pytorch学习笔记（二十五）：VGG

1. VGG块

VGG块的组成规律是：连续使用数个相同的填充为1、窗口形状为 $3\times 3$的卷积层后接上一个步幅为2、窗口形状为 $2\times 2$的最大池化层。卷积层保持输入的高和宽不变，而池化层则对其减半。我们使用vgg_block函数来实现这个基础的VGG块，它可以指定卷积层的数量和输入输出通道数。

import time
import torch
from torch import nn, optim

import sys
sys.path.append("..") 

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def vgg_block(num_convs, in_channels, out_channels):
    blk = []
    for i in range(num_convs):
        if i == 0:
            blk.append(nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1))
        else:
            blk.append(nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1))
        blk.append(nn.ReLU())
    blk.append(nn.MaxPool2d(kernel_size=2, stride=2)) # 这里会使宽高减半
    return nn.Sequential(*blk)

2. VGG网络

与AlexNet和LeNet一样，VGG网络由卷积层模块后接全连接层模块构成。卷积层模块串联数个vgg_block，其超参数由变量conv_arch定义。该变量指定了每个VGG块里卷积层个数和输入输出通道数。全连接模块则跟AlexNet中的一样。

现在我们构造一个VGG网络。它有5个卷积块，前2块使用单卷积层，而后3块使用双卷积层。第一块的输入输出通道分别是1（因为下面要使用的Fashion-MNIST数据的通道数为1）和64，之后每次对输出通道数翻倍，直到变为512。因为这个网络使用了8个卷积层和3个全连接层，所以经常被称为VGG-11。

conv_arch = ((1, 1, 64), (1, 64, 128), (2, 128, 256), (2, 256, 512), (2, 512, 512))
# 经过5个vgg_block, 宽高会减半5次, 变成 224/32 = 7
fc_features = 512 * 7 * 7 # c * w * h
fc_hidden_units = 64 # 任意

下面我们实现VGG-11。

def vgg(conv_arch, fc_features, fc_hidden_units=4096):
    net = nn.Sequential()
    # 卷积层部分
    for i, (num_convs, in_channels, out_channels) in enumerate(conv_arch):
        # 每经过一个vgg_block都会使宽高减半
        net.add_module("vgg_block_" + str(i+1), vgg_block(num_convs, in_channels, out_channels))
    # 全连接层部分
    net.add_module("fc", nn.Sequential(d2l.FlattenLayer(),
                                 nn.Linear(fc_features, fc_hidden_units),
                                 nn.ReLU(),
                                 nn.Dropout(0.5),
                                 nn.Linear(fc_hidden_units, fc_hidden_units),
                                 nn.ReLU(),
                                 nn.Dropout(0.5),
                                 nn.Linear(fc_hidden_units, 10)
                                ))
    return net

下面构造一个高和宽均为224的单通道数据样本来观察每一层的输出形状。

net = vgg(conv_arch, fc_features, fc_hidden_units)
X = torch.rand(1, 1, 224, 224)

# named_children获取一级子模块及其名字(named_modules会返回所有子模块,包括子模块的子模块)
for name, blk in net.named_children(): 
    X = blk(X)
    print(name, 'output shape: ', X.shape)

输出：

vgg_block_1 output shape:  torch.Size([1, 64, 112, 112])
vgg_block_2 output shape:  torch.Size([1, 128, 56, 56])
vgg_block_3 output shape:  torch.Size([1, 256, 28, 28])
vgg_block_4 output shape:  torch.Size([1, 512, 14, 14])
vgg_block_5 output shape:  torch.Size([1, 512, 7, 7])
fc output shape:  torch.Size([1, 10])

可以看到，每次我们将输入的高和宽减半，直到最终高和宽变成7后传入全连接层。与此同时，输出通道数每次翻倍，直到变成512。因为每个卷积层的窗口大小一样，所以每层的模型参数尺寸和计算复杂度与输入高、输入宽、输入通道数和输出通道数的乘积成正比。VGG这种高和宽减半以及通道翻倍的设计使得多数卷积层都有相同的模型参数尺寸和计算复杂度。

3. 获取数据和训练模型

因为VGG-11计算上比AlexNet更加复杂，出于测试的目的我们构造一个通道数更小，或者说更窄的网络在Fashion-MNIST数据集上进行训练。

ratio = 8
small_conv_arch = [(1, 1, 64//ratio), (1, 64//ratio, 128//ratio), (2, 128//ratio, 256//ratio), 
                   (2, 256//ratio, 512//ratio), (2, 512//ratio, 512//ratio)]
net = vgg(small_conv_arch, fc_features // ratio, fc_hidden_units // ratio)
print(net)

输出：

Sequential(
  (vgg_block_1): Sequential(
    (0): Conv2d(1, 8, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (vgg_block_2): Sequential(
    (0): Conv2d(8, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (vgg_block_3): Sequential(
    (0): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU()
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (vgg_block_4): Sequential(
    (0): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU()
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (vgg_block_5): Sequential(
    (0): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU()
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU()
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (fc): Sequential(
    (0): FlattenLayer()
    (1): Linear(in_features=3136, out_features=512, bias=True)
    (2): ReLU()
    (3): Dropout(p=0.5)
    (4): Linear(in_features=512, out_features=512, bias=True)
    (5): ReLU()
    (6): Dropout(p=0.5)
    (7): Linear(in_features=512, out_features=10, bias=True)
  )
)

模型训练过程与上一节的AlexNet中的类似。

def load_data_fashion_mnist(batch_size, resize=None, root='~/Datasets/FashionMNIST'):
    """Download the fashion mnist dataset and then load into memory."""
    trans = []
    if resize:
        trans.append(torchvision.transforms.Resize(size=resize))
    trans.append(torchvision.transforms.ToTensor())
    
    transform = torchvision.transforms.Compose(trans)
    mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=True, transform=transform)
    mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=True, transform=transform)
    if sys.platform.startswith('win'):
        num_workers = 0  # 0表示不用额外的进程来加速读取数据
    else:
        num_workers = 4
    train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
    test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)

    return train_iter, test_iter

def train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs):
    net = net.to(device)
    print("training on ", device)
    loss = torch.nn.CrossEntropyLoss()
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n, batch_count, start = 0.0, 0.0, 0, 0, time.time()
        for X, y in train_iter:
            X = X.to(device)
            y = y.to(device)
            y_hat = net(X)
            l = loss(y_hat, y)
            optimizer.zero_grad()
            l.backward()
            optimizer.step()
            train_l_sum += l.cpu().item()
            train_acc_sum += (y_hat.argmax(dim=1) == y).sum().cpu().item()
            n += y.shape[0]
            batch_count += 1
        test_acc = evaluate_accuracy(test_iter, net)
        print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f, time %.1f sec'
              % (epoch + 1, train_l_sum / batch_count, train_acc_sum / n, test_acc, time.time() - start))

batch_size = 64
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = load_data_fashion_mnist(batch_size, resize=224)

lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)

输出：

training on  cuda
epoch 1, loss 0.0101, train acc 0.755, test acc 0.859, time 255.9 sec
epoch 2, loss 0.0051, train acc 0.882, test acc 0.902, time 238.1 sec
epoch 3, loss 0.0043, train acc 0.900, test acc 0.908, time 225.5 sec
epoch 4, loss 0.0038, train acc 0.913, test acc 0.914, time 230.3 sec
epoch 5, loss 0.0035, train acc 0.919, test acc 0.918, time 153.9 sec

小结

• VGG-11通过5个可以重复使用的卷积块来构造网络。根据每块里卷积层个数和输出通道数的不同可以定义出不同的VGG模型。