参考大佬的图片进行解读，Mobilenetv2中的block如下图所示
Mobilenetv3中的block如下图所示
可以发现，Mobilenetv3的block中加入了SE模块，更换了激活函数 。
SE模块下一节讲。
此处更新的激活函数在图中用NL(非线性)统一表示，因为用到的激活函数不一样，主要有hardswish、relu两种。
最后那个1x1降维投影层用的是线性激活(f(x)=x)，也可以理解为没用激活。
SE(Squeeze-and-Excitation) 模块类似于一个注意力模块，以在Mobilenetv3中的应用为例进行理解，如下图所示。
ReLu6激活函数如下图所示，相当于加了个最大值6进行限制。
hardswish激活函数如下图所示，相当于分成3段进行限制。
采用hardswish，计算速度相对较快，对量化过程友好
作者针对不同需求，通过NAS得到两种结构，一个是MobilenetV3-Large，结构如下图：
图中部分参数解释：
另一个是MobilenetV3-Small，结构如下图：
直接看代码注释即可，可运行
from typing import Callable, List, Optional
import torch
from torch import nn, Tensor
from torch.nn import functional as F
from functools import partial
# ------------------------------------------------------#
# 这个函数的目的是确保Channel个数能被8整除。
# 离它最近的8的倍数
# 很多嵌入式设备做优化时都采用这个准则
# ------------------------------------------------------#
def _make_divisible(ch, divisor=8, min_ch=None):
if min_ch is None:
min_ch = divisor
# int(v + divisor / 2) // divisor * divisor：四舍五入到8
new_ch = max(min_ch, int(ch + divisor / 2) // divisor * divisor)
# Make sure that round down does not go down by more than 10%.
if new_ch < 0.9 * ch:
new_ch += divisor
return new_ch
# -------------------------------------------------------------#
# Conv+BN+Acti经常会用到，组在一起
# -------------------------------------------------------------#
class ConvBNActivation(nn.Sequential):
def __init__(self,
in_planes: int,
out_planes: int,
kernel_size: int = 3,
stride: int = 1,
groups: int = 1,
norm_layer: Optional[Callable[..., nn.Module]] = None, # 卷积后的BN层
activation_layer: Optional[Callable[..., nn.Module]] = None): # 激活函数
padding = (kernel_size - 1) // 2
if norm_layer is None: # 没有传入，就默认使用BN
norm_layer = nn.BatchNorm2d
if activation_layer is None:
activation_layer = nn.ReLU6
super(ConvBNActivation, self).__init__(nn.Conv2d(in_channels=in_planes,
out_channels=out_planes,
kernel_size=kernel_size,
stride=stride,
padding=padding,
groups=groups,
bias=False), # 后面会用到BN层，故不使用bias
norm_layer(out_planes),
activation_layer(inplace=True))
# ------------------------------------------------------#
# 注意力模块：SE模块
# 就是两个FC层，节点个数、激活函数要注意要注意
# ------------------------------------------------------#
class SqueezeExcitation(nn.Module):
# squeeze_factor: int = 4：第一个FC层节点个数是输入特征矩阵的1/4
def __init__(self, input_c: int, squeeze_factor: int = 4):
super(SqueezeExcitation, self).__init__()
# 第一个FC层节点个数，也要是8的整数倍
squeeze_c = _make_divisible(input_c // squeeze_factor, 8)
# 通过卷积核大小为1x1的卷积替代FC层，作用相同
self.fc1 = nn.Conv2d(input_c, squeeze_c, 1)
self.fc2 = nn.Conv2d(squeeze_c, input_c, 1)
def forward(self, x: Tensor) -> Tensor:
# x有很多channel，通过output_size=(1, 1)实现每个channel变成1个数字
scale = F.adaptive_avg_pool2d(x, output_size=(1, 1))
scale = self.fc1(scale)
scale = F.relu(scale, inplace=True)
scale = self.fc2(scale)
# 此处的scale就是第二个FC层输出的数据
scale = F.hardsigmoid(scale, inplace=True)
