Li Feifei is an ox in the field of computer vision at Stanford University who has some advice on writing paper _advice

De-mystifying Good and good papers
by Fei-fei Li, 2009.03.01

Please remember this:
1000+ Computer Vision papers get published every
Only 5-10 are worth reading and remembering!

Since Many of your are writing your papers now, I thought the I ' d share these thoughts and you. I probably have said all of these in various points during our group and individual meetings. But as I continue my AC reviews This is the papers and 200+ reviews-between me and my AC partner), which are foll Owing points just keep coming up. Not enough people conduct-class. And not enough people write good papers.
-Every project and Every paper should is conducted and written with one singular purpose: *to genuinely The field of computer vision*. So if you are conceptualize and carry out of your work, you are need to is constantly asking the yourself this question in the most Tical Way You could– "Would I work define or reshape xxx (problem, field, technique) in the future?" This means publishing papers are not about ' this has not been published or written before, let me do it ', nor is it about Let me find a arcane little problem can get me a easy poster ". It's about "If I do", I could offer a better solution to this important problem, "or" If I does this, I could add a genu Inely new and important piece of knowledge to the field. " You should always conduct and the goal that it could is directly used by many people (or industry). In other words, your topic should have many ' customers ', and your solution is the one would they to use.
-A good 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 for you, N->\inf).
-A CVPR ' submission with a Caltech101 performance A-received 444 (3 weakly rejects) D. This are by far the highest performance I ' ve seen for Caltech101. So why are this paper rejected? Because it doesn ' t teach us anything, and no one would likely is using it for anything. It uses a known technique (at least for many people) with super already tweaked parameters for the dataset tha T 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 does we learn from your?" And the only sensible answer I could come up and is this Caltech101 is no longer a good dataset.
-Einstein used to say:everything should is made as simple as possible, but not simpler. Your Method/algorithm should is the most simple, coherent and principled one you could to the this solving. Computer vision, like many, areas of engineering and science, are 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 this no other simpler methods offer. A method in which your have to manually tune many parameters are not considered principled or coherent.
-This might sound corny and but it is true. You ' re PhD students in one of the best universities in the world. This means your embody the highest level of intellectualism of humanity today. This means your are not a technician and your are not a coding monkey. When you write your paper, you communicate and. That ' s what a paper is about. This is what you should approach your writing. You are need to feel proud of your paper not just for the "day" or week it is finished, but many for many.
-Set a high goal for yourself–the truth are, can achieve it as long as you put your heart in it! When you are your paper, ask yourself this question:is this going to be among the papers of 2009 that people would Remember in computer vision? If not, why? The truth is only 10+/-epsilon gets remembered every year. Most of the papers are just meaningless publication. A long string of mediocre papers on your resume can in best get for Google software engineer Job (if at all–2009.03 up Date:no, Google doesn ' t hire PhD for this anymore). A couple of seminal papers can get your 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 this withstands the test of time and randomness-badly written the papers. Period. It doesn ' t matter if the idea is good, the good of the result is, citations are good. Not on all. Writing is critical--"and" is ironic because engineers are the worst trained the writers among all disciplines in a unive Rsity. You are need to discipline yourself:leave time for writing, and I am deeply about writing, and write it over and over again till It ' s as polished as you can.
-Last but to least, please remember this rule:important problem (inspiring idea) + solid and novel Theory + convinc ing and analytical experiments + good writing = seminal, and excellent paper. If any such ingredients is weak, your paper, hence reviewer-scores, would suffer