Pytorch系列(六):猫狗大战3-MobileNet_V1&V2

---

Pytorch猫狗大战系列：
猫狗大战1-训练和测试自己的数据集
猫狗大战2-AlexNet
猫狗大战3-MobileNet_V1&V2
猫狗大战3-MobileNet_V3

---

TensorFlow 2.0猫狗大战系列
猫狗大战1、制作与读取record数据
猫狗大战2、训练与保存模型

---

一、Mobilenet系列理论整理

看我的另外一篇博客笔记, 传送门
MobileNet相关知识整理

二、Depthwise Separable Convolution和MobileNet_v1 分类网络结构

Depthwise Separable Convolution 可以大大减少卷积计算的参数量, 来看Pytorch的实现
下图中,左边是普通卷积,右边的深度可分离卷积看pytorch代码

def conv_bn(inp, oup, stride):
     return nn.Sequential(
         nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
         nn.BatchNorm2d(oup),
         nn.ReLU(inplace=True)
     )

def conv_dw(inp, oup, stride):
    return nn.Sequential(
        nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
        nn.BatchNorm2d(inp),
        nn.ReLU(inplace=True),

        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU(inplace=True),
    )

再看mobileNet-v1的网络结构
看pytorch实现,还有关于mobilenet的两个超参数, 这里默认为1,不做任何改动

class MobileNetV1(nn.Module):
    def __init__(self, num_classes=2):
        super(MobileNetV1, self).__init__()

        def conv_bn(inp, oup, stride):
            return nn.Sequential(
                nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
                nn.BatchNorm2d(oup),
                nn.ReLU(inplace=True)
            )

        def conv_dw(inp, oup, stride):
            return nn.Sequential(
                nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
                nn.BatchNorm2d(inp),
                nn.ReLU(inplace=True),

                nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
                nn.ReLU(inplace=True),
            )

        self.model = nn.Sequential(
            conv_bn(3, 32, 2),
            conv_dw(32, 64, 1),
            conv_dw(64, 128, 2),
            conv_dw(128, 128, 1),
            conv_dw(128, 256, 2),
            conv_dw(256, 256, 1),
            conv_dw(256, 512, 2),
            conv_dw(512, 512, 1),
            conv_dw(512, 512, 1),
            conv_dw(512, 512, 1),
            conv_dw(512, 512, 1),
            conv_dw(512, 512, 1),
            conv_dw(512, 1024, 2),
            conv_dw(1024, 1024, 1),
            nn.AvgPool2d(7),
        )
        self.fc = nn.Linear(1024, num_classes)

    def forward(self, x):
        x = self.model(x)
        x = x.view(-1, 1024)
        x = self.fc(x)
        return x

在猫狗大战数据集上准确率轻松达到0.98

三、MobileNet_V2

3.1 和MobileNet_v1的区别

转自https://blog.csdn.net/mzpmzk/article/details/82976871

3.2 和ResNet的区别

torchvison里有v2的实现

net = models.MobileNetV2(num_classes=2)

但50次epoch后,准确率只达到了0.78, 虽然还在进一步下降