return scale * x # 和原输入相乘，得到SE模块的输出
# ------------------------------------------------------#
# InvertedResidualConfig是参数配置文件
# ------------------------------------------------------#
class InvertedResidualConfig:
def __init__(self,
input_c: int,
kernel: int,
expanded_c: int, # bottleneck中的第一层1x1卷积升维，维度升到多少
out_c: int,
use_se: bool,
activation: str,
stride: int,
width_multi: float): # 和mobilenetv2中倍率因子相同，通过它得到每一层channels个数和基线的区别
self.input_c = self.adjust_channels(input_c, width_multi) # 倍率因子用在这儿了
self.kernel = kernel
self.expanded_c = self.adjust_channels(expanded_c, width_multi)
self.out_c = self.adjust_channels(out_c, width_multi)
self.use_se = use_se
# activation == "HS"，则self.use_hs==True
self.use_hs = activation == "HS" # whether using h-swish activation
self.stride = stride
# 静态方法
@staticmethod
def adjust_channels(channels: int, width_multi: float):
return _make_divisible(channels * width_multi, 8)
class InvertedResidual(nn.Module):
def __init__(self,
cnf: InvertedResidualConfig, # cnf是个config文件，对应的格式就是上面介绍的InvertedResidualConfig类
norm_layer: Callable[..., nn.Module]):
super(InvertedResidual, self).__init__()
if cnf.stride not in [1, 2]:
raise ValueError("illegal stride value.")
# 是否使用shortcut连接
self.use_res_connect = (cnf.stride == 1 and cnf.input_c == cnf.out_c)
layers: List[nn.Module] = [] # 定义一个空列表，里面元素类型为nn.module
activation_layer = nn.Hardswish if cnf.use_hs else nn.ReLU
# expand
if cnf.expanded_c != cnf.input_c: # 第一个bottleneck没有这个1x1卷积，故有这个if哦安短
layers.append(ConvBNActivation(cnf.input_c,
cnf.expanded_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=activation_layer))
# depthwise
layers.append(ConvBNActivation(cnf.expanded_c, # 上一层1x1输出通道数为cnf.expanded_c
cnf.expanded_c,
kernel_size=cnf.kernel,
stride=cnf.stride,
groups=cnf.expanded_c, # DW卷积
norm_layer=norm_layer,
activation_layer=activation_layer))
if cnf.use_se: # 是否使用se模块，只需要传入个input_channel
layers.append(SqueezeExcitation(cnf.expanded_c))
# project 降维1x1卷积层
layers.append(ConvBNActivation(cnf.expanded_c,
cnf.out_c,
kernel_size=1,
norm_layer=norm_layer,
# nn.Identity是一个线性激活，没进行任何处理
# 内部实现：直接return input
activation_layer=nn.Identity))
self.block = nn.Sequential(*layers)
self.out_channels = cnf.out_c
self.is_strided = cnf.stride > 1
def forward(self, x: Tensor) -> Tensor:
result = self.block(x)
if self.use_res_connect:
result += x
return result
# 继承来自nn.module类
class MobileNetV3(nn.Module):
def __init__(self,
inverted_residual_setting: List[InvertedResidualConfig], # 参数设置列表，列表里面每个元素类型是上面定义的那个类的形式
last_channel: int, # 倒数第二层channel个数
num_classes: int = 1000, # 需要分类的类别数
block: Optional[Callable[..., nn.Module]] = None,
norm_layer: Optional[Callable[..., nn.Module]] = None):
super(MobileNetV3, self).__init__()
if not inverted_residual_setting:
raise ValueError("The inverted_residual_setting should not be empty.")
