bn(x)x = self.relu(x)return x

Stem模块: 卷积层组+池化层

# Stem:BasicConv2d+MaxPool2d
class Stem(nn.Module):def __init__(self, in_channels):super(Stem, self).__init__()# conv3*3(32 stride2 valid)self.conv1 = BasicConv2d(in_channels, 32, kernel_size=3, stride=2)# conv3*3(32 valid)self.conv2 = BasicConv2d(32, 32, kernel_size=3)# conv3*3(64)self.conv3 = BasicConv2d(32, 64, kernel_size=3, padding=1)# maxpool3*3(stride2 valid) & conv3*3(96 stride2 valid)self.maxpool4 = nn.MaxPool2d(kernel_size=3, stride=2)self.conv4 = BasicConv2d(64, 96, kernel_size=3, stride=2)# conv1*1(64)+conv3*3(96 valid)self.conv5_1_1 = BasicConv2d(160, 64, kernel_size=1)self.conv5_1_2 = BasicConv2d(64, 96, kernel_size=3)# conv1*1(64)+conv7*1(64)+conv1*7(64)+conv3*3(96 valid)self.conv5_2_1 = BasicConv2d(160, 64, kernel_size=1)self.conv5_2_2 = BasicConv2d(64, 64, kernel_size=(7, 1), padding=(3, 0))self.conv5_2_3 = BasicConv2d(64, 64, kernel_size=(1, 7), padding=(0, 3))self.conv5_2_4 = BasicConv2d(64, 96, kernel_size=3)# conv3*3(192 valid) & maxpool3*3(stride2 valid)self.conv6 = BasicConv2d(192, 192, kernel_size=3, stride=2)self.maxpool6 = nn.MaxPool2d(kernel_size=3, stride=2)def forward(self, x):x1_1 = self.maxpool4(self.conv3(self.conv2(self.conv1(x))))x1_2 = self.conv4(self.conv3(self.conv2(self.conv1(x))))x1 = torch.cat([x1_1, x1_2], 1)x2_1 = self.conv5_1_2(self.conv5_1_1(x1))x2_2 = self.conv5_2_4(self.conv5_2_3(self.conv5_2_2(self.conv5_2_1(x1))))x2 = torch.cat([x2_1, x2_2], 1)x3_1 = self.conv6(x2)x3_2 = self.maxpool6(x2)x3 = torch.cat([x3_1, x3_2], 1)return x3

Inception_ResNet-A模块: 卷积层组+池化层

# Inception_ResNet_A:BasicConv2d+MaxPool2d
class Inception_ResNet_A(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3red, ch3x3, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext, scale=1.0):super(Inception_ResNet_A, self).__init__()# 缩减指数self.scale = scale# conv1*1(32)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(32)+conv3*3(32)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3red, 1),BasicConv2d(ch3x3red, ch3x3, 3, stride=1, padding=1))# conv1*1(32)+conv3*3(48)+conv3*3(64)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, 3, stride=1, padding=1),BasicConv2d(ch3x3X2_1, ch3x3X2_2, 3, stride=1, padding=1))# conv1*1(384)self.conv = BasicConv2d(ch1x1+ch3x3+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)# 拼接x_res = torch.cat((x0, x1, x2), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)

Inception_ResNet-B模块: 卷积层组+池化层

# Inception_ResNet_B:BasicConv2d+MaxPool2d
class Inception_ResNet_B(nn.Module):def __init__(self, in_channels, ch1x1, ch_red, ch_1, ch_2, ch1x1ext, scale=1.0):super(Inception_ResNet_B, self).__init__()# 缩减指数self.scale = scale# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(128)+conv1*7(160)+conv1*7(192)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch_red, 1),BasicConv2d(ch_red, ch_1, (1, 7), stride=1, padding=(0, 3)),BasicConv2d(ch_1, ch_2, (7, 1), stride=1, padding=(3, 0)))# conv1*1(1154)self.conv = BasicConv2d(ch1x1+ch_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)

