how to give an IW talk

advertisement
How to Give an
Excellent IW Talk
Thomas Funkhouser
Princeton University
Why This Title?
Why This Title?
There have been previous talks on this subject:
 How to give a great research talk, Ju
 How to give a good talk, Fleet and Hertzmann
 How to give a good research talk, Jones et al.
 How to give a good research talk, Zeller
 How to give a good research talk, Scott
 Give a good research talk, Xu
 How to give good presentations, Carlton & Jacob
 How to give a bad talk, Paterson
Why This Title?
There have been previous talks on this subject:
 How to give a great research talk, Ju
 How to give a good talk, Fleet and Hertzmann
 How to give a good research talk, Jones et al.
 How to give a good research talk, Zeller
 How to give a good research talk, Scott
 Give a good research talk, Xu
 How to give good presentations, Carlton & Jacob
 How to give a bad talk, Paterson
At Princeton, we strive for EXCELLENCE!
(why settle for just “good” or “great”?)
Motivation
Giving excellent research talks helps you …
 Communicate your ideas to others
 Get useful feedback to guide your next steps
 Encourage people to learn more about your work
 Garner interest in your ideas
Giving excellent talks
is one of the
important skills to learn
as a student
www.michaelpeggs.com-
Goal of this Information Session
Teach you how to design excellent research talks
 Stimulate critical thinking about talk design
 Enhance awareness of common pitfalls
 Provide concrete guidance for your upcoming talk
Main takeaway:
guidelines
for designing
excellent talks
Plan for this Information Session
Part 1: Discuss basic principles of slide design
 Colors, fonts, text, figures, animations, etc.
 Layouts, context, consistency, etc.
Part 2: Provide a roadmap for how to organize a talk
 Suggest a particular flow of ideas
 Explain why that flow of ideas is good
 Discuss alternatives
Part 3: Show example talks from past semesters
 Flip through slides of some recent talks
 Discuss design principles that permeate these talks
 Evaluate what works well and what does not
Plan for this Information Session
Part 1: Discuss basic principles of slide design
 Colors, fonts, text, figures, animations, etc.
 Layouts, context, consistency, etc.
Part 2: Provide a roadmap for how to organize a talk
 Suggest a particular flow of ideas
 Explain why that flow of ideas is good
 Discuss alternatives
Part 3: Show example talks from past semesters
 Flip through slides of some recent talks
 Discuss design principles that permeate these talks
 Evaluate what works well and what does not
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Choose colors carefully
(what you see on your laptop is
not what your audience sees)
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Dark letters against a light background
Light letters against a dark background
Do this
Part 1: Slide Design
Basic principles of slide design
Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
It is a common mistake is to highlight a word
by making it lighter on a white background
It is a common mistake is to highlight a word
by making it darker on a dark background
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
28pt: Arial, Cambria, Times, Courier,
24pt: Arial, Cambria, Times, Courier, Monoty
22pt: Arial, Cambria, Times, Courier, Monotype
20pt: Arial, Cambria, Times, Courier, Monotype
18pt: Arial, Cambria, Times, Courier, Monotype
16pt: Arial, Cambria, Times, Courier, Monotype
14pt: Arial, Cambria, Times, Courier, Monotype
12pt: Arial, Cambria, Times, Courier, Monotype
Choose fonts that are easiest to read
Part 1: Slide Design
Basic principles of slide design
 Colors
Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Ann Marie Carlton & Daniel Jacob
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do any of this
Part 1: Slide Design
Basic principles of slide design
 Colors
Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
24pt fonts or larger
(depending on size of room)
Sans serif fonts
Arial
Helvetica
Tahoma
Verdana
Do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
Text
 Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
One message per slide
Only short sentences or phrases
Enough text to follow the story
Not all the details
Do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
(Unless you plan to define every variable
and step through every equation)
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
Equations
 Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Do this
(Only include equations
that help tell your story)
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Michael Li
Don’t do this
Michael Li
Do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
Plot Title
Y Label (units)
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Method 1
Method 2
Method 3
X Label (units)
Do this
Part 1: Slide Design
Basic principles of slide design
Plot Title
Y Label (units)
 Colors
 Fonts
 Text
 Equations
Figures
 Animations
 Layout
 Context
 Consistency
 etc.
Method 3
X Label (units)
Or this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
Animations
 Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
Animations
 Layout
 Context
 Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
Animations
 Layout
 Context
 Consistency
 etc.
