AlexNet

总结：

• AlexNet是更大更深的LeNet，10×参数个数，260×计算复杂度
• 新加入了丢弃法、ReLU、最大池化层、数据增强
• AlexNet赢下了2012ImageNet竞赛后，标志着新的一轮神经网络热潮的开始。

AlexNet代码实现

1. 导入相关库

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

2. 定义网络模型

net = nn.Sequential(
    nn.Conv2d(1, 96, kernel_size=11, stride=4, padding=1), nn.ReLU(),  # [1, 96, 54, 54]
    nn.MaxPool2d(kernel_size=3, stride=2),  # [1, 96, 26, 26]

    nn.Conv2d(96, 256, kernel_size=5, padding=2), nn.ReLU(),  # [1, 256, 26, 26]
    nn.MaxPool2d(kernel_size=3, stride=2),  # [1, 256, 12, 12]

    nn.Conv2d(256, 384, kernel_size=3, padding=1), nn.ReLU(),  # [1, 384, 12, 12]
    nn.Conv2d(384, 384, kernel_size=3, padding=1), nn.ReLU(),  # [1, 384, 12, 12]
    nn.Conv2d(384, 256, kernel_size=3, padding=1), nn.ReLU(),  # [1, 256, 12, 12]
    nn.MaxPool2d(kernel_size=3, stride=2),  # [1, 256, 5, 5]

    nn.Flatten(),  # [1, 256, 5, 5] --> [1, 256*5*5]= [1, 6400]

    nn.Linear(6400, 4096), nn.ReLU(),  # [1, 6400] --> [1, 4096]
    nn.Dropout(p=0.5),
    nn.Linear(4096, 4096), nn.ReLU(),  # [1, 4096] --> [1, 4096]
    nn.Dropout(p=0.5),

    nn.Linear(4096, 1000)  # [1, 4096] --> [1, 4096]
)

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

3. 加载数据集

batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)

4. 训练模型

lr, num_epochs = 0.01, 20
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())