Inception_ResNet-C模块: 卷积层组+池化层

# Inception_ResNet_C:BasicConv2d+MaxPool2d
class Inception_ResNet_C(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext,  scale=1.0, activation=True):super(Inception_ResNet_C, self).__init__()# 缩减指数self.scale = scale# 是否激活self.activation = activation# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(192)+conv1*3(224)+conv3*1(256)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, (1, 3), stride=1, padding=(0, 1)),BasicConv2d(ch3x3X2_1, ch3x3X2_2, (3, 1), stride=1, padding=(1, 0)))# conv1*1(2048)self.conv = BasicConv2d(ch1x1+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)if self.activation:return self.relu(x + self.scale * x_res)return x + self.scale * x_res

redutionA模块: 卷积层组+池化层

# redutionA:BasicConv2d+MaxPool2d
class redutionA(nn.Module):def __init__(self, in_channels, k, l, m, n):super(redutionA, self).__init__()# conv3*3(n stride2 valid)self.branch1 = nn.Sequential(BasicConv2d(in_channels, n, kernel_size=3, stride=2),)# conv1*1(k)+conv3*3(l)+conv3*3(m stride2 valid)self.branch2 = nn.Sequential(BasicConv2d(in_channels, k, kernel_size=1),BasicConv2d(k, l, kernel_size=3, padding=1),BasicConv2d(l, m, kernel_size=3, stride=2))# maxpool3*3(stride2 valid)self.branch3 = nn.Sequential(nn.MaxPool2d(kernel_size=3, stride=2))def forward(self, x):branch1 = self.branch1(x)branch2 = self.branch2(x)branch3 = self.branch3(x)# 拼接outputs = [branch1, branch2, branch3]return torch.cat(outputs, 1)

redutionB模块: 卷积层组+池化层

# redutionB:BasicConv2d+MaxPool2d
class redutionB(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3_1, ch3x3_2, ch3x3_3, ch3x3_4):super(redutionB, self).__init__()# conv1*1(256)+conv3x3(384 stride2 valid)self.branch_0 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_1, 3, stride=2, padding=0))# conv1*1(256)+conv3x3(288 stride2 valid)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_2, 3, stride=2, padding=0),)# conv1*1(256)+conv3x3(288)+conv3x3(320 stride2 valid)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_3, 3, stride=1, padding=1),BasicConv2d(ch3x3_3, ch3x3_4, 3, stride=2, padding=0))# maxpool3*3(stride2 valid)self.branch_3 = nn.MaxPool2d(3, stride=2, padding=0)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)x3 = self.branch_3(x)return torch.cat((x0, x1, x2, x3), dim=1)

