resnet调用修改输入 resnet代码详解

转载

mob6454cc6c1f4a 2024-05-07 10:18:37

文章标签 resnet调用修改输入 pytorch 深度学习神经网络 ide 文章分类 架构后端开发

Resnet

设计起因是随着网络层增加反而梯度下降困难，甚至起到反作用，因此加入残差结构。

残差网络原理就是"正常梯度+消失梯度=正常梯度"，只要自身的梯度是正常的，就算加上多层后出现的消失的梯度也是正常的值，这样能够保证梯度正常反向传播。

Resnet设计了两类残差块Basic_block和Bottleneck，分别用于不同层次的Resnet。完整代码

Basic_block

是两层的残差块,用于resnet18/34：

resnet调用修改输入 resnet代码详解_pytorch

代码部分：

class BasicBlock(nn.Module):
"""
inplane是输入的通道数,plane是输出的通道数,expansion是对输出通道数的倍乘.
"""
    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
"""
首先residual=x保存自身的值,然后x经过两个(conv+bn+relu)得到out,再将out+=residual.
"""
    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

Bottleneck

三层的残差块，用于resnet50/101/152：
第一个1x1的卷积层用于降维，第二个3x3层用于处理，第三个1x1层用于升维，这样减少了计算量。

第一个1x1的卷积把256维channel降到64维，然后在最后通过1x1卷积恢复，整体上用的参数数目：1x1x256x64 +3x3x64x64 + 1x1x64x256 = 69632，而不使用bottleneck的话就是两个3x3x256的卷积，参数数目: 3x3x256x256x2 = 1179648，差了16.94倍。

resnet调用修改输入 resnet代码详解_resnet调用修改输入_02

class Bottleneck(nn.Module):
"""
区别：
使用了3次卷积对通道数进行压缩,再放大:
conv1卷积核是(1*1)[inplanes, planes]
conv2卷积核(3*3)[planes, planes]
conv3卷积核(1*1)[planes, planes * self.expansion] #expansion=4
注意：
Conv2d都将bias=False，因为卷积后都经过BN，归一化后在变换重构部分加的beta 就相当于+bias。
"""
   expansion = 4
   def __init__(self, inplanes, planes, stride=1, downsample=None):
       super(Bottleneck, self).__init__()
       self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
       self.bn1 = nn.BatchNorm2d(planes)
       self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,padding=1, bias=False)
       self.bn2 = nn.BatchNorm2d(planes)
       self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1, bias=False)
       self.bn3 = nn.BatchNorm2d(planes * self.expansion)
       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 out

ResNet网络

conv1+bn1+relu+maxpool：(224,224,3)->(112,112,64)->(56,56,64)
其中conv1卷积核大小7*7,stride=2,p=3,通道从3->64。maxpool为2倍下采样。

self.inplanes = 64
  self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,bias=False) 
  self.bn1 = nn.BatchNorm2d(64)
  self.relu = nn.ReLU(inplace=True)
  self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) #2倍下采样

然后构造layer1-layer4：

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)

layers是叠加层数，如resnet50：[3, 4, 6, 3]

def resnet50(pretrained=False, **kwargs):
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)

其中_make_layer的构造：

# block：基础块的类型，是BasicBlock，还是Bottleneck
# planes：当前块的输入输入通道数
# blocks：块的数目
def _make_layer(self, block, planes, blocks, stride=1):
"""
downsample的作用：
拼接layers第一步是block(self.inplanes, planes, stride, downsample),其中不管经过Bottleneck
还是BasicBlock都会有这一步：
	if self.downsample is not None:
		residual = self.downsample(x)
然后有：
	out += residual
out在Bottleneck中通道数是inplanes->planes * self.expansion,在BasicBlock是inplanes->planes.

