网络中的网络（NiN）¶

NiN块

In [1]:

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

def nin_block(in_channels, out_channels, kernel_size, strides, padding):
    return nn.Sequential(
        nn.Conv2d(in_channels, out_channels, kernel_size, strides, padding),
        nn.ReLU(), nn.Conv2d(out_channels, out_channels, kernel_size=1),
        nn.ReLU(), nn.Conv2d(out_channels, out_channels, kernel_size=1),
        nn.ReLU())

NiN模型

In [2]:

net = nn.Sequential(
    nin_block(1, 96, kernel_size=11, strides=4, padding=0),
    nn.MaxPool2d(3, stride=2),
    nin_block(96, 256, kernel_size=5, strides=1, padding=2),
    nn.MaxPool2d(3, stride=2),
    nin_block(256, 384, kernel_size=3, strides=1, padding=1),
    nn.MaxPool2d(3, stride=2), nn.Dropout(0.5),
    nin_block(384, 10, kernel_size=3, strides=1, padding=1),
    nn.AdaptiveAvgPool2d((1, 1)),
    nn.Flatten())

查看每个块的输出形状

In [3]:

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

Sequential output shape:	 torch.Size([1, 96, 54, 54])
MaxPool2d output shape:	 torch.Size([1, 96, 26, 26])
Sequential output shape:	 torch.Size([1, 256, 26, 26])
MaxPool2d output shape:	 torch.Size([1, 256, 12, 12])
Sequential output shape:	 torch.Size([1, 384, 12, 12])
MaxPool2d output shape:	 torch.Size([1, 384, 5, 5])
Dropout output shape:	 torch.Size([1, 384, 5, 5])
Sequential output shape:	 torch.Size([1, 10, 5, 5])
AdaptiveAvgPool2d output shape:	 torch.Size([1, 10, 1, 1])
Flatten output shape:	 torch.Size([1, 10])

训练模型

In [4]:

lr, num_epochs, batch_size = 0.1, 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())

loss 0.331, train acc 0.878, test acc 0.878
3182.2 examples/sec on cuda:0