完整代码

Inception-ResNet的输入图像尺寸是299×299

Inception-ResNet-V1

import torch
import torch.nn as nn
from torchsummary import summary# 卷积组: Conv2d+BN+ReLU
class BasicConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0):super(BasicConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding)self.bn = nn.BatchNorm2d(out_channels)self.relu = nn.ReLU(inplace=True)def forward(self, x):x = self.conv(x)x = self.bn(x)x = self.relu(x)return x# Stem:BasicConv2d+MaxPool2d
class Stem(nn.Module):def __init__(self, in_channels):super(Stem, self).__init__()# conv3x3(32 stride2 valid)self.conv1 = BasicConv2d(in_channels, 32, kernel_size=3, stride=2)# conv3*3(32 valid)self.conv2 = BasicConv2d(32, 32, kernel_size=3)# conv3*3(64)self.conv3 = BasicConv2d(32, 64, kernel_size=3, padding=1)# maxpool3*3(stride2 valid)self.maxpool4 = nn.MaxPool2d(kernel_size=3, stride=2)# conv1*1(80)self.conv5 = BasicConv2d(64, 80, kernel_size=1)# conv3*3(192 valid)self.conv6 = BasicConv2d(80, 192, kernel_size=1)# conv3*3(256 stride2 valid)self.conv7 = BasicConv2d(192, 256, kernel_size=3, stride=2)def forward(self, x):x = self.maxpool4(self.conv3(self.conv2(self.conv1(x))))x = self.conv7(self.conv6(self.conv5(x)))return x# Inception_ResNet_A:BasicConv2d+MaxPool2d
class Inception_ResNet_A(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3red, ch3x3, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext, scale=1.0):super(Inception_ResNet_A, self).__init__()# 缩减指数self.scale = scale# conv1*1(32)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(32)+conv3*3(32)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3red, 1),BasicConv2d(ch3x3red, ch3x3, 3, stride=1, padding=1))# conv1*1(32)+conv3*3(32)+conv3*3(32)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, 3, stride=1, padding=1),BasicConv2d(ch3x3X2_1, ch3x3X2_2, 3, stride=1, padding=1))# conv1*1(256)self.conv = BasicConv2d(ch1x1+ch3x3+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)# 拼接x_res = torch.cat((x0, x1, x2), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)# Inception_ResNet_B:BasicConv2d+MaxPool2d
class Inception_ResNet_B(nn.Module):def __init__(self, in_channels, ch1x1, ch_red, ch_1, ch_2, ch1x1ext, scale=1.0):super(Inception_ResNet_B, self).__init__()# 缩减指数self.scale = scale# conv1*1(128)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(128)+conv1*7(128)+conv1*7(128)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch_red, 1),BasicConv2d(ch_red, ch_1, (1, 7), stride=1, padding=(0, 3)),BasicConv2d(ch_1, ch_2, (7, 1), stride=1, padding=(3, 0)))# conv1*1(896)self.conv = BasicConv2d(ch1x1+ch_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)# Inception_ResNet_C:BasicConv2d+MaxPool2d
class Inception_ResNet_C(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext,  scale=1.0, activation=True):super(Inception_ResNet_C, self).__init__()# 缩减指数self.scale = scale# 是否激活self.activation = activation# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(192)+conv1*3(192)+conv3*1(192)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, (1, 3), stride=1, padding=(0, 1)),BasicConv2d(ch3x3X2_1, ch3x3X2_2, (3, 1), stride=1, padding=(1, 0)))# conv1*1(1792)self.conv = BasicConv2d(ch1x1+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)if self.activation:return self.relu(x + self.scale * x_res)return x + self.scale * x_res# redutionA:BasicConv2d+MaxPool2d
class redutionA(nn.Module):def __init__(self, in_channels, k, l, m, n):super(redutionA, self).__init__()# conv3*3(n stride2 valid)self.branch1 = nn.Sequential(BasicConv2d(in_channels, n, kernel_size=3, stride=2),)# conv1*1(k)+conv3*3(l)+conv3*3(m stride2 valid)self.branch2 = nn.Sequential(BasicConv2d(in_channels, k, kernel_size=1),BasicConv2d(k, l, kernel_size=3, padding=1),BasicConv2d(l, m, kernel_size=3, stride=2))# maxpool3*3(stride2 valid)self.branch3 = nn.Sequential(nn.MaxPool2d(kernel_size=3, stride=2))def forward(self, x):branch1 = self.branch1(x)branch2 = self.branch2(x)branch3 = self.branch3(x)# 拼接outputs = [branch1, branch2, branch3]return torch.cat(outputs, 1)# redutionB:BasicConv2d+MaxPool2d
class redutionB(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3_1, ch3x3_2, ch3x3_3, ch3x3_4):super(redutionB, self).__init__()# conv1*1(256)+conv3x3(384 stride2 valid)self.branch_0 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_1, 3, stride=2, padding=0))# conv1*1(256)+conv3x3(256 stride2 valid)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_2, 3, stride=2, padding=0),)# conv1*1(256)+conv3x3(256)+conv3x3(256 stride2 valid)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_3, 3, stride=1, padding=1),BasicConv2d(ch3x3_3, ch3x3_4, 3, stride=2, padding=0))# maxpool3*3(stride2 valid)self.branch_3 = nn.MaxPool2d(3, stride=2, padding=0)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)x3 = self.branch_3(x)return torch.cat((x0, x1, x2, x3), dim=1)class Inception_ResNetv1(nn.Module):def __init__(self, num_classes = 1000, k=192, l=192, m=256, n=384):super(Inception_ResNetv1, self).__init__()blocks = []blocks.append(Stem(3))for i in range(5):blocks.append(Inception_ResNet_A(256,32, 32, 32, 32, 32, 32, 256, 0.17))blocks.append(redutionA(256, k, l, m, n))for i in range(10):blocks.append(Inception_ResNet_B(896, 128, 128, 128, 128, 896, 0.10))blocks.append(redutionB(896,256, 384, 256, 256, 256))for i in range(4):blocks.append(Inception_ResNet_C(1792,192, 192, 192, 192, 1792, 0.20))blocks.append(Inception_ResNet_C(1792, 192, 192, 192, 192, 1792, activation=False))self.features = nn.Sequential(*blocks)self.conv = BasicConv2d(1792, 1536, 1)self.global_average_pooling = nn.AdaptiveAvgPool2d((1, 1))self.dropout = nn.Dropout(0.8)self.linear = nn.Linear(1536, num_classes)def forward(self, x):x = self.features(x)x = self.conv(x)x = self.global_average_pooling(x)x = x.view(x.size(0), -1)x = self.dropout(x)x = self.linear(x)return xif __name__ == '__main__':device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")model = Inception_ResNetv1().to(device)summary(model, input_size=(3, 229, 229))

