trick2-mobilenetv1、mobilenetv2、mobilenetv3替换YOLO主干

---

前言

以YOLOV4为例，分别用Mobilenetv1,Mobilenetv2,Mobilenetv3替换YOLOV4主干。

一、YOLOV4主干网络

二、Mobilenetv1,Mobilenetv2,Mobilenetv3构建

1.Mobilenetv1构建（深度可分离卷积）

代码如下（示例）：

import torch
import torch.nn as nn


def conv_bn(inp, oup, stride = 1):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU6(inplace=True)
    )
    
def conv_dw(inp, oup, stride = 1):
    return nn.Sequential(
        # part1
        nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
        nn.BatchNorm2d(inp),
        nn.ReLU6(inplace=True),

        # part2
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU6(inplace=True),
    )

class MobileNetV1(nn.Module):
    def __init__(self):
        super(MobileNetV1, self).__init__()
        self.stage1 = nn.Sequential(
            # 416,416,3 -> 208,208,32
            conv_bn(3, 32, 2),
            # 208,208,32 -> 208,208,64
            conv_dw(32, 64, 1), 

            # 208,208,64 -> 104,104,128
            conv_dw(64, 128, 2),
            conv_dw(128, 128, 1),

            # 104,104,128 -> 52,52,256
            conv_dw(128, 256, 2),
            conv_dw(256, 256, 1), 
        )
            # 52,52,256 -> 26,26,512
        self.stage2 = nn.Sequential(
            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),
        )
            # 26,26,512 -> 13,13,1024
        self.stage3 = nn.Sequential(
            conv_dw(512, 1024, 2),
            conv_dw(1024, 1024, 1),
        )
        self.avg = nn.AdaptiveAvgPool2d((1,1))
        self.fc = nn.Linear(1024, 1000)

    def forward(self, x):
        x = self.stage1(x)
        x = self.stage2(x)
        x = self.stage3(x)
        x = self.avg(x)
        # x = self.model(x)
        x = x.view(-1, 1024)
        x = self.fc(x)
        return x

def mobilenet_v1(pretrained=False, progress=True):
    model = MobileNetV1()
    if pretrained:
        state_dict = torch.load('./model_data/mobilenet_v1_weights.pth')
        model.load_state_dict(state_dict, strict=True)
    return model

if __name__ == "__main__":
    import torch
    from torchsummary import summary

    # 需要使用device来指定网络在GPU还是CPU运行
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    model = mobilenet_v1().to(device)
    summary(model, input_size=(3, 416, 416))

2.Mobilenetv2构建(倒残差结构)

代码如下（示例）：

from torch import nn
from torchvision.models.utils import load_state_dict_from_url

model_urls = {
    'mobilenet_v2': 'https://download.pytorch.org/models/mobilenet_v2-b0353104.pth',
}


def _make_divisible(v, divisor, min_value=None):
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v

class ConvBNReLU(nn.Sequential):
    def __init__(self, in_planes, out_planes, kernel_size=3, stride=1, groups=1):
        padding = (kernel_size - 1) // 2
        super(ConvBNReLU, self).__init__(
            nn.Conv2d(in_planes, out_planes, kernel_size, stride, padding, groups=groups, bias=False),
            nn.BatchNorm2d(out_planes),
            nn.ReLU6(inplace=True)
        )

class InvertedResidual(nn.Module):
    def __init__(self, inp, oup, stride, expand_ratio):
        super(InvertedResidual, self).__init__()
        self.stride = stride
        assert stride in [1, 2]

        hidden_dim = int(round(inp * expand_ratio))
        self.use_res_connect = self.stride == 1 and inp == oup

        layers = []
        if expand_ratio != 1:
            layers.append(ConvBNReLU(inp, hidden_dim, kernel_size=1))
            
        layers.extend([
            ConvBNReLU(hidden_dim, hidden_dim, stride=stride, groups=hidden_dim),

            nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
            nn.BatchNorm2d(oup), 
        ])
        self.conv = nn.Sequential(*layers)

    def forward(self, x):
        if self.use_res_connect:
            return x + self.conv(x)
        else:
            return self.conv(x)


class MobileNetV2(nn.Module):
    def __init__(self, num_classes=1000, width_mult=1.0, inverted_residual_setting=None, round_nearest=8):
        super(MobileNetV2, self).__init__()
        block = InvertedResidual
        input_channel = 32
        last_channel = 1280

        if inverted_residual_setting is None:
            inverted_residual_setting = [
                # t, c, n, s
                # 208,208,32 -> 208,208,16
                [1, 16, 1, 1],
                # 208,208,16 -> 104,104,24
                [6, 24, 2, 2],
                # 104,104,24 -> 52,52,32
                [6, 32, 3, 2],

