《最强码农的水论文攻略》之水论文大法第三期--以IEEE access为例

相信大家看了我的前几期教程以后，实验指定都是做的差不多了，看来大家也已经到了要水论文的阶段了吧，嘿嘿嘿~~~本期，俺就教大家如何无痛入门，水一篇论文~~~~

 我先随便写一篇毫无意义的论文在下面，尽量保证每个章节，每个地方都有数据，然后再教大家如何屎盆子镶金边

———————————————————————————————————————————

基于改进YOLOv8的头部检测

Jason wu

摘要：头部检测那可是现代目标检测领域的一个重要的研究方向呀，不仅可以上课查看学生有没有低头睡觉，也可以用来判断一个班级学习氛围到底好不好，还可以放在公共场所判断人流量之类的，总而言之，言而总之，对于头部检测的研究还是很有必要滴。传统的头部检测方法速度慢，精度低，俺们提出了一种新的头部检测算法，往YOLOv8的骨干网络和颈部网络连接处插了三个CBAM注意力模块，然后在SCUT-HEAD头部检测数据集上进行了实验，实验结果表明，俺们的模型那是大大的好啊，速度快，精度高，模型小，完完全全那是吊打YOLOv8，所以，俺们认为本文提出的模型在给小学生增加学习压力以及防止大学生不上早八具有重要意义。

索引术语：YOLOv8, CBAM，头部检测，小学生

1. 引言

头部检测怎么怎么重要，小学生上课怎么怎么不认真，大学生怎么怎么猛猛逃课，俺这里就不赘述了，总是研究就是有意义。

2. 相关工作

传统的头部检测是在拿什么方法做，RGB图像啦，深度学习啦，机器学习啦，或者直接热力图分析啥的，这部分多看看论文，多看看综述，指定能写出来的，总之就是以上工作俺都觉得一般，不如我的工作的一根。

3. 算法设计

3.1 YOLOv8算法介绍

YOLOv8这个模型那是大大滴好啊，好在哪儿呢，好就好在模型小，结构优秀，既有YOLOv7的c2f结构，又有YOLOv6的解耦头，还有分布式焦点损失，甚至还有YOLOX的提前10轮关闭数据增强这一骚操作，总之就是高级得很，除了名字是自己的，其他全是缝的。

3.2 YOLOv8的不足与改进思路

YOLOv8就这，俺也觉得他这模型一般般，咋一般般捏，我觉得它特征提的不够深，直接提完就传给颈部网络开始特征金字塔了，我觉得8太行，我的建议是加个注意力，少壮不努力，毕业注意力嘛。

3.3 JSON-YOLOv8

直接开始展示咋改的，我这里就手绘一个改进图，不细说了。

再介绍一下什么是CBAM

这里直接看论文就完了，把CBAM的公式，原理啥的推一推，结束。

4. 实验与结果

4.1 实验配置

本实验的实验配置那是相当上档次的，2022年买的外星人m15r7，12代i7，加上3060（6GB），上次在咸鱼那人直接跟我说4000给我收了，我人都麻了，4000还不到我买的时候的零头。

这里说说操作系统，win11，pytorch2.0.1，python3.10.8，输入图像尺寸640*640，只开100轮，学习率，权重衰减系数啥的直接看代码，我这里就不介绍了。

4.2 评估系数

介绍一下什么是准确度（Precision）、召回率（Recall）、平均准确度（Average Precision, AP）、平均准确率（mean Average Precision, mAP）和F1分数（F1 Score）。

公式啥的这里放一放，举个例子：

平均准确率（mean Average Precision, mAP）：

其中，K是检测到的目标类别总数，APi是第i个类别的平均准确度。

4.3 消融实验

本次实验就改了一丢丢，就做那个加和不加的实验就完了，令原模型为baseline,本文改的模型前面取了个名字，叫JSON-YOLOv8，这里就可以摆摆数据结果啥的。俺直接上图

