1. 内容来源
2. Pytorch实现
2.1 ResNet块
(图1源:ResNet_哔哩哔哩_bilibili)
在图1中,颜色越深代表越小的模型,包含的区域小,颜色浅代表深度模型,包含的区域大。蓝色星为真实最优解。
在上左图中,随着模型加深,包含的区域变大,能抵达的最优解距离真实最优解距离先变小后变大,说明出现了偏移的问题,模型加深不一定能带来精度提升。而Resnet通过把加深的模型部分直接与小模型的输出相加,保证模型加深不会导致精度下降。
(图2源:ResNet_哔哩哔哩_bilibili)
图2为传统层层堆叠加深网络的方式和Resnet的对比图,虚线部分为增加的网络,Resnet通过将小模型输出(x)与增加部分(f(x))相加,保证f(x)不会导致新模型的偏移。
Residual块结构:
(图3源:ResNet_哔哩哔哩_bilibili)
代码实现:
import torch
from torch import nn
from torch.nn import functional as F
from d2l import torch as d2l
class Residual(nn.Module):
def __init__(self,input_channels, num_channels, use_1x1conv=False, strides=1):
super().__init__()
self.conv1 = nn.Conv2d(input_channels, num_channels, kernel_size=3, padding=1, stride=strides)
self.conv2 = nn.Conv2d(num_channels, num_channels, kernel_size=3, padding=1)
if use_1x1conv:
self.conv3 = nn.Conv2d(input_channels, num_channels, kernel_size=1, stride=strides)
else:
self.conv3 = None
self.bn1 = nn.BatchNorm2d(num_channels)
self.bn2 = nn.BatchNorm2d(num_channels)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
Y = F.relu(self.bn1(self.conv1(x)))
Y = self.bn2(self.conv2(Y))
if self.conv3:
x = self.conv3(x)
Y += x
return F.relu(Y)
Resnet块中包括N个Residual串联而得,且首个Residual块会让特征图长宽减半,之后的Residual块保持长宽不变,代码实现如下:
def resnet_block(input_channels, num_channels, num_residuals, first_block=False):
blk = []
for i in range(num_residuals):
if i == 0 and not first_block:
# Half the height and width
blk.append(Residual(input_channels, num_channels, use_1x1conv=True, strides=2))
else:
# Remain the height and width
blk.append(Residual(num_channels, num_channels))
return blk
2.2 ResNet构建
(图4源:ResNet_哔哩哔哩_bilibili)
ResNet网络架构如图4,与本系列中之前的VGG和GoogleNet类似,分为5stage,详细结构如下:
Stage1
- 卷积层1(输入通道1,输出通道64,7*7核,2步长,2填充)
- 批量归一层1(64通道)
- ReLU
- 最大池化层1(3*3核,2步长,1填充)
Stage2
- ResNet块(输入64通道,输出64通道,串联2Residual)
Stage3
- ResNet块(输入64通道,输出128通道,串联2Residual)
Stage4
- ResNet块(输入128通道,输出256通道,串联2Residual)
Stage5
- ResNet块(输入256通道,输出512通道,串联2Residual)
- 全局平均池化(输出维度(1, 1))
- 展开
- 线性层(512->10)
代码实现:
b1 = nn.Sequential(
nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=2),
nn.BatchNorm2d(64),
nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
)
b2 = nn.Sequential(*resnet_block(64, 64, 2, first_block=True))
b3 = nn.Sequential(*resnet_block(64, 128, 2))
b4 = nn.Sequential(*resnet_block(128, 256, 2))
b5 = nn.Sequential(*resnet_block(256, 512, 2))
net = nn.Sequential(b1, b2, b3, b4, b5,
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Linear(512, 10)
)
CHECK:
Sequential output shape: torch.Size([1, 64, 56, 56])
Sequential output shape: torch.Size([1, 64, 56, 56])
Sequential output shape: torch.Size([1, 128, 28, 28])
Sequential output shape: torch.Size([1, 256, 14, 14])
Sequential output shape: torch.Size([1, 512, 7, 7])
AdaptiveAvgPool2d output shape: torch.Size([1, 512, 1, 1])
Flatten output shape: torch.Size([1, 512])
Linear output shape: torch.Size([1, 10])
通过打印每一stage的输出,可以看到除1/2stage,均让特征图长宽减半,通道数倍增
2.3 模型训练
# Training
lr, num_epochs, batch_size = 0.05, 10, 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())
结果:
loss 0.009, train acc 0.998, test acc 0.912
2357.6 examples/sec on cuda:0
与之前模型对比发现训练集拟合效果暴增,但也带来较大的过拟合问题,但最终测试集精确度相较于之前的GoogLeNet的0.891有一定程度提升。