Presentation on theme: "Microsoft Research Vision Technology A Perspective on doing Effective Research* Sing Bing Kang MSR Redmond * Caveat: My (rather biased) perspective!"— Presentation transcript:
Microsoft Research Vision Technology A Perspective on doing Effective Research* Sing Bing Kang MSR Redmond * Caveat: My (rather biased) perspective!
Microsoft Research Vision Technology Why do research? Passion for discovering new things Making the world a better place (and for company) Yeah, right: Opportunity to “show off” Prestige Other perks: –Visiting exotic places (conferences) –Flexible working hours
Microsoft Research Vision Technology Starting a project So you have a great idea You go through the usual phases: –Literature search (anybody else done this?) –Idea refinement –Implementation –Write up An important component: Documenting
Microsoft Research Vision Technology Documenting Do this often (as written notes or in LaTex) When: –Literature search –Defining new ideas/concepts, motivation –Explaining experiments and results Keeps project in perspective Useful for patent filing
Microsoft Research Vision Technology Observed progression deadline Amount of documentation Time
Microsoft Research Vision Technology More desirable progression deadline Amount of documentation Literature search Refining, experimenting Defining new ideas/concepts Distribute load! Time
Microsoft Research Vision Technology Documenting: Approaches Top-down (skeleton, then fill) Bottom-up (write details, then reorg) Linear (progressing from begin to end) Any combination of the above
Microsoft Research Vision Technology Documenting in the beginning Goals of project Motivation Prior relevant work –Critical review –Point out similarities and differences –Why your proposal is new and better More critical for collaboration
Microsoft Research Vision Technology Literature search An absolute must in the beginning Useful resources: –USC’s Annotated Computer Vision Bibliography –CMU’s Computer Vision Homepage –Univ. of Edinburgh’s CVonline –NEC’s CiteSeer (the ultimate in egosurfing!) Links in
Microsoft Research Vision Technology Documenting throughout The document should be evolving Ideas and concepts (with figures) Results Forces you to review project constantly
Microsoft Research Vision Technology Typical contents of final paper Abstract Introduction Body Experimental results and interpretation Conclusions/summary, future work
Microsoft Research Vision Technology Paper: Abstract Purpose: Give overall picture, entice reader Do: –Be brief –Give relevant high-level descriptions –Indicate why your work is new, better Don’t: –Cut and paste from introduction or conclusions –Give details
Microsoft Research Vision Technology Paper: Introduction Motivate your work Show you are well-read (prior work) Indicate why your work is new and better (not just different) Provide a roadmap to your paper (depends)
Microsoft Research Vision Technology Paper: Body If concept/system is complex, provide a block diagram, e.g.: Inputs Substep 1 Substep 2 Conditions Step 3…
Microsoft Research Vision Technology Paper: Results Present results clearly –Graphs properly labeled, check font size –Be careful with color figures (contrast) –Always define abbreviations –Check for conflicting symbols Don’t just describe, interpret!
Microsoft Research Vision Technology Paper: Discussion Describe alternative techniques Elaborate effects of assumptions Reveal problems (breaking points), explain Suggest applications
Microsoft Research Vision Technology Paper: Conclusions Repeat what is great about work Mention how general the work is Hint on future work
Microsoft Research Vision Technology General tips on writing (or How to not annoy reviewers) Always check spelling One paragraph, one main point Avoid long, convoluted sentences Have connecting statements between sections Tables, figures must be described in text Check diction (get someone to proofread)
Microsoft Research Vision Technology Collaborative effort Research projects often involve >= 2 people Requires regular meetings Documentation very important! Be inventive: –If data not forthcoming, generate plausible ones for your own progress –Show results effectively (e.g., intermediate 3D data -> VRML)
Microsoft Research Vision Technology Giving presentations Content is as important as delivery Plan content well: –What is your primary message? –Who is your audience? –How long is your talk?
Microsoft Research Vision Technology Level of detail (Adapted from Takeo Kanade’s advice) Time Level of detail Enough high-level stuff to enable most to understand Some details to satisfy those familiar with area, and to show you really know your work
Microsoft Research Vision Technology Giving presentations: Slides Slides should typical not exceed 6-8 lines Very bad: –Photocopied directly from paper –Unintelligible scribbles –Distracting animations, “cute” irrelevant graphics Readability test: Measure width of screen W, step back 8W.
Microsoft Research Vision Technology Giving presentations: Delivery If unfamiliar, write notes and memorize Provide linking statements for next slide Using a laser pointer? To reduce shaking: –Keep upper arm close to body –Move only lower arm and hand
Microsoft Research Vision Technology Interaction with audience Keep eye contact with audience Don’t mumble; speak loudly and clearly Don’t block audience’s view If question is complex, defer to after talk Relax: People want you to succeed
Microsoft Research Vision Technology Presentations: Observe and practice! When attending others’ presentations, observe. Note what you like. Note what you don’t like. Analyze! Most important: Practice, practice, practice!
Microsoft Research Vision Technology Attending conferences Make full use of it Talk to other people Attend talks/posters, ask questions Notice how others deliver talks/posters, how they handle questions
Microsoft Research Vision Technology Finally (seriously)… A bad attitude can hinder your career Consider these: –Word gets around quickly –People won’t lie to preserve your reputation Own up to what you don’t know Use the “reasonable person” principle!
Microsoft Research Vision Technology Sample resources Guides to Research (http://www.useekufind.com/uiresear.htm, Guide for Writing Research Papers (http://webster.commnet.edu/mla.htm)http://webster.commnet.edu/mla.htm Giving Presentations (http://bmrc.berkeley.edu/courseware/cs160/spring 99/Lectures/14-Presentations/sld001.htm)
Microsoft Research Vision Technology Handouts “The science of scientific writing” “A letter to research students” “How to give a good research talk” “How to be a good graduate student/advisor” “The general pattern of the scientific method” “Computer vision methodologies”