从这张图里一看，俺们的模型那真是大大滴好啊，哈哈哈哈

4.3 对比实验

俺在看看俺们的结果跟其他模型比咋样捏

总之就是精度高的比我大，精度低的，那还叫什么，精度都没我高，爪巴。推理比我快的精度低，推理比我慢的，那更别叫，总结一句话，俺的模型最屌。

4.4 结果可视化对比实验

这里我就随便放张图了，这实验我确实没整过，嘿嘿

从图里面一看就知道，俺们的模型那确实比其他的模型强得多啊，有图为证，不解释好吧。

5. 总结

基于对现在头部检测的方法深入分析，俺们提出了一种新的头部检测算法，即JSON-YOLOv8，俺们往里加了一个CBAM注意力，加完以后模型那是嘎嘎好，性能提升非常明显。

为了验证俺们的有效性，俺们还在公共数据集上进行了实验，实验结果证明，俺们的模型吊打一切，其他的模型完全不是对手啊。

最后，俺们未来的研究方向可能还有针对过马路闯红灯的啊，看俺的帖子不点赞啊，对这些娃进行头部检测，我直接一个逮到。

———————————————————————————————————————————

OK，以上废话就到这，接下来，俺就教大家，如何给上面那坨答辩，改成依托精致的答辩。

一. 环境搭建

工欲善其事必先利其器，想要写出一篇编辑一眼望过去，还能接受的论文，首先，那必须是各种科研绘图，公式编辑，以及格式啥的直接拉满呐哥哥，现在大部分期刊交论文，要么是直接交latex，要么交pdf，用word写真的很麻烦，答应我，咱都研究生了，别再用word写论文了，豪嘛！

1.1 latex+vscode下载与搭建

俺知道很多人写论文用的是overleaf，但是我之前用的时候发现了一个问题，字数太多的时候，直接要交钱了，不给我用了，我能受那气嘛，我直接，重新下载一个latex，在论文格式这方面，不让人卡脖子嗷。

latex可以直接在清华源下载，链接如下：

Index of /CTAN/systems/texlive/Images/ | 清华大学开源软件镜像站 | Tsinghua Open Source Mirror

就下这个，最大的。

具体安装教程这里不再赘述，这里给一个安装教程的链接

【LaTex】LaTex的下载与安装（超详细、超简洁）_latex下载-CSDN博客

注意，不要下那个texstudio。

安装完毕以后，直接在开始下载vscode，vscode链接如下：

Visual Studio Code - Code Editing. Redefined

下载完毕以后，安装我就不写了，如果你也是搞深度学习的，但是不会安装vscode，我的建议是，重开。

在扩展里下载一个latex workshop

直接搜latex，应该第一个就是。

下载完毕以后，直接新建一个文件夹，然后再里面新建一个xxx.tex文件，然后复制以下代码到文件里

\documentclass{article}
\begin{document}
Hello, world!
\end{document}

如果执行成功的话，你就会看到他给你新建了一个pdf文件

到这里，你的屎盆子镶金边已经完成一半辣！但是只有这个还是8太行，俺们还是需要画图的，如果只有内容没有图，那论文也不太容易中。

1.2 visio下载

画网络结构图啥的，其实比较好用的是visio，这里为什么是visio，我后面会解释，但是，俺们先下载。

直接微信小程序搜索软件智库，然后直接在软件库里找就完了，这里需要注意的是，一定要按照他那个要求一步一步来，下载完了以后，在安装的时候需要将所有防火墙全部关闭，如果没有关闭的话，破解软件就用不了了。总之，下载完了之后的结果如下：

别问，问就是盗版。

1.3 simpletex下载

在写论文的时候，一大问题就是公式不好搞，如果一个字一个字敲的话，有的专业的同学公式多的，那不知道敲到猴年马月去了，关键是如果要用latex写的话，公式也不好转，所以，俺们直接下载一个simpletex，公式直接截图，就可以转成各种格式，根本不用动手敲，嘿嘿嘿~~~

链接如下：

SimpleTex主页——致力于提供先进的公式识别/文档识别解决方案

这一套整完以后，俺们基本就可以开始套格式了。

1.4 latex模板下载

首先，去IEEE access的官网把他们的latex压缩包下下来，如果你登不上去，我需要告诉你的是，IEEE access官网登录是要挂梯子的，如果你说你没有梯子，我就

