动手学深度学习（Pytorch版）代码实践 -卷积神经网络-29残差网络ResNet

29残差网络ResNet

import torch  
from torch import nn  
from torch.nn import functional as F 
import liliPytorch as lp  
import matplotlib.pyplot as plt

# 定义一个继承自nn.Module的残差块类
class Residual(nn.Module):
    def __init__(self, input_channels, num_channels, use_1x1conv=False, strides=1):
        super().__init__()
        # 第一个卷积层，使用3x3的卷积核，填充为1，步幅为指定值
        self.conv1 = nn.Conv2d(input_channels, num_channels, kernel_size=3, padding=1, stride=strides)
        # 第二个卷积层，使用3x3的卷积核，填充为1
        self.conv2 = nn.Conv2d(num_channels, num_channels, kernel_size=3, padding=1)
        
        # 可选的1x1卷积层，用于匹配输入输出通道数和步幅
        if use_1x1conv:
            self.conv3 = nn.Conv2d(input_channels, num_channels, kernel_size=1, stride=strides)
        else:
            self.conv3 = None
        
        # 批量归一化层
        self.bn1 = nn.BatchNorm2d(num_channels)
        self.bn2 = nn.BatchNorm2d(num_channels)
        # 为什么需要两个不同的批量归一化层?
        # 1.不同的位置，不同的输入特征
        # 2.独立的参数和统计数据
    
    def forward(self, X):
        # 先通过第一个卷积层、批量归一化层和ReLU激活函数
        Y = F.relu(self.bn1(self.conv1(X)))
        # 然后通过第二个卷积层和批量归一化层
        Y = self.bn2(self.conv2(Y))
        # 如果定义了conv3，则通过conv3调整X
        if self.conv3:
            X = self.conv3(X)
        # 将输入X加到输出Y上实现残差连接
        Y += X
        # 通过ReLU激活函数
        return F.relu(Y)

# 创建一个包含输入和输出形状一致的残差块实例，并测试其输出形状
# blk = Residual(3, 3)
# X = torch.rand(4, 3, 6, 6)
# Y = blk(X)
# print(Y.shape)  # 预期输出形状：torch.Size([4, 3, 6, 6])

# 创建一个包含1x1卷积和步幅为2的残差块实例，并测试其输出形状
# blk = Residual(3, 6, use_1x1conv=True, strides=2)
# print(blk(X).shape)  # 预期输出形状：torch.Size([4, 6, 3, 3])

# 定义一个包含初始卷积层、批量归一化层、ReLU激活函数和最大池化层的顺序容器
b1 = nn.Sequential(
    nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
    nn.BatchNorm2d(64),
    nn.ReLU(),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
)

# 定义一个函数，用于创建由多个残差块组成的模块
def resnet_block(input_channels, num_channels, num_residuals, first_block=False):
    blk = []
    for i in range(num_residuals):
        # 如果是第一个残差块且不是第一个模块，则使用1x1卷积和步幅为2
        if i == 0 and not first_block:
            blk.append(Residual(input_channels, num_channels, use_1x1conv=True, strides=2))
        else:
            blk.append(Residual(num_channels, num_channels))
    return blk

# 创建由残差块组成的各个模块
# *符号有多种用途，但在函数调用时，*符号主要用于将列表或元组解包。
# *resnet_block()的作用是将列表中的元素逐个传递给nn.Sequential
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))

# 创建整个ResNet模型
net = nn.Sequential(
    b1, 
    b2, 
    b3, 
    b4, 
    b5,
    nn.AdaptiveAvgPool2d((1, 1)),  # 自适应平均池化层
    nn.Flatten(),  # 展平层
    nn.Linear(512, 176)  # 全连接层，输出10类
)

# 测试整个网络的输出形状
X = torch.rand(size=(1, 1, 96, 96))
for layer in net:
    X = layer(X)
    print(layer.__class__.__name__, 'output shape:\t', X.shape)
# Sequential output shape:         torch.Size([1, 64, 24, 24])
# Sequential output shape:         torch.Size([1, 64, 24, 24])
# Sequential output shape:         torch.Size([1, 128, 12, 12])
# Sequential output shape:         torch.Size([1, 256, 6, 6])
# Sequential output shape:         torch.Size([1, 512, 3, 3])
# AdaptiveAvgPool2d output shape:  torch.Size([1, 512, 1, 1])
# Flatten output shape:    torch.Size([1, 512])
# Linear output shape:     torch.Size([1, 10])

# 设置训练参数
lr, num_epochs, batch_size = 0.05, 10, 256
# 加载训练和测试数据
train_iter, test_iter = lp.loda_data_fashion_mnist(batch_size, resize=96)
# 训练模型
lp.train_ch6(net, train_iter, test_iter, num_epochs, lr, lp.try_gpu())
# 显示训练结果
plt.show()

# loss 0.009, train acc 0.998, test acc 0.920
# 2306.3 examples/sec on cuda:0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115

运行结果：