亮点
- 引入了残差结构
- 使用Batch Normalization加速训练(丢弃dropout)
这两个方法，解决了梯度消失和梯度爆炸等问题，使得构建深层网络成为可能
残差结构
计算量
左边是ResNet18/34的残差结构，右边是ResNet101/152的残差结构
1.左边计算量：3x3x256x256+3x3x256x256=1179648
2.右边计算量：1x1x256x64+3x3x64x64+1x1x64x256=69632
由此可见，右边一个残差结构的计算量更少，原因是右边的残差结构，使用阿一个1*1的卷积核，用来降维和升维
虚线残差结构
上图有两个残差结构，它们的区别在于：
- 左图的输入(Input)直接和输出(Output)相加，而右图的输入(Input2)需要经过一个1*1的卷积核，才能与输出(Output2)相加。
- 虚线残差第一个卷积的步距stride=2,而实线残差结构stride=1
下图中，1号框的64表示卷积核的个数，等于输出深度。2号框的既是虚线残差结构
代码解析
resnet18/34的残差结构
BasicBlock如下图所示，它包括实线残差和虚线残差两种结构
# resnet18和resnet34的残差结构 class BasicBlock(nn.Module): # 卷积核个数改变的倍数，如果一样，则expansion=1 expansion = 1 # ----------------------------残差结构-------------------------------------------- # in_channel 输入特征矩阵的深度 # out_channel 输出特征矩阵的深度，既是卷积核的个数 # stride 步距，当stride=1时，宽高不变，当stride=2时，宽高为原来的一半 # downsample 下采样，默认为None。只有使用到虚线残差结构才设为True # bias=False 使用BN结构时，不需要使用偏置(bias) # ------------------------------------------------------------------------------- def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs): super(BasicBlock, self).__init__() self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel, kernel_size=3, stride=stride, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(out_channel) self.relu = nn.ReLU() self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel, kernel_size=3, stride=1, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(out_channel) self.downsample = downsample def forward(self, x): identity = x # 判断self.downsample是否为空，如果不为空，则进行实线的残差结构 if self.downsample is not None: identity = self.downsample(x) # 主支线的正向传播 out = self.conv1(x) out = self.bn1(out) out = self.relu(out) out = self.conv2(out) out = self.bn2(out) # 主线和残差相加 out += identity out = self.relu(out) return out
下面的代码段，对应的是一次卷积操作
self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,kernel_size=3, stride=stride, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(out_channel) self.relu = nn.ReLU()
重复两次，既有
# ----------------------------残差结构-------------------------------------------- # in_channel 输入特征矩阵的深度 # out_channel 输出特征矩阵的深度，既是卷积核的个数 # stride 步距，当stride=1时，宽高不变，当stride=2时，宽高为原来的一半 # downsample 下采样，默认为None。只有使用到虚线残差结构才设为True # bias=False 使用BN结构时，不需要使用偏置(bias) # ------------------------------------------------------------------------------- def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs): super(BasicBlock, self).__init__() self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel, kernel_size=3, stride=stride, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(out_channel) self.relu = nn.ReLU() self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel, kernel_size=3, stride=1, padding=1, bias=False) self.bn2 = nn.BatchNorm2d(out_channel) self.downsample = downsample
self.downsample = downsample对应的下采样，即是用一个1*1的卷积实现虚线残差结构。如果self.downsample为空，机型实线残差。如果不为空，则进行虚线的残差结构
out += identity进行残差结构+主线结构
resnet50/101/152的残差结构Bottleneck
class Bottleneck(nn.Module): # 卷积核个数改变的倍数，这里主干上的输出深度为输入深度的4倍，则expansion=4 expansion = 4 def __init__(self, in_channel, out_channel, stride=1, downsample=None, groups=1, width_per_group=64): super(Bottleneck, self).__init__() width = int(out_channel * (width_per_group / 64.)) * groups self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=width, kernel_size=1, stride=1, bias=False) # squeeze channels self.bn1 = nn.BatchNorm2d(width) # ----------------------------------------- self.conv2 = nn.Conv2d(in_channels=width, out_channels=width, groups=groups, kernel_size=3, stride=stride, bias=False, padding=1) self.bn2 = nn.BatchNorm2d(width) # ----------------------------------------- self.conv3 = nn.Conv2d(in_channels=width, out_channels=out_channel*self.expansion, kernel_size=1, stride=1, bias=False) # unsqueeze channels self.bn3 = nn.BatchNorm2d(out_channel*self.expansion) self.relu = nn.ReLU(inplace=True) self.downsample = downsample