Do this
(animated visualization)
http://www.pptalchemy.co.uk/
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
Animations
 Layout
 Context
 Consistency
 etc.
Do this
(demonstration)
Luke Li
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Layout
 Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Layout
 Context
 Consistency
 etc.
I like to have main point of slide across top,
supporting points in bullets, and
a figure filling most of the slide
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Layout
 Context
 Consistency
 etc.
I like to have main point of slide across top,
supporting points in bullets, and
a figure filling most of the slide
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Layout
 Context
 Consistency
 etc.
I like to have main point of slide across top,
supporting points in bullets, and
a figure filling most of the slide
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Layout
 Context
 Consistency
 etc.
Rarely two points per slide
Rarely three levels
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
Context
 Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Point
 Layout
being
Context discussed
 Consistency
 etc.
Part of Talk
Topic sentence
Content
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
Consistency
 etc.
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
Consistency
 etc.
Don’t do this
Part 1: Slide Design
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
 Layout
 Context
Consistency
 etc.
Don’t do this
Part 1: Slide Design
Part of Talk
Basic principles of slide design
 Colors
 Fonts
 Text
 Equations
 Figures
 Animations
Point
 Layout
being
 Context discussed
Consistency
 etc.
Topic sentence
Content
Do this
Plan for this Information Session
Part 1: Discuss basic principles of slide design
 Colors, fonts, text, figures, animations, etc.
 Layouts, context, consistency, etc.
Part 2: Provide a roadmap for how to organize a talk
 Suggest a particular flow of ideas
 Explain why that flow of ideas is good
 Discuss alternatives
Part 3: Show example talks from past semesters
 Flip through slides of some recent talks
 Discuss design principles that permeate these talks
 Evaluate what works well and what does not
Part 2: Flow of Ideas
Goal: organize your talk with a flow of ideas that …
 Makes a point
 Teaches the listener something they remember
 Tells a story
Part 2: Flow of Ideas
Goal: organize your talk with a flow of ideas that …
 Makes a point
 Teaches the listener something they remember
 Tells a story with a logical flow of ideas
Flow of Ideas
 Each idea should follow directly from previous one
 Details should be omitted unless necessary for story
 Level of detail should be tailored to audience
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Part 2: Flow of Ideas
A flow that works for almost any talk:
Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Motivation:
Establish importance of your
research area or topic
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Goal:
Define problem:
• Specify inputs and outputs
• Define assumptions
• Describe desirable properties
• Provide concrete example(s)
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Related work:
Provide taxonomy of previous
approaches to achieve your goal.
For each one, briefly describe the
key idea and explain why it doesn’t
achieve your goal completely
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Approach:
Present novel idea (Eureka!)
Explain why it is a good idea
(describe rationale)
Provide simple example showing
how your idea achieves your goal
in situation where previous
approaches would not
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Implementation:
Provide overview:
• System organization
• Sequence of steps (flow chart)
For most important steps (or issues):
• Challenge
• Approach
• Implementation
• Results
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
Results
 Conclusions
 Future work
 Acknowledgments
Results:
Present results of experiments
aimed at testing whether met goal
For each experiment:
• State goal of experiment
• Describe data set(s)
• Describe experiment setup
• Describe evaluation metric(s)
• Present results
• Discuss success/failure cases
• Explain implications
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
Conclusions
 Future work
 Acknowledgments
Conclusions:
Summarize key points
Restate main results
Describe implications
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
Future work
 Acknowledgments
Future work:
This semester: remaining steps,
pending evaluations, etc.
Next projects: questions to
investigate in a follow-up study
Long-term: how work could affect
direction of your field 
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
Acknowledgments
Acknowledgments:
Sources of code, data, etc.
Student collaborators
Faculty adviser
Funding
etc.
Part 2: Flow of Ideas
A flow that works for almost any talk:
 Motivation
 Goal
 Related work
 Approach
 Implementation
 Results
 Conclusions
 Future work
 Acknowledgments
Plan for this Information Session
Part 1: Discuss basic principles of slide design
 Colors, fonts, text, figures, animations, etc.
 Layouts, context, consistency, etc.