 IEEE access官网链接如下：

Preparing Your Article - IEEE Access

 直接在这里就可以找到了，大部分的刊都会在官网上放出latex模板，稍微花点时间都可以找到，我之前还有看见有人在网上卖这个的，我的龟龟，这玩意儿也能卖钱的吗，钱这么好挣？

拿到模板之后，直接解压，在vscode里打开，就能看见如下：

现在，你的任何改动，只需要写在accesss.tex文件里，然后一点运行，就可以在pdf里直接生成了，举个例子，我要把名字改成我的标题，即Head detection based on improved YOLOv8，首先，在pdf里找到你要改的地方，然后按住ctrl键，点击你要改的地方，左边的access.tex文件就会自动跳转到那里，即便你看不懂代码也没关系。

改完之后，一按运行，pdf里就改完了，怎么样，是不是非常简单。

如果以上操作都没什么问题，俺们就可以开始下一步了。

二. 科研绘图

2.1 文字部分直接往模板里面套，这都没什么可说的，但是绘图这里可能第一次写论文的会犯错，因为大多数出版社对论文中的图片清晰度是有要求的，不是随便画画就可以的，这就导致了一个问题，大多数同学如果只会用ppt画图，不调整格式的话，大多数出来的图分辨率都不够，怎么办呢，欸，别急，这里有三种绘图方法，一种是用PPT画图，一种是用visio画图，还有一种是用matplotlib配合seaborn画图，三种方法都能保证你出来的图，清晰度一定是够的。

2.1 ppt绘图

相信大多数同学首先用的指定都是ppt绘图，但是ppt画出来多半图像清晰度不够，怎么办呢，欸，别急，只需要小小调整一下即可。

举个例子，上面这张图，如果我要把他放到我的论文里，那么首先，我们需要调整图像整体的分辨率大小，依次点击[文件]→[更多]→[选项]→[高级]→[默认分辨率]，这里一般杂志社要求都是300dpi，我都是调到最高，这里有个高保真的选项。

然后，依次点击[设计]→[幻灯片大小]，这里把宽度和高度调整到原本的4倍即可，比如我的是

 所以我应该把他调整到135.468×76.2，一般在这个尺寸下，出来的图应该是远超300dpi的。

我们可以试一下这一张的图尺寸到底有多大。

依次点击[文件]→[另存为]→[选项]→[桌面]→[保存类型]→[PNG可移植网络图形格式]

可以看到，我们这张图的尺寸已经到了5120*2880，远不止300dpi。

2.2 visio绘图

这里以controlnet的结构图为例

 把图画完以后，依次点击[文件]→[另存为]→[选项]→[桌面]→[保存类型]→[可移植网络图形格式]

你会看到如下界面

如果你不知道该怎么弄的话，我的建议是调大点，给自己留一点裕量，点击分辨率，自定义

 

 可以看到，我们保存的图的尺寸也是远超要求的，后期如果还需要什么其他的调整，我们也有裕量。

2.3 matplotlib+seaborn绘图

这里，如果有一定python基础的人，肯定知道matplotlib是什么，那么你根本不用担心图像尺寸大小的问题，在代码里一个参数就可以指定图像的dpi，根本不用担心尺寸不够，但是如果代码能力确实不行，那么我的建议是，gpt，当然了，这里我还是把seaborn的官方文档放在下面，以防大家还不知道seaborn能画哪些图。

seaborn: statistical data visualization — seaborn 0.13.2 documentation (pydata.org)

 如果你是有一定基础的人， 那么其实你只需要看官方文档，那么大部分的图你就都可以画了，这里我针对的是完全没基础的人，你直接gpt就完了，跟他说，你要用seaborn画一个什么什么图，我在这里以上面那张图为例，我需要画四个折线图，那么代码如下：