                # 52,52,32 -> 26,26,64
                [6, 64, 4, 2],
                # 26,26,64 -> 26,26,96
                [6, 96, 3, 1],
                
                # 26,26,96 -> 13,13,160
                [6, 160, 3, 2],
                # 13,13,160 -> 13,13,320
                [6, 320, 1, 1],
            ]

        if len(inverted_residual_setting) == 0 or len(inverted_residual_setting[0]) != 4:
            raise ValueError("inverted_residual_setting should be non-empty "
                             "or a 4-element list, got {}".format(inverted_residual_setting))

        input_channel = _make_divisible(input_channel * width_mult, round_nearest)
        self.last_channel = _make_divisible(last_channel * max(1.0, width_mult), round_nearest)

        # 416,416,3 -> 208,208,32
        features = [ConvBNReLU(3, input_channel, stride=2)]

        for t, c, n, s in inverted_residual_setting:
            output_channel = _make_divisible(c * width_mult, round_nearest)
            for i in range(n):
                stride = s if i == 0 else 1
                features.append(block(input_channel, output_channel, stride, expand_ratio=t))
                input_channel = output_channel

        features.append(ConvBNReLU(input_channel, self.last_channel, kernel_size=1))
        self.features = nn.Sequential(*features)

        self.classifier = nn.Sequential(
            nn.Dropout(0.2),
            nn.Linear(self.last_channel, num_classes),
        )

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out')
                if m.bias is not None:
                    nn.init.zeros_(m.bias)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.ones_(m.weight)
                nn.init.zeros_(m.bias)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                nn.init.zeros_(m.bias)

    def forward(self, x):
        x = self.features(x)
        x = x.mean([2, 3])
        x = self.classifier(x)
        return x

def mobilenet_v2(pretrained=False, progress=True):
    model = MobileNetV2()
    if pretrained:
        state_dict = load_state_dict_from_url(model_urls['mobilenet_v2'], model_dir="model_data",
                                              progress=progress)
        model.load_state_dict(state_dict)

    return model

if __name__ == "__main__":
    print(mobilenet_v2())

3.Mobilenetv3构建（bneck结构）

代码如下（示例）：

import math

import torch
import torch.nn as nn


def _make_divisible(v, divisor, min_value=None):
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    # Make sure that round down does not go down by more than 10%.
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v

class h_sigmoid(nn.Module):
    def __init__(self, inplace=True):
        super(h_sigmoid, self).__init__()
        self.relu = nn.ReLU6(inplace=inplace)

    def forward(self, x):
        return self.relu(x + 3) / 6


class h_swish(nn.Module):
    def __init__(self, inplace=True):
        super(h_swish, self).__init__()
        self.sigmoid = h_sigmoid(inplace=inplace)

    def forward(self, x):
        return x * self.sigmoid(x)


class SELayer(nn.Module):
    def __init__(self, channel, reduction=4):
        super(SELayer, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
                nn.Linear(channel, _make_divisible(channel // reduction, 8)),
                nn.ReLU(inplace=True),
                nn.Linear(_make_divisible(channel // reduction, 8), channel),
                h_sigmoid()
        )

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y


def conv_3x3_bn(inp, oup, stride):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
        nn.BatchNorm2d(oup),
        h_swish()
    )


def conv_1x1_bn(inp, oup):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        h_swish()
    )


class InvertedResidual(nn.Module):
    def __init__(self, inp, hidden_dim, oup, kernel_size, stride, use_se, use_hs):
        super(InvertedResidual, self).__init__()
        assert stride in [1, 2]

        self.identity = stride == 1 and inp == oup

        if inp == hidden_dim:
            self.conv = nn.Sequential(
                # dw
                nn.Conv2d(hidden_dim, hidden_dim, kernel_size, stride, (kernel_size - 1) // 2, groups=hidden_dim, bias=False),
                nn.BatchNorm2d(hidden_dim),
                h_swish() if use_hs else nn.ReLU(inplace=True),
                # Squeeze-and-Excite
                SELayer(hidden_dim) if use_se else nn.Identity(),
                # pw-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )
        else:
            self.conv = nn.Sequential(

                # pw
                nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False),
                nn.BatchNorm2d(hidden_dim),
                h_swish() if use_hs else nn.ReLU(inplace=True),