Part 2: Provide a roadmap for how to organize a talk
 Suggest a particular flow of ideas
 Explain why that flow of ideas is good
 Discuss alternatives
Part 3: Show example talks from past semesters
 Flip through slides of some recent talks
 Discuss design principles that permeate these talks
 Evaluate what works well and what does not
Part 3: Example
Let’s look an example talk and discuss how/whether
it follows the proposed principles …
Finding Surface Correspondences
Using Symmetry Axis Curves
Tianqiang Liu
Vladimir G. Kim
Thomas Funkhouser
Princeton
University
Motivation
Computing maps between surfaces is
critical for many scientific applications
Medicine
Paleontology
Facial
Recognition
Goal
Compute a map between two surfaces
 Injective
 Non-rigid
 Smooth
 Shape preserving
 Align semantic features
 Automatic
 Efficient
Previous Work
Possible approaches:
 Search combinations of feature correspondences
 Align after embedding in high-dimensional space
 Align with a low-dimensional transformation
Previous Work
Possible approaches:
Search combinations of feature correspondences
 Align after embedding in high-dimensional space
 Align with a low-dimensional transformation
Branch and bound
Priority-driven search
etc.
Least squares conformal map
aligning corresponding feature points
Feature points
[Zeng et al., 2008]
Previous Work
Possible approaches:
 Search combinations of feature correspondences
Align after embedding in high-dimensional space
 Align with a low-dimensional transformation
Flip, reorder,
closest points, etc.
Eigenfunctions of the Laplacian
[Lombaert et al. 2011]
Previous Work
Possible approaches:
 Search combinations of feature correspondences
 Align after embedding in high-dimensional space
Align with a low-dimensional transformation
Mobius Voting
[Lipman et al., 2009]
Previous Work
Possible approaches:
 Search combinations of feature correspondences
 Align after embedding in high-dimensional space
Align with multiple low-dimensional transformations
Blended Intrinsic Maps (BIM)
[Kim et al., 2011]
Problem
Current methods focus mainly on preserving
local shape features
Observations
1. Preserving global shape features is important
for most real-world correspondence problems
2. Reflection symmetry axes are stable, detectable,
alignable, global shape features
Aligned
Symmetry
Axes
Approach
Detect symmetry axes, align them, and then
extrapolate correspondence to rest of surfaces
Aligned
Symmetry
Axes
Surface
Correspondence
Computational Pipeline
Symmetry Axis
Detection
Symmetry Axis
Alignment
Correspondence
Extrapolation
Consider only
Genus-0 Surfaces
Computational Pipeline
Symmetry Axis
Detection
Symmetry Axis
Alignment
Correspondence
Extrapolation
Consider only
Genus-0 Surfaces
Symmetry Axis Detection
Given a mesh, extract potential symmetry axes
1. Find symmetry map(s)
2. Extract stationary curve(s)
3. Estimate axis quality
Input Mesh
Symmetry Axis Detection
Given a mesh, extract potential symmetry axes
1. Find symmetry map(s)
2. Extract stationary curve(s)
3. Estimate axis quality
Feature
Points (F)
Blended
Intrinsic
Map (M)
Symmetry Map (M)
Symmetry Axis Detection
Given a mesh, extract potential symmetry axes
1. Find symmetry map(s)
2. Extract stationary curve(s)
3. Estimate axis quality
... where d(p, M(p) = 0
Symmetry Axis
Curve (C)
Distance Function (d(p, M(p))
Symmetry Axis Detection
Given a mesh, extract potential symmetry axes
1. Find symmetry map(s)
2. Extract stationary curve(s)
3. Estimate axis quality
Symmetry Axis
Curve (C)
QAxis(C) = Length(C) ∙
Computational Pipeline
Symmetry Axis
Detection
Symmetry Axis
Alignment
Correspondence
Extrapolation
Consider only
Genus-0 Surfaces
Symmetry Axis Alignment
Given symmetry axis curve(s) for two meshes,
find best pairwise alignment
 For each pair of curves
• For each curve starting point ...
