VGG块

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

def vgg_block(num_convs, in_channels, out_channels):
    '''
    vgg块, 卷积层数量，输入通道数量，输出通道数量
    :param num_convs 块中卷积层数量
    :param in_channels 块中输入通道数量
    :param out_channels 块中输出通道数量
    '''
    # vgg块
    layers = []
    for _ in range(num_convs):
        layers.append(nn.Conv2d(in_channels, out_channels,
                                kernel_size=3, padding=1))
        layers.append(nn.ReLU())
        # 前1个卷积层的输出通道是下1个卷积层的输入通道
        in_channels = out_channels
    layers.append(nn.MaxPool2d(kernel_size=2,stride=2))
    return nn.Sequential(*layers)

VGG网络

# 5个vgg块，(vgg块中卷积层数量,输出通道数)
conv_arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512))

# 给我一个vgg的架构
def vgg(conv_arch):
    conv_blks = []
    in_channels = 1
    for (num_convs, out_channels) in conv_arch:
        conv_blks.append(vgg_block(num_convs, in_channels, out_channels))
        in_channels = out_channels
    # *tuple入参
    return nn.Sequential(
        *conv_blks, 
        # 展平层
        nn.Flatten(),
        # 最后1个卷积层的输出通道数x最后1个卷积层输出图片宽*高
        nn.Linear(out_channels * 7 * 7, 4096), 
        # 非线性单元
        # 简单线性函数模拟复杂非线性函数的根本
        # ReLU数值稳定性好，避免梯度消失或爆炸
        nn.ReLU(), 
        # 可以使用复杂的模型，模型容量大的模型，使用正则化方法避免复杂模型过拟合
        # 正则化，限制w的取值范围，取得更平滑的曲线，降低过拟合，提高泛化能力
        # 丢弃法，以概率p置零隐藏层的全连接层的输出
        nn.Dropout(0.5),
        nn.Linear(4096, 4096), 
        nn.ReLU(), 
        nn.Dropout(0.5),
        # 10个类别
        nn.Linear(4096, 10))

net = vgg(conv_arch)

# 我们构造一个 单通道数据，来观察每一层输出的形状
X = torch.randn(size=(1, 1, 224, 224))
# ！这种方式可以观察每一块的输出的shape，而不用pytorch提供的函数
for blk in net:
    X = blk(X)
    print(blk.__class__.__name__,'output shape:\t',X.shape)

Sequential output shape:	 torch.Size([1, 64, 112, 112])
Sequential output shape:	 torch.Size([1, 128, 56, 56])
Sequential output shape:	 torch.Size([1, 256, 28, 28])
Sequential output shape:	 torch.Size([1, 512, 14, 14])
Sequential output shape:	 torch.Size([1, 512, 7, 7])
Flatten output shape:	 torch.Size([1, 25088])
Linear output shape:	 torch.Size([1, 4096])
ReLU output shape:	 torch.Size([1, 4096])
Dropout output shape:	 torch.Size([1, 4096])
Linear output shape:	 torch.Size([1, 4096])
ReLU output shape:	 torch.Size([1, 4096])
Dropout output shape:	 torch.Size([1, 4096])
Linear output shape:	 torch.Size([1, 10])

# vgg11（8个卷积层+3个全连接层） 卷积层加全连接层
# 由于VGG-11比AlexNet计算量更大，因此我们构建了一个通道数较少的网络
ratio = 4
# 把输出通道数除4，向下取整
small_conv_arch = [(pair[0], pair[1] // ratio) for pair in conv_arch]
net = vgg(small_conv_arch)

# 模型训练
lr, num_epochs, batch_size = 0.05, 10, 128
# Fashion-MNIST图像的分辨率=28x28 < ImageNet的分辨率，reshape到224×224
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())

loss 0.182, train acc 0.932, test acc 0.915
2260.1 examples/sec on cuda:0

总结

• vgg超参数，vgg块数量，vgg块中卷积层数量,输出通道数)
• conv_arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512))