summary可以打印网络结构和参数,方便查看搭建好的网络结构。

Inception-ResNet-V2

import torch
import torch.nn as nn
from torchsummary import summary# 卷积组: Conv2d+BN+ReLU
class BasicConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0):super(BasicConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding)self.bn = nn.BatchNorm2d(out_channels)self.relu = nn.ReLU(inplace=True)def forward(self, x):x = self.conv(x)x = self.bn(x)x = self.relu(x)return x# Stem:BasicConv2d+MaxPool2d
class Stem(nn.Module):def __init__(self, in_channels):super(Stem, self).__init__()# conv3*3(32 stride2 valid)self.conv1 = BasicConv2d(in_channels, 32, kernel_size=3, stride=2)# conv3*3(32 valid)self.conv2 = BasicConv2d(32, 32, kernel_size=3)# conv3*3(64)self.conv3 = BasicConv2d(32, 64, kernel_size=3, padding=1)# maxpool3*3(stride2 valid) & conv3*3(96 stride2 valid)self.maxpool4 = nn.MaxPool2d(kernel_size=3, stride=2)self.conv4 = BasicConv2d(64, 96, kernel_size=3, stride=2)# conv1*1(64)+conv3*3(96 valid)self.conv5_1_1 = BasicConv2d(160, 64, kernel_size=1)self.conv5_1_2 = BasicConv2d(64, 96, kernel_size=3)# conv1*1(64)+conv7*1(64)+conv1*7(64)+conv3*3(96 valid)self.conv5_2_1 = BasicConv2d(160, 64, kernel_size=1)self.conv5_2_2 = BasicConv2d(64, 64, kernel_size=(7, 1), padding=(3, 0))self.conv5_2_3 = BasicConv2d(64, 64, kernel_size=(1, 7), padding=(0, 3))self.conv5_2_4 = BasicConv2d(64, 96, kernel_size=3)# conv3*3(192 valid) & maxpool3*3(stride2 valid)self.conv6 = BasicConv2d(192, 192, kernel_size=3, stride=2)self.maxpool6 = nn.MaxPool2d(kernel_size=3, stride=2)def forward(self, x):x1_1 = self.maxpool4(self.conv3(self.conv2(self.conv1(x))))x1_2 = self.conv4(self.conv3(self.conv2(self.conv1(x))))x1 = torch.cat([x1_1, x1_2], 1)x2_1 = self.conv5_1_2(self.conv5_1_1(x1))x2_2 = self.conv5_2_4(self.conv5_2_3(self.conv5_2_2(self.conv5_2_1(x1))))x2 = torch.cat([x2_1, x2_2], 1)x3_1 = self.conv6(x2)x3_2 = self.maxpool6(x2)x3 = torch.cat([x3_1, x3_2], 1)return x3# Inception_ResNet_A:BasicConv2d+MaxPool2d
class Inception_ResNet_A(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3red, ch3x3, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext, scale=1.0):super(Inception_ResNet_A, self).__init__()# 缩减指数self.scale = scale# conv1*1(32)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(32)+conv3*3(32)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3red, 1),BasicConv2d(ch3x3red, ch3x3, 3, stride=1, padding=1))# conv1*1(32)+conv3*3(48)+conv3*3(64)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, 3, stride=1, padding=1),BasicConv2d(ch3x3X2_1, ch3x3X2_2, 3, stride=1, padding=1))# conv1*1(384)self.conv = BasicConv2d(ch1x1+ch3x3+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)# 拼接x_res = torch.cat((x0, x1, x2), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)# Inception_ResNet_B:BasicConv2d+MaxPool2d
class Inception_ResNet_B(nn.Module):def __init__(self, in_channels, ch1x1, ch_red, ch_1, ch_2, ch1x1ext, scale=1.0):super(Inception_ResNet_B, self).__init__()# 缩减指数self.scale = scale# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(128)+conv1*7(160)+conv1*7(192)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch_red, 1),BasicConv2d(ch_red, ch_1, (1, 7), stride=1, padding=(0, 3)),BasicConv2d(ch_1, ch_2, (7, 1), stride=1, padding=(3, 0)))# conv1*1(1154)self.conv = BasicConv2d(ch1x1+ch_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)return self.relu(x + self.scale * x_res)# Inception_ResNet_C:BasicConv2d+MaxPool2d