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

precision_data = pd.read_excel('new1.xlsx')
recall_data = pd.read_excel('new2.xlsx')
mAP50_data = pd.read_excel('new3.xlsx')
mAP50_95_data = pd.read_excel('new4.xlsx')

fig, axes = plt.subplots(2, 2, figsize=(14, 7),dpi=600)

sns.lineplot(ax=axes[0, 0], data=recall_data)
axes[0, 0].set_title('Recall(B)')
sns.lineplot(ax=axes[0, 1], data=mAP50_data)
axes[0, 1].set_title('mAP50(B)')
sns.lineplot(ax=axes[1, 0], data=mAP50_95_data)
axes[1, 0].set_title('mAP50-95(B)')
sns.lineplot(ax=axes[1, 1], data=precision_data)
axes[1, 1].set_title('Precision(B)')

plt.savefig('new1.png')
plt.tight_layout()
plt.show()

可以看到，只需要输入你的数据结果的文件地址，他就可以直接把你的四个xlsx文件直接画成折线图，非常方便，且因为我在代码中已经指定了dpi=600，所以这里也不用担心清晰度的问题。

三. 结合IEEE access套模板

其实套模板这里没什么可说的，如果只是文字的话，大家都会套，但是其中如果有图片和表格的话，可能大家会比较麻烦，我可以直接把我套完模板代码给大家看看，参考一下。

\documentclass{ieeeaccess}
\usepackage{cite}
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{algorithmic}
\usepackage{graphicx}
\usepackage{textcomp}

\usepackage{bm}
\makeatletter
\AtBeginDocument{\DeclareMathVersion{bold}
\SetSymbolFont{operators}{bold}{T1}{times}{b}{n}
\SetSymbolFont{NewLetters}{bold}{T1}{times}{b}{it}
\SetMathAlphabet{\mathrm}{bold}{T1}{times}{b}{n}
\SetMathAlphabet{\mathit}{bold}{T1}{times}{b}{it}
\SetMathAlphabet{\mathbf}{bold}{T1}{times}{b}{n}
\SetMathAlphabet{\mathtt}{bold}{OT1}{pcr}{b}{n}
\SetSymbolFont{symbols}{bold}{OMS}{cmsy}{b}{n}
\renewcommand\boldmath{\@nomath\boldmath\mathversion{bold}}}
\makeatother

\def\BibTeX{{\rm B\kern-.05em{\sc i\kern-.025em b}\kern-.08em
    T\kern-.1667em\lower.7ex\hbox{E}\kern-.125emX}}

%Your document starts from here ___________________________________________________
\begin{document}
\history{Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.}
\doi{10.1109/ACCESS.2024.0429000}

\title{Head detection based on improved YOLOv8}
\author{\uppercase{jason wu}\authorrefmark{1}}

\address[1]{National Institute of Standards and
    Technology, Boulder, CO 80305 USA (e-mail: wuj400646@gmail.com)}

\corresp{Corresponding author: jason wu (e-mail: wuj400646@gmail.com).}


\begin{abstract}
    Head detection is an important research direction in the field of modern object detection. It can not only be used to check whether students are sleeping with their heads down in class, but also to determine whether the learning atmosphere in a class is good or not. It can also be used in public places to judge the flow of people. In short, research on head detection is still necessary. The traditional head detection methods are slow and have low accuracy. We propose a new head detection algorithm by inserting three CBAM attention modules at the connection between the backbone network and neck network of YOLOv8. Then, we conducted experiments on the SCUT-HEAD head detection dataset. The experimental results show that our model is very good, fast, accurate, small, and completely misses YOLOv8. Therefore, we believe that the model proposed in this paper is of great significance in increasing learning pressure for primary school students and preventing them from not attending early 8th class.
\end{abstract}

\begin{keywords}
    YOLOv8, CBAM, Head Detection, Elementary School Students.
\end{keywords}

\titlepgskip=-21pt

\maketitle

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Introduction}
\label{sec:introduction}
\PARstart{H}{ow} important is head examination, how elementary school students are not serious in class, and how college students skip classes aggressively? I won't go into detail here, research is always meaningful.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Related work}
What methods are used for traditional head detection, such as RGB images, deep learning, machine learning, or direct heat map analysis? Please read more papers, reviews, and specify what can be written. In short, I think the above work is average and not as good as my own work.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Algorithm design}
\subsection{Introduction to YOLOv8 Algorithm}
The YOLOv8 model is really great, but what's good about it is that it's small and has an excellent structure. It has both YOLOv7's C2F structure and YOLOv6's decoupling head, as well as distributed focus loss and even YOLOX's tricky operation of closing data augmentation 10 rounds in advance. Overall, it's very advanced, except for its own name, everything else is sewn together.

