resnet系列模型
resnet系列结构图
resnet的网络构成都是1个卷积+4个残差+1个全连接网络。黄框为resnet50的结构，50 = 1+（3+4+6+3）*3+1，其中3、4、6、3的意思是有3个这样的残差块。。。
两种残差块
BasicBlock BottleBlock
残差块有2种，左侧的BasicBlock适用于较浅的resnet18及resnet34，右侧的Bottleneck适用于较深的resnet50及以上。
为什么要设计2种结构，我的理解：通过1*1减少通道数，从而减少参数量，节约显存，构建更深的网络。
对图中的256、64不必太过在意，不是每层都是一样的，以系列图及代码为准。
BasicBlock中的3*3都是下面这个
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
padding=1, bias=False)对BasicBlock，结构为：3*3卷积——块归一化——Relu激活——3*3卷积——块归一化——与输入相加（这一步可能涉及将输入下采样后相加）——Relu激活
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = nn.BatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.BatchNorm2d(planes)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out对BottleBlock，结构为：1*1卷积——块归一化——Relu激活——3*3卷积——块归一化——Relu激活——1*1卷积——块归一化——与输入相加（这一步可能涉及将输入下采样后相加）——Relu激活
在BottleBlock中两个1*1的卷积核都改变了输入输出通道数。
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(Bottleneck, self).__init__()
# 如：256->64
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) # inplanes == 4*planes 见上图可知，输入通道数是输出通道数量的4倍
self.bn1 = nn.BatchNorm2d(planes)
# 如 64->64
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
# 如：64->256
self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False) # 输出通道数是输入通道数量的4倍
self.bn3 = nn.BatchNorm2d(planes * 4)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return outResNet类
class ResNet(nn.Module):
def __init__(self, block, layers, num_classes=1000):
self.inplanes = 64 # 经过开始一个卷积块后，在4个块前，通道数由rgb3通道变为64
super(ResNet, self).__init__()
# 开始：1个卷积层
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False) # resnet第一个卷积将输入3通道图像转化为64维
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=0, ceil_mode=True) # 减小特征图尺寸，112->56
# 中间：4个块，每个块包含多层残差block
# stride 只有第2、3、4块的第一个block才为2，通过卷积下采样，实现特征图更小，通道更多
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
# 最后：1个全连接层
self.avgpool = nn.AvgPool2d(7)
self.fc = nn.Linear(512 * block.expansion, num_classes) # block.expansion为1、4，全连接层的输入为512、2048通道
for m in self.modules():
# 卷积层初始化
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))
# bn初始化
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def _make_layer(self, block, planes, blocks, stride=1):
# block：指定basicblock或bottleblock
# planes：block 第一个卷积的输出通道数
# block：这个块有几个残差结构
# stride：值为1或2，改变这个块第一个残差结构第一层的特征图尺寸
downsample = None
# stride != 1：第2、3、4个块
# self.inplanes != planes * block.expansion：上层的输出不是第一个残差结构第一层的输出通道数的block.expansion倍,涵盖了所有使用了bottleblock的块，和resnet18、resnet34的2、3、4个块
# 下采样是通过卷积实现的，目的是改变通道数量或通过stride改变尺寸，叫下采样不是特别恰当
# resnet系列图中一共有4*5=20个块，只有resnet18和resnet34的第一个块（共2个）没做下采样，其他18个块都在第一个残差结构里做下采样了（即用了botttleblock的18、101、152系列，及18、34的2、3、4块）
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample)) # 所有块的第一层可能会涉及到下采样（除了2个不做下采样）；2、3、4块的第一个残差结构stride为2，目的是改变特征图尺寸
self.inplanes = planes * block.expansion # 下个残差结构的输入通道数是上个结构第一层输出通道数的block.expansion倍，其实看系列图每个块中最后一个卷积结构就可以知道
for i in range(1, blocks):
layers.append(block(self.inplanes, planes))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x
基于类及配置构建resnet网络
以两种Block为例说明
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
import torch.nn.functional as F
import torch
__all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101',
'resnet152']
model_urls = {
'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
'resnet101': 'https://download.pytorch.org/models/resnet101-5d3mb4d8f.pth',
'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
}
def resnet34(pretrained=False):
"""Constructs a ResNet-34 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [3, 4, 6, 3])
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet34']))
return model
def resnet50(pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 4, 6, 3])
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
return model
本文章为转载内容，我们尊重原作者对文章享有的著作权。如有内容错误或侵权问题，欢迎原作者联系我们进行内容更正或删除文章。