                # dw
                nn.Conv2d(hidden_dim, hidden_dim, kernel_size, stride, (kernel_size - 1) // 2, groups=hidden_dim, bias=False),
                nn.BatchNorm2d(hidden_dim),

                # Squeeze-and-Excite
                SELayer(hidden_dim) if use_se else nn.Identity(),

                h_swish() if use_hs else nn.ReLU(inplace=True),

                # pw-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )

    def forward(self, x):
        if self.identity:
            return x + self.conv(x)
        else:
            return self.conv(x)


class MobileNetV3(nn.Module):
    def __init__(self, num_classes=1000, width_mult=1.):
        super(MobileNetV3, self).__init__()
        # setting of inverted residual blocks
        self.cfgs = [
            #`   k,   t,   c, SE,HS,s 
                # 208,208,16 -> 208,208,16
                [3,   1,  16, 0, 0, 1],

                # 208,208,16 -> 104,104,24
                [3,   4,  24, 0, 0, 2],
                [3,   3,  24, 0, 0, 1],

                # 104,104,24 -> 52,52,40
                [5,   3,  40, 1, 0, 2],
                [5,   3,  40, 1, 0, 1],
                [5,   3,  40, 1, 0, 1],

                # 52,52,40 -> 26,26,80
                [3,   6,  80, 0, 1, 2],
                [3, 2.5,  80, 0, 1, 1],
                [3, 2.3,  80, 0, 1, 1],
                [3, 2.3,  80, 0, 1, 1],

                # 26,26,80 -> 26,26,112
                [3,   6, 112, 1, 1, 1],
                [3,   6, 112, 1, 1, 1],

                # 26,26,112 -> 13,13,160
                [5,   6, 160, 1, 1, 2],
                [5,   6, 160, 1, 1, 1],
                [5,   6, 160, 1, 1, 1]
        ]

        input_channel = _make_divisible(16 * width_mult, 8)
        # 416,416,3 -> 208,208,16
        layers = [conv_3x3_bn(3, input_channel, 2)]

        block = InvertedResidual
        for k, t, c, use_se, use_hs, s in self.cfgs:
            output_channel = _make_divisible(c * width_mult, 8)
            exp_size = _make_divisible(input_channel * t, 8)
            layers.append(block(input_channel, exp_size, output_channel, k, s, use_se, use_hs))
            input_channel = output_channel
        self.features = nn.Sequential(*layers)

        self.conv = conv_1x1_bn(input_channel, exp_size)
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        output_channel = _make_divisible(1280 * width_mult, 8) if width_mult > 1.0 else 1280
        self.classifier = nn.Sequential(
            nn.Linear(exp_size, output_channel),
            h_swish(),
            nn.Dropout(0.2),
            nn.Linear(output_channel, num_classes),
        )

        self._initialize_weights()

    def forward(self, x):
        x = self.features(x)
        x = self.conv(x)
        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.classifier(x)
        return x

    def _initialize_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2. / n))
                if m.bias is not None:
                    m.bias.data.zero_()
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()
            elif isinstance(m, nn.Linear):
                n = m.weight.size(1)
                m.weight.data.normal_(0, 0.01)
                m.bias.data.zero_()

def mobilenet_v3(pretrained=False, **kwargs):
    model = MobileNetV3(**kwargs)
    if pretrained:
        state_dict = torch.load('./model_data/mobilenetv3-large-1cd25616.pth')
        model.load_state_dict(state_dict, strict=True)
    return model

三、获得主干网络中的三个有效特征层（yolo4.py)

1.导入库

import torch
import torch.nn as nn
from collections import OrderedDict
from nets.mobilenet_v1 import mobilenet_v1
from nets.mobilenet_v2 import mobilenet_v2
from nets.mobilenet_v3 import mobilenet_v3

2.Mobilenetv1

class MobileNetV1(nn.Module):
    def __init__(self, pretrained = False):
        super(MobileNetV1, self).__init__()
        self.model = mobilenet_v1(pretrained=pretrained)

    def forward(self, x):
        out3 = self.model.stage1(x)
        out4 = self.model.stage2(out3)
        out5 = self.model.stage3(out4)
        return out3, out4, out5

3.Mobilenetv2

class MobileNetV2(nn.Module):
    def __init__(self, pretrained = False):
        super(MobileNetV2, self).__init__()
        self.model = mobilenet_v2(pretrained=pretrained)

    def forward(self, x):
        out3 = self.model.features[:7](x)
        out4 = self.model.features[7:14](out3)
        out5 = self.model.features[14:18](out4)
        return out3, out4, out5

