VGG神经网络（李沐老师课程）

vgg网络结构图

Vgg主要思想来源

为什么不用5*5卷积核，在过程发现多个卷积核为3的效果比少量的大卷积核效果要好

VGG的贡献：网络结构更加的规整

VGG网络，LeNet网络，AlexNet的贡献

各个网络的对比

代码：

import torch
from torch import nn
from d2l import torch as d2l
​
"""
该函数有三个参数，分别对应于卷积层的数量num_convs、输入通道的数量in_channels 和输出通道的数量out_channels.
"""
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(kernel_size=2,stride=2))
    return nn.Sequential(*layers)
​
conv_arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512))
​
"""
VGG-11
"""
def vgg(conv_arch):
    conv_blks = []
    in_channels = 1
    # 卷积层部分
    for (num_convs, out_channels) in conv_arch:
        conv_blks.append(vgg_block(num_convs, in_channels, out_channels))
        in_channels = out_channels
​
    return nn.Sequential(
        *conv_blks, nn.Flatten(),
        # 全连接层部分
        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))
​
net = vgg(conv_arch)
"""
每个层输出的形状
"""
X = torch.randn(size=(1, 1, 224, 224))
for blk in net:
    X = blk(X)
    print(blk.__class__.__name__,'output shape:\t',X.shape)
"""
通道数较少的网络，足够用于训练Fashion-MNIST数据集
"""
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())
d2l.plt.show()