elif not (isinstance(inverted_residual_setting, List) and
all([isinstance(s, InvertedResidualConfig) for s in inverted_residual_setting])):
raise TypeError("The inverted_residual_setting should be List[InvertedResidualConfig]")
if block is None:
block = InvertedResidual
# 将norm_layer设置为BN
# partial()给输入函数BN指定默认参数，简化之后的函数参数量
if norm_layer is None:
norm_layer = partial(nn.BatchNorm2d, eps=0.001, momentum=0.01)
layers: List[nn.Module] = []
# building first layer 就是普通的conv
firstconv_output_c = inverted_residual_setting[0].input_c
layers.append(ConvBNActivation(3,
firstconv_output_c,
kernel_size=3,
stride=2,
norm_layer=norm_layer,
activation_layer=nn.Hardswish))
# building inverted residual blocks
for cnf in inverted_residual_setting:
layers.append(block(cnf, norm_layer))
# building last several layers
lastconv_input_c = inverted_residual_setting[-1].out_c
lastconv_output_c = 6 * lastconv_input_c # small：96->576; Large:160->960
layers.append(ConvBNActivation(lastconv_input_c,
lastconv_output_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=nn.Hardswish))
self.features = nn.Sequential(*layers)
self.avgpool = nn.AdaptiveAvgPool2d(1)
self.classifier = nn.Sequential(nn.Linear(lastconv_output_c, last_channel),
nn.Hardswish(inplace=True),
nn.Dropout(p=0.2, inplace=True),
nn.Linear(last_channel, num_classes))
# initial weights
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.GroupNorm)):
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_impl(self, x: Tensor) -> Tensor:
x = self.features(x)
x = self.avgpool(x) # 到这后面不再需要高和宽的维度了
x = torch.flatten(x, 1) # 故进行展平处理
x = self.classifier(x)
return x
def forward(self, x: Tensor) -> Tensor:
return self._forward_impl(x)
def mobilenet_v3_large(num_classes: int = 1000,
reduced_tail: bool = False) -> MobileNetV3:
"""
Constructs a large MobileNetV3 architecture from
"Searching for MobileNetV3" <https://arxiv.org/abs/1905.02244>.
weights_link:
https://download.pytorch.org/models/mobilenet_v3_large-8738ca79.pth
Args:
num_classes (int): number of classes
reduced_tail (bool): 需要的话, 设为True, 可以进一步减小网络
If True, reduces the channel counts of all feature layers
between C4 and C5 by 2. It is used to reduce the channel redundancy in the
backbone for Detection and Segmentation.
"""
width_multi = 1.0 # 调整channel个数，默认1.0
bneck_conf = partial(InvertedResidualConfig, width_multi=width_multi) # partial()给输入函数指定默认参数
# 给类里的方法传入参数 有了上面一行，这行有必要吗？
adjust_channels = partial(InvertedResidualConfig.adjust_channels, width_multi=width_multi)
reduce_divider = 2 if reduced_tail else 1
inverted_residual_setting = [
# input_c, kernel, expanded_c, out_c, use_se, activation, stride
bneck_conf(16, 3, 16, 16, False, "RE", 1),
bneck_conf(16, 3, 64, 24, False, "RE", 2), # C1
bneck_conf(24, 3, 72, 24, False, "RE", 1),
bneck_conf(24, 5, 72, 40, True, "RE", 2), # C2
bneck_conf(40, 5, 120, 40, True, "RE", 1),
bneck_conf(40, 5, 120, 40, True, "RE", 1),
bneck_conf(40, 3, 240, 80, False, "HS", 2), # C3
bneck_conf(80, 3, 200, 80, False, "HS", 1),
bneck_conf(80, 3, 184, 80, False, "HS", 1),
bneck_conf(80, 3, 184, 80, False, "HS", 1),
bneck_conf(80, 3, 480, 112, True, "HS", 1),
bneck_conf(112, 3, 672, 112, True, "HS", 1),
bneck_conf(112, 5, 672, 160 // reduce_divider, True, "HS", 2), # C4
bneck_conf(160 // reduce_divider, 5, 960 // reduce_divider, 160 // reduce_divider, True, "HS", 1),
bneck_conf(160 // reduce_divider, 5, 960 // reduce_divider, 160 // reduce_divider, True, "HS", 1),
]
last_channel = adjust_channels(1280 // reduce_divider) # C5 # 倒数第二个全连接层节点个数
return MobileNetV3(inverted_residual_setting=inverted_residual_setting,
last_channel=last_channel,
num_classes=num_classes)
def mobilenet_v3_small(num_classes: int = 1000,
reduced_tail: bool = False) -> MobileNetV3:
"""
Constructs a large MobileNetV3 architecture from
"Searching for MobileNetV3" <https://arxiv.org/abs/1905.02244>.