4.Mobilenetv3

class MobileNetV3(nn.Module):
    def __init__(self, pretrained = False):
        super(MobileNetV3, self).__init__()
        self.model = mobilenet_v3(pretrained=pretrained)

    def forward(self, x):
        out3 = self.model.features[:7](x)
        out4 = self.model.features[7:13](out3)
        out5 = self.model.features[13:16](out4)
        return out3, out4, out5

四、YOLOV4主干特征提取网络的替换（yolo4.py)

1.在YoloBody定义backbone=“mobilenetv2”

class YoloBody(nn.Module):
    def __init__(self, anchors_mask, num_classes, backbone="mobilenetv2", pretrained=False):

2.判断backbone是否是上面预先定义的类别

 if backbone == "mobilenetv1":
            #---------------------------------------------------#   
            #   52,52,256；26,26,512；13,13,1024
            #---------------------------------------------------#
            self.backbone   = MobileNetV1(pretrained=pretrained)
            in_filters      = [256, 512, 1024]
        elif backbone == "mobilenetv2":
            #---------------------------------------------------#   
            #   52,52,32；26,26,92；13,13,320
            #---------------------------------------------------#
            self.backbone   = MobileNetV2(pretrained=pretrained)
            in_filters      = [32, 96, 320]
        elif backbone == "mobilenetv3":
            #---------------------------------------------------#   
            #   52,52,40；26,26,112；13,13,160
            #---------------------------------------------------#
            self.backbone   = MobileNetV3(pretrained=pretrained)
            in_filters      = [40, 112, 160]

3.关于通道不匹配错误的问题，需要修改卷积使用的输入通道数。

3.1 首先定义三个有效特征层的输出通道数是多少。

52,52,256；26,26,512；13,13,1024（mobilenetv1）
52,52,32；26,26,92；13,13,320（mobilenetv2）
52,52,40；26,26,112；13,13,160（mobilenetv3）

in_filters      = [256, 512, 1024]
in_filters      = [32, 96, 320]
in_filters      = [40, 112, 160]

3.2 然后需要修改卷积使用的输入通道数

    self.conv1           = make_three_conv([512, 1024], in_filters[2]) #1024->in_filters[2]
    self.conv_for_P4     = conv2d(in_filters[1], 256,1)                #512->in_filters[1]
    self.conv_for_P3     = conv2d(in_filters[0], 128,1)                #256->in_filters[0]

4.参数量（大量的参数是在PAnet里面）

五、PAnet加强特征提取网络修改，使参数量更小（yolo4.py)

5.1 思路：PAnet大部分使用了3x3卷积，而在mobilenetv1里面提到过可以将深度可分离卷积替换3x3卷积，即可实现参数量的大幅度缩小。将下面深度可分离卷积用在yolo4.py中。

def conv_dw(inp, oup, stride = 1):
    return nn.Sequential(
        # part1
        nn.Conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
        nn.BatchNorm2d(inp),
        nn.ReLU6(inplace=True),

        # part2
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU6(inplace=True),
    )

5.2 在三次卷积块和五次卷积块以及yolo-head都会用到3x3卷积，用深度可分离卷积进行替换。用下面的方式全部进行替换。

conv2d(filters_list[0],filters_list[1],3)修改为下面的代码块：
conv_dw(filters_list[0], filters_list[1])

5.3 参数量变化

六、训练参数详解（train.py）

6.1 backbone

backbone        = "mobilenetv1"

6.2 model_path(要和backbone相对应，比如采用主干是mobilenetv1,对应的权值文件也是mobilenetv1，即根骨不同主干和权值）

model_path      = 'model_data/yolov4_mobilenet_v1_voc.pth'

七、利用训练好的模型进行预测（predict.py）,在yolo.py文件中更改三个地方：model_path(训练好的权值文件logs),classes_path(类别文件),backbone(与训练好的权值文件logs主干特征提取网络相对应).

        "model_path"        : 'model_data/yolov4_mobilenet_v1_voc.pth',
        "classes_path"      : 'model_data/voc_classes.txt',
        "backbone"          : 'mobilenetv1',

总结

完成了主干的修改和PAnet的修改，主干部分就是将yolo主干替换成mobilenet，PAnet部分是利用mobilenetv1的思想，利用深度可分离卷积3x3卷积＋1x1卷积替换普通的卷积块。根据这个思想可以进一步减少yolo的参数。