class Inception_ResNet_C(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3redX2, ch3x3X2_1, ch3x3X2_2, ch1x1ext,  scale=1.0, activation=True):super(Inception_ResNet_C, self).__init__()# 缩减指数self.scale = scale# 是否激活self.activation = activation# conv1*1(192)self.branch_0 = BasicConv2d(in_channels, ch1x1, 1)# conv1*1(192)+conv1*3(224)+conv3*1(256)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch3x3redX2, 1),BasicConv2d(ch3x3redX2, ch3x3X2_1, (1, 3), stride=1, padding=(0, 1)),BasicConv2d(ch3x3X2_1, ch3x3X2_2, (3, 1), stride=1, padding=(1, 0)))# conv1*1(2048)self.conv = BasicConv2d(ch1x1+ch3x3X2_2, ch1x1ext, 1)self.relu = nn.ReLU(inplace=True)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)# 拼接x_res = torch.cat((x0, x1), dim=1)x_res = self.conv(x_res)if self.activation:return self.relu(x + self.scale * x_res)return x + self.scale * x_res# redutionA:BasicConv2d+MaxPool2d
class redutionA(nn.Module):def __init__(self, in_channels, k, l, m, n):super(redutionA, self).__init__()# conv3*3(n stride2 valid)self.branch1 = nn.Sequential(BasicConv2d(in_channels, n, kernel_size=3, stride=2),)# conv1*1(k)+conv3*3(l)+conv3*3(m stride2 valid)self.branch2 = nn.Sequential(BasicConv2d(in_channels, k, kernel_size=1),BasicConv2d(k, l, kernel_size=3, padding=1),BasicConv2d(l, m, kernel_size=3, stride=2))# maxpool3*3(stride2 valid)self.branch3 = nn.Sequential(nn.MaxPool2d(kernel_size=3, stride=2))def forward(self, x):branch1 = self.branch1(x)branch2 = self.branch2(x)branch3 = self.branch3(x)# 拼接outputs = [branch1, branch2, branch3]return torch.cat(outputs, 1)# redutionB:BasicConv2d+MaxPool2d
class redutionB(nn.Module):def __init__(self, in_channels, ch1x1, ch3x3_1, ch3x3_2, ch3x3_3, ch3x3_4):super(redutionB, self).__init__()# conv1*1(256)+conv3x3(384 stride2 valid)self.branch_0 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_1, 3, stride=2, padding=0))# conv1*1(256)+conv3x3(288 stride2 valid)self.branch_1 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_2, 3, stride=2, padding=0),)# conv1*1(256)+conv3x3(288)+conv3x3(320 stride2 valid)self.branch_2 = nn.Sequential(BasicConv2d(in_channels, ch1x1, 1),BasicConv2d(ch1x1, ch3x3_3, 3, stride=1, padding=1),BasicConv2d(ch3x3_3, ch3x3_4, 3, stride=2, padding=0))# maxpool3*3(stride2 valid)self.branch_3 = nn.MaxPool2d(3, stride=2, padding=0)def forward(self, x):x0 = self.branch_0(x)x1 = self.branch_1(x)x2 = self.branch_2(x)x3 = self.branch_3(x)return torch.cat((x0, x1, x2, x3), dim=1)class Inception_ResNetv2(nn.Module):def __init__(self, num_classes = 1000, k=256, l=256, m=384, n=384):super(Inception_ResNetv2, self).__init__()blocks = []blocks.append(Stem(3))for i in range(5):blocks.append(Inception_ResNet_A(384,32, 32, 32, 32, 48, 64, 384, 0.17))blocks.append(redutionA(384, k, l, m, n))for i in range(10):blocks.append(Inception_ResNet_B(1152, 192, 128, 160, 192, 1152, 0.10))blocks.append(redutionB(1152, 256, 384, 288, 288, 320))for i in range(4):blocks.append(Inception_ResNet_C(2144,192, 192, 224, 256, 2144, 0.20))blocks.append(Inception_ResNet_C(2144, 192, 192, 224, 256, 2144, activation=False))self.features = nn.Sequential(*blocks)self.conv = BasicConv2d(2144, 1536, 1)self.global_average_pooling = nn.AdaptiveAvgPool2d((1, 1))self.dropout = nn.Dropout(0.8)self.linear = nn.Linear(1536, num_classes)def forward(self, x):x = self.features(x)x = self.conv(x)x = self.global_average_pooling(x)x = x.view(x.size(0), -1)x = self.dropout(x)x = self.linear(x)return xif __name__ == '__main__':device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")model = Inception_ResNetv2().to(device)summary(model, input_size=(3, 229, 229))