Bottleneck与BasicBlock相似，仅在几个地方有些许差异
1.expansion = 4:conv1和conv3的输出深度分别为64、256，它们相差了4倍，即是expansion，它在self.conv3 = nn.Conv2d(in_channels=width, out_channels=out_channel*self.expansion,kernel_size=1, stride=1, bias=False)中被使用
2.self.conv2 = nn.Conv2d(in_channels=width, out_channels=width, groups=groups,kernel_size=3, stride=stride, bias=False, padding=1)：注意stride=stride，它对应着虚线残差结构中的conv2，stride=2
一层layer的结构(_make_layer()函数)
# ----------------------------一个layer的结构-------------------------------------- # block 残差结构,有BasicBlock、Bottleneck # channel 与blocks_num对应,残差结构的卷积核数目，为一个列表。如resnet18为[2,2,2,2] # block_num 该层一共包含了几个残差块，即使执行的次数 # ------------------------------------------------------------------------------- def _make_layer(self, block, channel, block_num, stride=1): downsample = None # 判断是resnet18/resnet34还是resnet50/resnet101/resnet152 if stride != 1 or self.in_channel != channel * block.expansion: downsample = nn.Sequential( nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(channel * block.expansion)) # 若有downsample则为虚线残差结构，否则，仍是虚线残差结构 layers = [] layers.append(block(self.in_channel, channel, downsample=downsample, stride=stride, groups=self.groups, width_per_group=self.width_per_group)) self.in_channel = channel * block.expansion # 实线残差结构 for _ in range(1, block_num): layers.append(block(self.in_channel, channel, groups=self.groups, width_per_group=self.width_per_group)) return nn.Sequential(*layers)
block：残差结构,有BasicBlock、Bottleneck
channel：与blocks_num对应,残差结构的卷积核数目，为一个列表。如resnet18为[2,2,2,2]
block_num：该层一共包含了几个残差块，即使执行的次数
1.downsample = None默认下采样为空，即默认不执行1*1卷积核的虚线残差结构
2.if stride != 1 or self.in_channel != channel * block.expansion:默认in_channel=64，channel=64；对于18、34来说：expansion=1，所以self.in_channel != channel * block.expansion成立，故不进入if语句。对于50、101来说：expansion=4，所以不相等，进入if语句
3.if语句就是一个1*1的卷积操作
下图的【💔红色】1对应着代码1部分；【💚绿色】2代表的代码2部分
ResNet主网络
class ResNet(nn.Module): def __init__(self, block, blocks_num, num_classes=1000, include_top=True, groups=1, width_per_group=64): super(ResNet, self).__init__() self.include_top = include_top self.in_channel = 64 self.groups = groups self.width_per_group = width_per_group self.conv1 = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2, padding=3, bias=False) self.bn1 = nn.BatchNorm2d(self.in_channel) self.relu = nn.ReLU(inplace=True) self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) self.layer1 = self._make_layer(block, 64, blocks_num[0]) self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2) self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2) self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2) if self.include_top: self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # output size = (1, 1) self.fc = nn.Linear(512 * block.expansion, num_classes) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
代码仓库
Image_Classification_Net/ResNet at main · yzfzzz/Image_Classification_Net (github.com)
