Skip to content

Latest commit

 

History

History
241 lines (159 loc) · 8.33 KB

Awesome-resources-for-better-writing-of-computer-vision-papers.md

File metadata and controls

241 lines (159 loc) · 8.33 KB
Raw

Advice for better writing

This page is an effort to gather the advice I can find on how to write a good paper, with as focus on top-tier computer vision conferences (CVPR, ECCV, ICCV)

Follow guidelines

Read author and reviewer guidelines to know how not to get rejectet and how to get accepted

CVPR 2018 reviewer's guidelines

What to Look For

Minor flaws can be corrected and shouldn't be a reason to reject a paper

Embrace novel, brave concepts

The fact that a proposed method does not exceed the state of the art accuracy on an existing benchmark dataset is not grounds for rejection by itself.

Acceptance and rejection decisions should not be determined solely by the method's raw performance

It is important to weigh both the novelty and potential impact of the work alongside the reported performance

Each paper that is accepted should be technically sound and make a contribution to the field.

Writing advice from those who know best !

Jitendra Malik's write a good paper slides 2018

Possible introduction style: What did you do (Fig. 1), How did you do it? (Fig. 2)

The best way to write a paper is to first give a talk on it

Introduction section: The most important section of a paper. For me, once I have finished reading the introduction, I have formed an opinion of whether to accept or reject the paper

Bill Freeman's 2018 slides on writing a good CVPR submission

I can’t stand “future work” sections. It’s hard to think of a weaker way to end a paper

Omit needless words

Figure captions should be self-contained and the caption should tell the reader what to notice about the figure

There are perceived pressures to over-sell, hide drawbacks, and disparage others’ work. Don’t succumb.

(and previous 2014 version)

Practical tips

LaTeX

Structuring your latex project, and per-section structure recommendations (introduction, related work, ...) Typically - a *main* file named `00submission_id.tex` - references in the main file to each section is a **separate file**

00submission_id.tex

\documentclass[10pt,twocolumn,letterpaper]{article}

\usepackage{times}
\usepackage{epsfig}
\usepackage{graphicx}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage[dvipsnames]{xcolor}
\usepackage[numbers,sort,compress]{natbib}
\begin{document}

\title{My Awesome Paper}

\author{
 Me \textsuperscript{1} \thanks{This work was performed during an internship at Microsoft.}
 \qquad Author 2 \textsuperscript{2}
 \qquad Author 3 \textsuperscript{2}\\
 \\ \\
 {\normalsize \textsuperscript{1}My affiliation \textsuperscript{2} Other affiliations} \\
}

\maketitle

\input{abstract}
%%%%%%%%% BODY TEXT
\input{introduction}
\input{relatedwork}
\input{method}
\input{experiments}
\input{conclusion}
%%%%%%%%% BODY TEXT
\clearpage \newpage
{\small
\bibliographystyle{ieee_fullname}
\bibliography{egbib}
}

abstract.tex

\begin{abstract}
We did something awesome and wrote about it.
\end{abstract}

introduction.tex

\section{Introduction}
% Why the task you address is important, (if it is a new problem) problem statement, and applications

% Challenges and the overview of pioneering works 

% Remaining limitations

% Our method in a glance 

% Clearly stated contributions
Our contributions can be summarized as follows:
\begin{itemize}
	\item{We propose a new method for ...}
	\item{We demonstrate...}
	% \item{We further present...}
\end{itemize}

% The summary of the results

relatedwork.tex

\section{Related Work}
\label{sec:related}

% Define the scope of the related work, grouping the works by theme (tackling the same problem, or using similar methods, ...)

We first review the literature on {relevant field}.
Then, we focus on methods which {relevant approaches}, and {other related topic}.


\paragraph{relevant field.}
Most approaches in the literature tackle the problem of estimating either hand or object pose, separately.

% ...

...

One color per author for in-document comments

Assign colors to authors in main latex file:

\newcommand{\Yana}[1]{\textcolor{red}{#1}}
\newcommand{\Author2}[1]{\textcolor{green}{#1}}
% Add author with custom color custom colors
\definecolor{mycolor}{RGB}{219, 122, 48}
\newcommand{\Author3}[1]{\textcolor{mycolor}{#1}}

Add comments in text

\Yana{I think we should ...}
\Author2{Should we keep ... ?}
Bibliography (bibtex)
  • use natbib so that references are sorted in ascending number ([13, 3, 34] --> [3, 13, 34])
\usepackage[numbers,sort,compress]{natbib}
  • Make references homogeneous
    • Good sources of bibtex files are usually the publication websites
    • To minimize efforts to make homogeneous, remove all unecessary fields (month, page numbers, abstract, url, ...) from the bibtex
    • Check that all names are in full letters (no initials except for middle names)
    • Everytime a conference is referenced, it should have exactly the same name, typically the one the conference uses in their own offical bibtex ({Advances in Neural Information Processing Systems} for NeurIPS, {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)} for CVPR, {The European Conference on Computer Vision (ECCV)} for ECCV), {The IEEE International Conference on Computer Vision (ICCV)} for ICCV, ... (Can't get it wrong by taking the official version !)
    • in a final pass, look at how the references are rendered in the PdF, and check for consistency
Appendix sections with Letter numbering 
\usepackage{titletoc}
\renewcommand{\thesection}{\Alph{section}}
	
% ...
\begin{document}
\section{First appendix section}

ArXiv

Cleaning latex for ArXiv

ArXiv compiles latex and makes the source files downloadable. You should therefore clean them before you submit !

  • delete any unused source files and figures
  • use arxiv-latex-cleaner to create ready-to-zip-and-submit latex files

Advice for better rebuttals

How we write rebuttals, by Devi Parikh

Practical tips

Major tips:

  • Acknowledge reviewer effort : Thank the reviewers for their effort
  • Remind reviewers of the positive aspects: "Provide a summary of the reviews, highlighting positive things that reviewers said about your work."
  • Meaningful order of answers : "Start with the biggest concerns that you have good answers for and work your way to less clear-cut responses and minor points."
  • Tone should be receptive and enthousiastic overall
  • Point to missed points "If something a reviewer asked for was already in the paper, say so. Give them line/Table/Figure numbers, and then restate it in the rebuttal."

Minor cute bonuses:

  • Color code reviewers
  • Mention that requested experiments can not be provided if venue doesn't allow it