所以downsample作用是保证out+=residual步骤，将输入的特征图和残差结构中卷积操作后的特征尺寸一致
才能相加.
self.inplanes->planes*block.expansion，其中BasicBlock.expansion=1，Bottleneck.expansion=4.
"""
    downsample = None
    if stride != 1 or self.inplanes != planes * block.expansion:
        downsample = nn.Sequential(
            conv1x1(self.inplanes, planes * block.expansion, stride),
            nn.BatchNorm2d(planes * block.expansion),
        )
	#拼接blocks个layer
    layers = []
    layers.append(block(self.inplanes, planes, stride, downsample))
    self.inplanes = planes * block.expansion
    for _ in range(1, blocks):
        layers.append(block(self.inplanes, planes))
    return nn.Sequential(*layers)

最后平均池化+全连接.

self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)

完整代码：
我对参数初始化的意义不是很明白！for m in self.modules()和zero_init_residual部分

class ResNet(nn.Module):
   #参数block指明残差块是两层或三层，参数layers指明每个卷积层需要的残差块数量，num_classes指明分类数，zero_init_residual是否初始化为0
   def __init__(self, block, layers, num_classes=1000, zero_init_residual=False):
       super(ResNet, self).__init__()
       self.inplanes = 64
       self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,bias=False) 
       self.bn1 = nn.BatchNorm2d(64)
       self.relu = nn.ReLU(inplace=True)
       self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) #2倍下采样
       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)
       self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
       self.fc = nn.Linear(512 * block.expansion, num_classes)
       #对模型卷积层参数以及BN层参数的初始化操作
       for m in self.modules():
           if isinstance(m, nn.Conv2d):
               #kaiming高斯初始化，目的是使得Conv2d卷积层反向传播的输出的方差都为1
               nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
           elif isinstance(m, nn.BatchNorm2d):
               #初始化m.weight，即gamma的值为1；m.bias即beta的值为0
               nn.init.constant_(m.weight, 1)
               nn.init.constant_(m.bias, 0)
       # 在每个残差分支中初始化最后一个BN，即BatchNorm2d
       # 以便残差分支以零开始，并且每个残差块的行为类似于一个恒等式。
       if zero_init_residual:
           for m in self.modules():
               if isinstance(m, Bottleneck):#Bottleneck的最后一个BN是m.bn3
                   nn.init.constant_(m.bn3.weight, 0)
               elif isinstance(m, BasicBlock):#BasicBlock的最后一个BN是m.bn2
                   nn.init.constant_(m.bn2.weight, 0)
   def _make_layer(self, block, planes, blocks, stride=1):
       downsample = None
       if stride != 1 or self.inplanes != planes * block.expansion:
           downsample = nn.Sequential(
               conv1x1(self.inplanes, planes * block.expansion, stride),
               nn.BatchNorm2d(planes * block.expansion),
           )
       layers = []
       layers.append(block(self.inplanes, planes, stride, downsample))
       self.inplanes = planes * block.expansion
       for _ 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

不同层次网络实现

resnet调用修改输入 resnet代码详解_ide_03

#18层的resnet
def resnet18(pretrained=False, **kwargs):
    model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
    if pretrained:#是否使用已经训练好的预训练模型，在此基础上继续训练
        model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
    return model
#34层的resnet
def resnet34(pretrained=False, **kwargs):
    model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
    if pretrained:#是否使用已经训练好的预训练模型，在此基础上继续训练
        model.load_state_dict(model_zoo.load_url(model_urls['resnet34']))
    return model
#50层的resnet
def resnet50(pretrained=False, **kwargs):
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
    if pretrained:#是否使用已经训练好的预训练模型，在此基础上继续训练
        model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
    return model
#101层的resnet
def resnet101(pretrained=False, **kwargs):
    model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
    if pretrained:#是否使用已经训练好的预训练模型，在此基础上继续训练
        model.load_state_dict(model_zoo.load_url(model_urls['resnet101']))
    return model
#152层的resnet
def resnet152(pretrained=False, **kwargs):
    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
    if pretrained:#是否使用已经训练好的预训练模型，在此基础上继续训练
        model.load_state_dict(model_zoo.load_url(model_urls['resnet152']))
    return model