summary可以打印网络结构和参数,方便查看搭建好的网络结构。


总结

尽可能简单、详细的介绍了Inception-ResNet将Inception和ResNet结合的作用和过程,讲解了Inception-ResNet模型的结构和pytorch代码。


https://www.xjx100.cn/news/3091915.html

相关文章

ueditor整合到thinkPHP里

<?phpnamespace app\ueditor\controller;use think\Controller;class Ueditor extends Controller {//首页public function upload(){//header(Access-Control-Allow-Origin: http://www.baidu.com); //设置http://www.baidu.com允许跨域访问//header(Access-Control-Allow…

请求的接口响应状态为已取消的原因

有趣的iframe问题 今天遇到一个问题&#xff0c;当点击了按钮----跳转页面时----F12键点击网络中的状态报了已取消&#xff0c;类型是 document说明是前端页面的问题&#xff0c;如果是xhr那可能是接口的问题。 原本是写了3个iframe,页面刷新的时候请求了第一个iframe,然后就…

第四代智能井盖传感器,实时守护井盖位安全

城市管理中井盖的安全问题始终是一个不容忽视的方面。传统的巡检方式不仅效率低下&#xff0c;无法实现实时监测&#xff0c;而且很难准确掌握井盖的异动状态。因此智能井盖传感器的应用具有重要意义。这种智能传感器可以帮助政府实时掌握井盖的状态&#xff0c;一旦发现异常情…

Vue 优雅的减少请求次数

文章目录 前言Java代码Vue代码改造前改造后前言 当开发一个应用程序或者网站时,我们经常需要从后端获取一些数据来展示给用户。在某些情况下,我们可能需要获取多个属性的值,而不是单独获取每个属性的值。这时,我们可以使用一次请求获取多个属性的值,以减少网络请求的次数…

影视行业如何远程完整快速传输大文件?

影视行业是一个充满创意和协作的领域。在影视制作中&#xff0c;涉及到多个环节和部门&#xff0c;包括编剧、导演、摄影、剪辑、配音、视效等。这些环节和部门通常分布在不同的地点&#xff0c;甚至不同的国家。因此&#xff0c;影视制作过程中需要频繁进行远程传输&#xff0…

深度学习动物识别 - 卷积神经网络 机器视觉 图像识别 计算机竞赛

文章目录 0 前言1 背景2 算法原理2.1 动物识别方法概况2.2 常用的网络模型2.2.1 B-CNN2.2.2 SSD 3 SSD动物目标检测流程4 实现效果5 部分相关代码5.1 数据预处理5.2 构建卷积神经网络5.3 tensorflow计算图可视化5.4 网络模型训练5.5 对猫狗图像进行2分类 6 最后 0 前言 &#…

【图数据库实战】gremlin语法

Gremlin 是 Apache TinkerPop 的图遍历语言。Gremlin 是一种函数式数据流语言&#xff0c;使用户能够简洁地表达对其应用程序属性图的复杂遍历&#xff08;或查询&#xff09;。每个 Gremlin 遍历都由一系列&#xff08;可能嵌套的&#xff09;步骤组成。步骤对数据流执行原子操…

20天拿下华为OD笔试之【DP/贪心】2023B-观看文艺汇演【欧弟算法】全网注释最详细分类最全的华为OD真题题解

文章目录 【DP/贪心】2023B-观看文艺汇演题目描述与示例输入输出示例一输入输出说明 示例二输入输出说明 示例三输入输出 解题思路原始数据处理贪心思想求解问题动态规划求解问题 代码解法一&#xff1a;贪心时空复杂度 解法二&#xff1a;DP时空复杂度 华为OD算法冲刺训练 【D…