[Tutorail Summary] CVPR 2020 “How to Write” series go through

How to write a good paper? How to write a good review? How to write a good rebuttal? How are reviews used in the decision process?…….. summarizing key points from the 5hr tutorial…

Tutorial website:

CVPR 2020 Tutorial

Tutorial full session recording (youtube):

Timestamps:

• 9:00 — How to write a good paper
• 40:40 — How to write a good review
• >> 1:06:00 — How to be a good reviewer
• >> 1:28:10 — How to write a good review (as a newcomer in the field)
• >> 1:58:20 — Novelty is in the eyes of the beholder
• >> 2:27:30 — Reviewing and teaching how to review
• 2:55:30 — How to write a good rebuttal
• 3:26:20 — How are reviews used in the decision process?
  >> How to be a good reviewer … as part of The System
• >> 3:53:30 — Inside the Decision Process: Notes from CVPR 2020
• 4:24:30 — Panel Discussion

How to write a good paper (9:00)

Only creative original and good papers are effective.

Academic environment: we are not secluded monks, but rather a busy marketplace where everyone is trying to get everyone’s attention.

<Paper Structure: (or maybe the flow of a paper)>

• Stating the problem you are addressing, keeping the audience in mind. WHY SHOULD THEY CARE?
• Brief other solutions, why they are not satisfactory.
• Explain your own solution, compare and say why it's better.
• Talk about related works with similar techniques/experiments that are applied to different questions.

<Paper paragraph organizations>

• A dynamite Introduction — easy to read, easy for anyone to see what your paper is about, what problem it solves, why it is interesting, what is new (and what’s not), what is neat. Do it right upfront.
• Related Work
• Main Idea — often useful to use a toy example to convey the main idea.
  (model design, algorithms, methods,…)
• Experiments — must have a quantitative comparison.
  (for new problems: compare to a naive solution, adapt prior works, or compare to a limited version of your solution)
• Conclusions — what this opens up, how this can change the way we approach computer vision problems.
  “How is the world a better place because of your work?”

How not to end: Future work
Future work = “enumerate the list of things we couldn’t do because we run out of time” “list of good ideas you can do before we get a chance”
Better to talk in generalities: “our work allows us to do the following”, “to tackle this kind of problem”…

<General Writing Tips>

Idea: Keep the reader in mind

What does the reader know so far? What does the reader expect next and why?
→ give a talk, see where people get confused.

Be concise

Consider readership

(most people glance over title >> abstract >> figures…)

Figure captions: self-contained (don’t rely on main text), tell the reader what to notice about the figure

Equations: don’t force the reader to memorize what equation (3) is, say “insert the XYZ term in equation (3) into the ABC in equation (4)” instead of “insert Eq. (3) into Eq. (4)”

Gracious Tone

e.g. “the idea has also been discussed… the reader is urged to review these works for a more complete picture of the field.”

<Quick and easy way to reject a paper>

(Area Chair has the task of rejecting 75% of the paper)

• Do the authors not deliver what they promised?
• Are important references missing? (author not up-to-date for this problem)
• Are the results too incremental? (too similar to previous paper)
• Are the results believable?
• Is the paper poorly written?
• Are their mistakes or incorrect statement?

Often most paper is borderline — some are cockroaches and hard to find a reason to reject; some are puppies with 6 toes…there are blatant flaws.

Good writing is rewriting

Novelty (in terms of reviewing) (1:58:20)

How to write a good rebuttal (2:55:30)

Q and A:

How to write a rebuttal for bad reviews.
……….. cannot draw a summary, maybe just believe in yourself and be calm to accept flaws in your works.

Decision process (2 talks)

3:33:38 — shares an interesting story about missing reviews and chasing after reviewers.

3:42:30 — shares an interesting story about saving papers from bad reviews.

3:45:50 — motivating talk to give a good review.