[转载]Stanford华人教授李飞飞写给她学生的一封信，如何做好研究以及写好PA

 De-mystifying Good Research and Good Papers 

By Fei-Fei Li, 2009.03.01 

 

Please remember this:  

1000+ computer vision papers get published every year! 

Only 5-10 are worth reading and remembering! 

 

Since many of you are writing your papers now, I thought that I'd share these thoughts with you. I probably have said all these at various points during our group and individual meetings. But as I continue my AC reviews these
days (that's 70 papers and 200+ reviews -- between me and my AC partner), these following points just keep coming up. Not enough people conduct first class research. And not enough people
write good papers.  

- Every research project and every paper should be conducted and written with one singular purpose: *to genuinely advance the field of computer vision*. So when you conceptualize and carry out your work, you need to be constantly
asking yourself this question in the most critical way you could – “Would my work define or reshape xxx (problem, field, technique) in the future?” This means publishing papers is NOT
about "this has not been published or written before, let me do it", nor is it about “let me find an arcane little problem that can get me an easy poster”. It's about "if I do this, I could offer a better
solution to this important problem," or “if I do this, I could add a genuinely new and important piece of knowledge to the field.” You should always conduct research with the goal that it could be directly used by many people
(or industry). In other words, your research topic should have many ‘customers’, and your solution would be the one they want to use. 

- A good research project is not about the past (i.e. obtaining a higher performance than the previous N papers). It's about the future (i.e. inspiring N future papers to follow and cite you, N->\inf).  

- A CVPR'09 submission with a Caltech101 performance of 95% received 444 (3 weakly rejects) this year, and will be rejected. 

This is by far the highest performance I've seen for Caltech101. So why is this paper rejected? Because it doesn't teach us anything, and no one will likely be using it for anything. It uses a known technique (at least for
many people already) with super tweaked parameters custom-made for the dataset that is no longer a good reflection of real-world image data. It uses a BoW representation without object
level understanding. All reviewers (from very different angles) asked the same question 

"what do we learn from your method?" And the only sensible answer I could come up with is that Caltech101 is no longer a good dataset.  

- Einstein used to say: everything should be made as simple as possible, but not simpler. Your method/algorithm should be the most simple, coherent and principled one you could think of for solving this problem. Computer
vision research, like many other areas of engineering and science research, is about problems, not equations. No one appreciates a complicated graphical model with super fancy inference
techniques that essentially achieves the same result as a simple SVM -- unless it offers deeper understanding of your data that no other simpler methods could offer. A method in which you have to manually tune
many parameters is not considered principled or coherent.  

 - This might sound corny, but it is true. You're PhD students in one of the best universities in the world. This means you embody the highest level of intellectualism of humanity today. This means you are NOT a technician
and you are NOT a coding monkey. When you write your paper, you communicate  and . That's what a paper is about. This is how you should approach your writing. You need to feel proud of
your paper not just for the day or week it is finished, but many for many years to come. 

 - Set a high goal for yourself – the truth is, you can achieve it as long as you put your heart in it! When you think of your paper, ask yourself this question:  Is this going to be among the 10 papers of 2009 that people
will remember in computer vision? If not, why not? The truth is only 10+/-epsilon gets remembered every year. Most of the papers are just meaningless publication games. A long string of
mediocre papers on your resume can at best get you a Google software engineer job (if at all – 2009.03 update: no, Google doesn’t hire PhD for this anymore). A couple of seminal papers can get
you a faculty job in a top university. This is the truth that most graduate students don't know, or don't have a chance to know.  

- Review process is highly random. But there is one golden rule that withstands the test of time and randomness -- badly written papers get bad reviews. Period. It doesn't matter if the idea is good, result is good, citations
are good. Not at all. Writing is critical -- and this is ironic because engineers are the worst trained writers among all disciplines in a university. You need to discipline yourself:
leave time for writing, think deeply about writing, and write it over and over again till it's as polished as you can think of.  

 - Last but not the least, please remember this rule: important problem (inspiring idea) + solid and novel theory +  convincing and analytical experiments + good writing = seminal research + excellent paper. If any of these
ingredients is weak, your paper, hence reviewer scores, would suffer. 