\subsection{Shortcomings and Improvement Strategies of YOLOv8}
YOLOv8 is just like that\cite{b1}. I also think his model is average, it's just average. I don't think its feature extraction is deep enough, so it's just passed on to the neck network to start the feature pyramid. I think YOLOv8 is great. My suggestion is to add attention. If you don't work hard in your youth\cite{b2}, focus on graduation.

\subsection{JSON-YOLOv8}
Let's start showing how we improved it directly. I'll just draw a hand drawn improvement diagram here, without going into detail\cite{b3}.

\Figure[t!](topskip=0pt, botskip=0pt, midskip=0pt)[height=4cm,width=10cm]{123.png}
{ \textbf{this is a figure.}\label{fig1}}

Let me introduce again what is CBAM

\Figure[t!](topskip=0pt, botskip=0pt, midskip=0pt)[height=2cm,width=7cm]{234.png}
{ \textbf{this is a figure.}\label{fig2}}

Just read the paper directly here, push forward the formulas and principles of CBAM, and end.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Experiment and Results}
\subsection{Experimental Configuration}
The experimental setup for this experiment is quite high-end. I bought an Alienware m15r7 in 2022, with a 12th Gen i7 and a 3060 (6GB). Someone on Xianyu offered to buy it from me for 4000 RMB, which left me stunned because 4000 RMB is not even a fraction of what I paid for it\cite{b4}.

Now, let's talk about the operating system: Windows 11, PyTorch 2.0.1, Python 3.10.8. The input image size is 640x640, and we only run for 100 epochs. As for the learning rate, weight decay coefficient, and so on, just refer to the code. I won't go into details here.

\subsection{Evaluation coefficient}
Introduction to Precision, Recall, Average Precision (AP), mean Average Precision (mAP), and F1 Score

For example:
\[ \text{mAP} = \frac{1}{K} \sum_{i=1}^{K} \text{AP}_i \]
where \( K \) is the total number of detected object classes, and \( \text{AP}_i \) is the Average Precision for the \( i \)-th class.

\subsection{Ablation Study}
In this experiment, we made a small modification and conducted experiments with and without the modification. Let the original model be the baseline. The modified model in this paper is named JSON-YOLOv8. Here we can present the data results. I'll directly show the figures.

\Figure[t!](topskip=0pt, botskip=0pt, midskip=0pt)[height=7cm,width=14cm]{new1.png}
{ \textbf{this is a figure.}\label{fig3}}

From this picture, our model looks really great, hahaha

\subsection{Comparative experiment}
I'm looking at how our results compare to other models

\Figure[t!](topskip=0pt, botskip=0pt, midskip=0pt)[height=7cm,width=7cm]{qweqweqw.png}
{ \textbf{this is a figure.}\label{fig4}}

In short, those with higher accuracy are bigger than me, while those with lower accuracy are not even as accurate as me. Reasoning faster than me has lower accuracy, reasoning slower than me, let alone that. In summary, my model is the most amazing.

\subsection{Comparative experiment}
I'll just put a random picture here. I haven't really worked on this experiment, hehe

\Figure[t!](topskip=0pt, botskip=0pt, midskip=0pt)[height=5cm,width=14cm]{sewsss.png}
{ \textbf{this is a figure.}\label{fig5}}

From the graph, it is evident that our model is indeed much stronger than other models. With the graph as evidence, it is better not to explain.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Summary}
Based on a thorough analysis of current head detection methods, we propose a new head detection algorithm, namely JSON-YOLOv8. We have added a CBAM attention to the algorithm, and after adding it, the model performs well with significant performance improvement.
In order to verify our effectiveness, we also conducted experiments on public datasets, and the experimental results proved that our model outperforms everything, and other models are completely unmatched.
Finally, our future research direction may also focus on those who cross the street and run red lights. If you don't like my post, I can directly catch these children by conducting head tests.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\begin{thebibliography}{00}

    \bibitem{b1} G. O. Young, ``Synthetic structure of industrial plastics,'' in \emph{Plastics,} 2\textsuperscript{nd} ed., vol. 3, J. Peters, Ed. New York, NY, USA: McGraw-Hill, 1964, pp. 15--64.

