参考：https://zh-v2.d2l.ai/chapter_convolutional-modern/vgg.html

VGG模型

VGG结构共有5个VGG块，主要模型为VGG-16和VGG-19，分别对应16层和19层结构（包括全连接层），所有卷积层均使用3x3的卷积核，每一个VGG块由若干相同的卷积层构成

VGG核AlexNet的比较，相当于是将AlexNet一部分抽象出来形成VGG块，然后堆叠VGG块以增加模型层数

复现（pytorch）

可以看到每一个VGG块中保持图像的长宽不变，由于kernel size = 3，根据公式计算得padding = 1

同时块内卷积层之间通道数也是相同的

实现代码如下

import torch
from torch import nn
from d2l import torch as d2l

"""
    定义vgg块
    num_convs:      块中卷积层数量
    in_channels:    块输入通道数
    out_channels:   块输出通道数
"""


def vgg_block(num_convs, in_channels, out_channels):
    layers = []
    for _ in range(num_convs):
        layers.append(nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1))
        layers.append(nn.ReLU())
        in_channels = out_channels
    layers.append(nn.MaxPool2d(2))
    return nn.Sequential(*layers)


# 定义vgg网络模型
def vgg(arch):
    blks = []
    in_channels = 1
    for (num_convs, out_channels) in arch:
        blks.append(vgg_block(num_convs, in_channels, out_channels))
        in_channels = out_channels
    return nn.Sequential(*blks, nn.Flatten(),
                         nn.Linear(out_channels * 7 * 7, 4096), nn.ReLU(), nn.Dropout(0.5),
                         nn.Linear(4096, 4096), nn.ReLU(), nn.Dropout(0.5),
                         nn.Linear(4096, 10))


# 声明vgg的块（卷积层数，通道数）
conv_arch = ((2, 64), (2, 128), (4, 256), (4, 512), (4, 512))
# conv_arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512))
net = vgg(conv_arch)

# X = torch.randn(size=(1, 1, 224, 224))
# for blk in net:
#     X = blk(X)
#     print(blk.__class__.__name__, 'output shape:\t', X.shape)


ratio = 4
small_conv_arch = [(pair[0], pair[1] // ratio) for pair in conv_arch]
net = vgg(small_conv_arch)
lr, num_epochs, batch_size = 0.05, 10, 128
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())