请记住一点：

 每年有1000篇以上的计算机视觉论文被发表！

 但只有5-10篇值得阅读并记住！

现在你开始写自己的论文，我认为应该与你分享一些看法。当然，一些话或观点可能已经在我们的组会或约谈中说过，但是最近随着我继续

(?)回顾论文（包括的70篇论文和200多篇综述），一些观点也发生了变化。

做好一流研究的人还不够！发表一流文章的人还不够！

-每个研究项目每一篇文章都应该有一个特定的目标：去推动计算机视觉的发展。

 所以当你提出一个概念并完成你的工作时，你应该不停的严格的问你自己这样一个问题-“我的工作在将来能不能定义或改善视觉里的一个问

题、领域或技术”。这也就意味着发表论文不再是因为“这个问题之前还没有被发表过或写过，那我来做”，也不是“去找别人都没有注意到

的问题从而更容易发文章”。而是“如果我做了这个事，我可以提出一种更好的解决方法来解决这个重要的问题”或者是“如果我做了这个事

，我可以为这个领域增加一个确实是新的并且重要的知识”。你也可以用这个目标来指引你的研究就是它可以被许多人或工业上使用。换句话

说就是，你的研究主题应该有很多客户并且你的解决方法也是他们想要的。


-一个好的研究项目不是关于过去（比如，比之前N篇论文获得更好的性能），而是关于未来（比如，可以激发N篇论文，引用并改进你的，N->inf）。



-09年CVPR上提交的一篇论文，它在Caltech101数据集上跑的性能好于其他的绝大部分（95%），但是3个星期内被拒掉了。实际上，它远远的好于在这个数据集上跑出来的最好的结果。但是，为什么这样的一篇论文被拒掉呢？因为它没有告诉我们任何新的东西，并且也不会被别人用来做其他的工作。它使用众所周知的处理方法，只是选用更适合Caltech101数据集的参数，这样并不能反映真实世界的图像集。可以说它只是针对表面表现做技巧处理，而不是从物体理解的水平去解决。所有的审稿人，虽然来自不同的角度，但都问了一个相同的问题，“我们能从你的方法中学到什么新的知识？”。我能想出最明智的回答就是Caltech101数据集不再是一个好的数据集。



-老爱曾经说过：所有的事情都应该尽可能的简洁，但是简洁不等于简单。你提出的方法或算法应该是解决你的问题的，那个尽可能的简洁、明了、通用的方法。计算机视觉的研究，就像很多其他的工程和科学研究领域是围绕着问题的，而不是去解算式。没人欣赏一个复杂的构思极其巧妙但是取得的效果却和简单的SVM方法相同的图像模型,除非这个模型可以更好的理解你的数据并且使用其他任何简单方法都不能做到这一点。一种方法需要手动去调整很多参数是不通用的方法。



-这听起来是有点陈词滥调了，但是都是大实话。记住你现在是一个博士生并在世界上最好的大学之一中读书。这意味着你掌握了当今人类最高的理性思维。这意味着你不只是个技术员或者一个程序猿。当你写论文时，你是在传达知识。你应该学会欣赏所写。你应该为你的论文感到自豪而不是数着日子，而是许多年许多年之后都能够自豪的事。



-为你自己设置一个高的目标，事实是，你能达成目标只要你把心和热情投入！当你思考你的论文时，请自问这样一个问题：这会不会成为09年人们会记住的十篇关于计算机视觉的论文之一？如果不是，为什么不是？记住每年只有十篇左右的文章会被人们记住，大部分论文只是毫无意思的出版游戏。也许在简历上罗列一串论文会让你获得一个Google软件工程师的职位。但其实一两篇很有创意的论文会使你在顶尖大学得到一份科研工作。事实上这是很多学生不知道的，或者是没有机会知道的。



-审核过程是随机的。但是一条金条玉律是--写作很差的论文只会得到差的审稿评价。不论你想法、结果、引用都是多么 好。毫不客气的说，写作是非常重要的，但是这对于工科学生来说好像有点讽刺，工科生在任何大学里s所有学科里受到的写作训练都是很少的。你需要自己有意去培养：花一点时间用来写作，深入想想关于如何写作，并且一遍又一遍的写直到达到你能想到最好的程度。



-最后，也是很重要的，请记住这样一个原则：重要的问题（或激发的灵感）+可靠的新方法+有足够说服力和有分析的实验+好的写作=好的研究+优秀的论文。如果其中任何一个环节薄弱，你的论文，就会遭遇不幸。。。