type
status
date
slug
password
summary
tags
category
icon
VGG Blocks
LeNet, AlexNet, and VGG all share a common design pattern: extract features exploiting spatial structure via a sequence of convolutions and pooling layers and post-process the representations via fully connected layers. The improvements upon LeNet by AlexNet and VGG mainly lie in how these later networks widen and deepen these two modules.
This design poses two major challenges. First, the fully connected layers at the end of the architecture consume tremendous numbers of parameters. For instance, even a simple model such as VGG-11 requires a monstrous matrix, occupying almost 400MB of RAM in single precision (FP32). This is a significant impediment to computation, in particular on mobile and embedded devices. After all, even high-end mobile phones sport no more than 8GB of RAM. At the time VGG was invented, this was an order of magnitude less (the iPhone 4S had 512MB). As such, it would have been difficult to justify spending the majority of memory on an image classifier.
Second, it is equally impossible to add fully connected layers earlier in the network to increase the degree of nonlinearity: doing so would destroy the spatial structure and require potentially even more memory.
The network in network (NiN) blocks (Lin et al., 2013) offer an alternative, capable of solving both problems in one simple strategy. They were proposed based on a very simple insight: (i) use 1×1 convolutions to add local nonlinearities across the channel activations and (ii) use global average pooling to integrate across all locations in the last representation layer. Note that global average pooling would not be effective, were it not for the added nonlinearities. Let’s dive into this in detail.
While AlexNet offered empirical evidence that deep CNNs can achieve good results, it did not provide a general template to guide subsequent researchers in designing new networks. In the following sections, we will introduce several heuristic concepts commonly used to design deep networks.
Progress in this field mirrors that of VLSI (very large scale integration) in chip design where engineers moved from placing transistors to logical elements to logic blocks (Mead, 1980). Similarly, the design of neural network architectures has grown progressively more abstract, with researchers moving from thinking in terms of individual neurons to whole layers, and now to blocks, repeating patterns of layers. A decade later, this has now progressed to researchers using entire trained models to repurpose them for different, albeit related, tasks. Such large pretrained models are typically called foundation models (Bommasani et al., 2021).
Back to network design. The idea of using blocks first emerged from the Visual Geometry Group (VGG) at Oxford University, in their eponymously-named VGG network (Simonyan and Zisserman, 2014). It is easy to implement these repeated structures in code with any modern deep learning framework by using loops and subroutines.
虽然AlexNet证明深层神经网络卓有成效，但它没有提供一个通用的模板来指导后续的研究人员设计新的网络。 在下面的几个章节中，我们将介绍一些常用于设计深层神经网络的启发式概念。
与芯片设计中工程师从放置晶体管到逻辑元件再到逻辑块的过程类似，神经网络架构的设计也逐渐变得更加抽象。研究人员开始从单个神经元的角度思考问题，发展到整个层，现在又转向块，重复层的模式。
使用块的想法首先出现在牛津大学的视觉几何组（visual geometry group）的VGG网络中。通过使用循环和子程序，可以很容易地在任何现代深度学习框架的代码中实现这些重复的架构。
经典卷积神经网络的基本组成部分是下面的这个序列：
- 带填充以保持分辨率的卷积层；
- 非线性激活函数，如ReLU；
- 汇聚层，如最大汇聚层。
而一个VGG块与之类似，由一系列卷积层组成，后面再加上用于空间下采样的最大汇聚层。在最初的VGG论文中 (Simonyan and Zisserman, 2014)，作者使用了带有3×3卷积核、填充为1（保持高度和宽度）的卷积层，和带有2×2汇聚窗口、步幅为2（每个块后的分辨率减半）的最大汇聚层。在下面的代码中，我们定义了一个名为
vgg_block的函数来实现一个VGG块。
注意到: 打印层的尺寸时，我们只看到8个结果，而不是11个结果。剩余的3层信息去哪了？(Sequential 和 Linear)
其中Sequential 5层，其中Linear3层，ReLU，Dropout这些层都不算的，因为它们不是神经元。
我们可以得知前两个Sequential层包含了1个卷积层，后三个Sequential层包含了两个卷积层。
- VGG-11使用可复用的卷积块构造网络。不同的VGG模型可通过每个块中卷积层数量和输出通道数量的差异来定义。
- 块的使用导致网络定义的非常简洁。使用块可以有效地设计复杂的网络。
- 在VGG论文中，Simonyan和Ziserman尝试了各种架构。特别是他们发现深层且窄的卷积（即3×3）比较浅层且宽的卷积更有效。
- 请参考VGG论文 (Simonyan and Zisserman, 2014)中的表1构建其他常见模型，如VGG-16或VGG-19。
Network in Network (NiN)
LeNet, AlexNet, and VGG all share a common design pattern: extract features exploiting spatial structure via a sequence of convolutions and pooling layers and post-process the representations via fully connected layers. The improvements upon LeNet by AlexNet and VGG mainly lie in how these later networks widen and deepen these two modules.
This design poses two major challenges. First, the fully connected layers at the end of the architecture consume tremendous numbers of parameters. For instance, even a simple model such as VGG-11 requires a monstrous matrix, occupying almost 400MB of RAM in single precision (FP32). This is a significant impediment to computation, in particular on mobile and embedded devices. After all, even high-end mobile phones sport no more than 8GB of RAM. At the time VGG was invented, this was an order of magnitude less (the iPhone 4S had 512MB). As such, it would have been difficult to justify spending the majority of memory on an image classifier.
Second, it is equally impossible to add fully connected layers earlier in the network to increase the degree of nonlinearity: doing so would destroy the spatial structure and require potentially even more memory.
The network in network (NiN) blocks (Lin et al., 2013) offer an alternative, capable of solving both problems in one simple strategy. They were proposed based on a very simple insight: (i) use 1×1 convolutions to add local nonlinearities across the channel activations and (ii) use global average pooling to integrate across all locations in the last representation layer. Note that global average pooling would not be effective, were it not for the added nonlinearities. Let’s dive into this in detail.
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 (Lin et al., 2013)
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1×1卷积层（如 6.4节中所述），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）。
图7.3.1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1×1的卷积层。这两个1×1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1×1。
Multi-Branch Networks (GoogLeNet)
GoogLeNet吸收了NiN中串联网络的思想，并在此基础上做了改进。 这篇论文的一个重点是解决了什么样大小的卷积核最合适的问题。 毕竟，以前流行的网络使用小到1×1，大到11×11的卷积核。 本文的一个观点是，有时使用不同大小的卷积核组合是有利的。 本节将介绍一个稍微简化的GoogLeNet版本：我们省略了一些为稳定训练而添加的特殊特性，现在有了更好的训练方法，这些特性不是必要的。
在GoogLeNet中，基本的卷积块被称为Inception块（Inception block）。这很可能得名于电影《盗梦空间》（Inception），因为电影中的一句话“我们需要走得更深”（“We need to go deeper”）。
Inception Blocks
Inception块由四条并行路径组成。
GoogLeNet Model
GoogLeNet一共使用9个Inception块和全局平均汇聚层的堆叠来生成其估计值。Inception块之间的最大汇聚层可降低维度。 第一个模块类似于AlexNet和LeNet，Inception块的组合从VGG继承，全局平均汇聚层避免了在最后使用全连接层。
上一篇
Deep learning Guide 4: Batch Normalization, ResNet, DenseNet, 面試經典題
下一篇
Deep learning Guide 2: LeNet model
- Author:tom-ci
- URL:https://www.tomciheng.com//article/d2lv-3
- Copyright:All articles in this blog, except for special statements, adopt BY-NC-SA agreement. Please indicate the source!