weights_link:
https://download.pytorch.org/models/mobilenet_v3_small-047dcff4.pth
Args:
num_classes (int): number of classes
reduced_tail (bool): If True, reduces the channel counts of all feature layers
between C4 and C5 by 2. It is used to reduce the channel redundancy in the
backbone for Detection and Segmentation.
"""
width_multi = 1.0
bneck_conf = partial(InvertedResidualConfig, width_multi=width_multi)
adjust_channels = partial(InvertedResidualConfig.adjust_channels, width_multi=width_multi)
reduce_divider = 2 if reduced_tail else 1
inverted_residual_setting = [
# input_c, kernel, expanded_c, out_c, use_se, activation, stride
bneck_conf(16, 3, 16, 16, True, "RE", 2), # C1
bneck_conf(16, 3, 72, 24, False, "RE", 2), # C2
bneck_conf(24, 3, 88, 24, False, "RE", 1),
bneck_conf(24, 5, 96, 40, True, "HS", 2), # C3
bneck_conf(40, 5, 240, 40, True, "HS", 1),
bneck_conf(40, 5, 240, 40, True, "HS", 1),
bneck_conf(40, 5, 120, 48, True, "HS", 1),
bneck_conf(48, 5, 144, 48, True, "HS", 1),
bneck_conf(48, 5, 288, 96 // reduce_divider, True, "HS", 2), # C4
bneck_conf(96 // reduce_divider, 5, 576 // reduce_divider, 96 // reduce_divider, True, "HS", 1),
bneck_conf(96 // reduce_divider, 5, 576 // reduce_divider, 96 // reduce_divider, True, "HS", 1)
]
last_channel = adjust_channels(1024 // reduce_divider) # C5
return MobileNetV3(inverted_residual_setting=inverted_residual_setting,
last_channel=last_channel,
num_classes=num_classes)
if __name__ == "__main__":
model = mobilenet_v3_small()
print(model)
from torchsummaryX import summary
summary(model, torch.randn(1,3,224,224))
输出：
MobileNetV3(
(features): Sequential(
(0): ConvBNActivation(
(0): Conv2d(3, 16, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(1): BatchNorm2d(16, eps=0.001, momentum=0.01, affine=True, track_running_stats=True)
(2): Hardswish()
)
(1): InvertedResidual(
...
(classifier): Sequential(
(0): Linear(in_features=576, out_features=1024, bias=True)
(1): Hardswish()
(2): Dropout(p=0.2, inplace=True)
(3): Linear(in_features=1024, out_features=1000, bias=True)
)
)
================================================================================================
Kernel Shape Output Shape \
Layer
0_features.0.Conv2d_0 [3, 16, 3, 3] [1, 16, 112, 112]
1_features.0.BatchNorm2d_1 [16] [1, 16, 112, 112]
...
123_classifier.Dropout_2 - -
124_classifier.Linear_3 1.025M 1.024M
------------------------------------------------------------------------------------------------
Totals
Total params 2.542856M
Trainable params 2.542856M
Non-trainable params 0.0
Mult-Adds 56.516456M
https://www.bilibili.com/video/BV1GK4y1p7uE/?spm_id_from=333.788
https://blog.csdn.net/m0_48742971/article/details/123438626
https://www.bilibili.com/video/BV1zT4y1P7pd/?spm_id_from=333.788