    \bibitem{b2} W.-K. Chen, \emph{Linear Networks and Systems.} Belmont, CA, USA: Wadsworth, 1993, pp. 123--135.

    \bibitem{b3} J. U. Duncombe, ``Infrared navigation---Part I: An assessment of feasibility,'' \emph{IEEE Trans. Electron Devices}, vol. ED-11, no. 1, pp. 34--39, Jan. 1959, 10.1109/TED.2016.2628402.

    \bibitem{b4} E. P. Wigner, ``Theory of traveling-wave optical laser,'' \emph{Phys. Rev}., vol. 134, pp. A635--A646, Dec. 1965.

\end{thebibliography}

\begin{IEEEbiography}[{\includegraphics[width=1in,height=1.25in,clip,keepaspectratio]{author1.png}}]{First A. Author} received the B.S. and M.S. degrees in aerospace engineering from
    the University of Virginia, Charlottesville, in 2001 and the Ph.D. degree in
    mechanical engineering from Drexel University, Philadelphia, PA, in 2008.

    From 2001 to 2004, he was a Research Assistant with the Princeton Plasma
    Physics Laboratory. Since 2009, he has been an Assistant Professor with the
    Mechanical Engineering Department, Texas A{\&}M University, College Station.
    He is the author of three books, more than 150 articles, and more than 70
    inventions. His research interests include high-pressure and high-density
    nonthermal plasma discharge processes and applications, microscale plasma
    discharges, discharges in liquids, spectroscopic diagnostics, plasma
    propulsion, and innovation plasma applications. He is an Associate Editor of
    the journal \emph{Earth, Moon, Planets}, and holds two patents.

    Dr. Author was a recipient of the International Association of Geomagnetism
    and Aeronomy Young Scientist Award for Excellence in 2008, and the IEEE
    Electromagnetic Compatibility Society Best Symposium Paper Award in 2011.
\end{IEEEbiography}


\begin{IEEEbiography}[{\includegraphics[width=1in,height=1.25in,clip,keepaspectratio]{author2.png}}]{Second B. Author} (M'76--SM'81--F'87) and all authors may include
    biographies. Biographies are often not included in conference-related
    papers. This author became a Member (M) of IEEE in 1976, a Senior
    Member (SM) in 1981, and a Fellow (F) in 1987. The first paragraph may
    contain a place and/or date of birth (list place, then date). Next,
    the author's educational background is listed. The degrees should be
    listed with type of degree in what field, which institution, city,
    state, and country, and year the degree was earned. The author's major
    field of study should be lower-cased.

    The second paragraph uses the pronoun of the person (he or she) and not the
    author's last name. It lists military and work experience, including summer
    and fellowship jobs. Job titles are capitalized. The current job must have a
    location; previous positions may be listed
    without one. Information concerning previous publications may be included.
    Try not to list more than three books or published articles. The format for
    listing publishers of a book within the biography is: title of book
    (publisher name, year) similar to a reference. Current and previous research
    interests end the paragraph.

    The third paragraph begins with the author's
    title and last name (e.g., Dr.\ Smith, Prof.\ Jones, Mr.\ Kajor, Ms.\ Hunter).
    List any memberships in professional societies other than the IEEE. Finally,
    list any awards and work for IEEE committees and publications. If a
    photograph is provided, it should be of good quality, and
    professional-looking. Following are two examples of an author's biography.
\end{IEEEbiography}

\EOD

\end{document}

出来的结果大概是这样的，最后一张图的原因是我确实没什么内容，所以排版会很难看，大家正常写的时候字数肯定是比我多的，那会儿就没这个问题了，我这里就是写个大概。

那么大概就是这样啦，hxd们，如果已经学的明白儿的了，那就速速开始动手写吧，看教程是看不出SCI的哦