.. _sec_nin:
网络中的网络（NiN）
===================
LeNet、AlexNet和VGG都有一个共同的设计模式：通过一系列的卷积层与汇聚层来提取空间结构特征；然后通过全连接层对特征的表征进行处理。
AlexNet和VGG对LeNet的改进主要在于如何扩大和加深这两个模块。
或者，可以想象在这个过程的早期使用全连接层。然而，如果使用了全连接层，可能会完全放弃表征的空间结构。
*网络中的网络*\ （\ *NiN*\ ）提供了一个非常简单的解决方案：在每个像素的通道上分别使用多层感知机
:cite:`Lin.Chen.Yan.2013`
NiN块
-----
回想一下，卷积层的输入和输出由四维张量组成，张量的每个轴分别对应样本、通道、高度和宽度。
另外，全连接层的输入和输出通常是分别对应于样本和特征的二维张量。
NiN的想法是在每个像素位置（针对每个高度和宽度）应用一个全连接层。
如果我们将权重连接到每个空间位置，我们可以将其视为\ :math:`1\times 1`\ 卷积层（如
:numref:`sec_channels`\ 中所述），或作为在每个像素位置上独立作用的全连接层。
从另一个角度看，即将空间维度中的每个像素视为单个样本，将通道维度视为不同特征（feature）。
:numref:`fig_nin`\ 说明了VGG和NiN及它们的块之间主要架构差异。
NiN块以一个普通卷积层开始，后面是两个\ :math:`1 \times 1`\ 的卷积层。这两个\ :math:`1 \times 1`\ 卷积层充当带有ReLU激活函数的逐像素全连接层。
第一层的卷积窗口形状通常由用户设置。
随后的卷积窗口形状固定为\ :math:`1 \times 1`\ 。
.. _fig_nin:
.. figure:: ../img/nin.svg
:width: 600px
对比 VGG 和 NiN 及它们的块之间主要架构差异。
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
from mindspore import nn, ops
from d2l import mindspore as d2l
def nin_block(in_channels, out_channels, kernel_size, strides, padding):
return nn.SequentialCell([
nn.Conv2d(in_channels, out_channels, kernel_size, strides, 'pad', 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()])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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())
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
net = nn.SequentialCell([
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(keep_prob=1-0.5),
nin_block(384, 10, kernel_size=3, strides=1, padding=1),
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten()])
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
[WARNING] ME(2919748:140037300815680,MainProcess):2026-05-21-02:27:53.297.000 [mindspore/nn/layer/basic.py:176] For Dropout, this parameter `keep_prob` will be deprecated, please use `p` instead.
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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),
# 标签类别数是10
nin_block(384, 10, kernel_size=3, strides=1, padding=1),
nn.AdaptiveAvgPool2d((1, 1)),
# 将四维的输出转成二维的输出，其形状为(批量大小,10)
nn.Flatten())
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
X = ops.randn(1, 1, 224, 224)
for blk in net:
X = blk(X)
print(blk.__class__.__name__,'output shape:\t',X.shape)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
SequentialCell output shape: (1, 96, 54, 54)
MaxPool2d output shape: (1, 96, 26, 26)
SequentialCell output shape: (1, 256, 26, 26)
MaxPool2d output shape: (1, 256, 12, 12)
SequentialCell output shape: (1, 384, 12, 12)
MaxPool2d output shape: (1, 384, 5, 5)
Dropout output shape: (1, 384, 5, 5)
SequentialCell output shape: (1, 10, 5, 5)
AdaptiveAvgPool2d output shape: (1, 10, 1, 1)
Flatten output shape: (1, 10)
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
X = torch.rand(size=(1, 1, 224, 224))
for layer in net:
X = layer(X)
print(layer.__class__.__name__,'output shape:\t', X.shape)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
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])
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
lr, num_epochs, batch_size = 0.05, 3, 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)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 1.820, train acc 0.302, test acc 0.504
6154.4 examples/sec
.. figure:: output_nin_8ad4f3_30_1.png
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
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())
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.395, train acc 0.856, test acc 0.849
5721.0 examples/sec on cuda:0
.. figure:: output_nin_8ad4f3_33_1.svg
.. raw:: html
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html
`讨论 `__
.. raw:: html
.. raw:: html