C21
C12
C11
Mesh 1
C13
C22
Mesh 2
Symmetry Axis Alignment
Find best alignment for a pair of starting points
C1i
C2j
Shape
Similarities
𝑄𝐴𝑙𝑖𝑔𝑛 𝐶1 , 𝐶2 , 𝑐 = 1/
𝐷(𝑃1,𝑘 , 𝑃2,𝑘 )
𝑃1,𝐾 ∈𝐶1
Symmetry Axis Alignment
O(mn) algorithm based on dynamic programming
 Dynamic time warping [Marzal et al., 2005]
Input:
Edit graph:
X
Y
Solution:
Y
X
Y
X
X
Y
n
X
m X
m
X
Y
Y
X Y Y
X
X
Y
X X Y
X
Y X X Y
n
Computational Pipeline
Symmetry Axis
Detection
Symmetry Axis
Alignment
Correspondence
Extrapolation
Consider only
Genus-0 Surfaces
Correspondence Extrapolation
Given an alignment between symmetry axes,
extrapolate correspondences to rest of surfaces
1. Find correspondences for extremal feature points
2. Interpolate correspondence to all points
Aligned
Symmetry Axes
Correspondence Extrapolation
Given an alignment between symmetry axes,
extrapolate correspondences to rest of surfaces
1. Find correspondences for extremal feature points
2. Interpolate correspondence to all points
Aligned
Symmetry Axes
Aligned Extremal
Feature Points
Correspondence Extrapolation
Given an alignment between symmetry axes,
extrapolate correspondences to rest of surfaces
1. Find correspondences for extremal feature points
2. Interpolate correspondence to all points
Aligned
Symmetry Axes
Aligned Extremal
Feature Points
Full
Surface Map
Timing
Computational complexity:
 O(F6SlogM + N3 + NFSlogM)




F = # feature points (~5-10)
S = # sample points on mesh (128)
M = # vertices on mesh (~10K)
N = # sample points on axis curve (200)
Computation time, in practice:
 ~1 min for symmetry axis detection (once per mesh)
 ~5 sec for symmetry axis alignment
 ~3 min for correspondence extrapolation
Example Results
Experimental Evaluation
Surface Correspondence Benchmark:
TOSCA
[Bronstein et al., 2008]
SCAPE
[Anguelov et al., 2004]
SHREC Watertight 2007
[Giorgi et al., 2007]
Experimental Evaluation
% Correspondences
Evaluation methodology:
100
90
80
70
60
50
40
30
20
10
0
0
SCAPE
Geodesic
Error
Ours
0.05
0.1
0.15
Geodesic Error
0.2
0.25
Experimental Evaluation
Comparison to Blended Intrinsic Maps (Benchmark):
Experimental Evaluation
Comparison to Blended Intrinsic Maps (SHREC):
Experimental Evaluation
Comparison to Blended Intrinsic Maps (SHREC):
Our method is better for 6 object categories
Experimental Evaluation
Comparison to Blended Intrinsic Maps (SHREC):
and worse for 3 object categories
Limitations
Some failure cases:
Non-descriptive
Symmetry Axes
Different Extremal
Feature Points
Poor Symmetry
Axis Extraction
Summary
Main idea:
 Leverage symmetry detection to
find surface correspondences
Rationale:
 Finding symmetry maps, aligning curves, and
extrapolating correspondences are easier than
finding surface correspondences
Results:
 Slightly better than previous best method (BIM)
on most object classes
Future Work
Short term:
 Improvements in symmetry map detection and
correspondence extrapolation
Long term:
 Partial symmetries, partial matching
 Finding correspondences in collections
 Considering other global shape features
Acknowledgements
Data
 Daniela Giorgi and [email protected] (Watertight’07)
 Drago Arguelov and Stanford University (SCAPE)
 Project TOSCA (Non-Rigid World)
Collaboration:
 Yaron Lipman (early planning of project)
Funding:
 NSF, NSERC, Adobe, Google, Intel
Thank You
Notes on Talk Presentation
Giving a talk is an opportunity
 Your chance to describe why what you did is cool!
 Your chance to convince others you are competent
 Why waste that opportunity by giving a bad talk?
Giving a talk does not have to be nerve-wracking
 You know more about topic than your audience
 Preparation and practice
is very calming
Ann Marie Carlton & Daniel Jacob
Summary
Slide design is important
 Colors, fonts, text, figures, animations, etc.
 Layouts, context, consistency, etc.
Probably best to follow suggested flow of ideas
 Motivation, goal, previous work, approach,
implementation, results, conclusion, future work
 Same flow used by most top researchers in the world
Designing effective talks is hard, but important
 Most people are not good at it, mostly because they
don’t think about the choices they are making
 You are now above the median in knowledge
on this subject 
Acknowledgments
Thanks to the IW coordinators for slides and advice:
 Alan Kaplan and Rob Fish
Thanks to the following people for ideas:
 Ju, Fleet & Hertzmann, Jones et al., Zeller, Scott,
Xu, Carlton & Jacob, Paterson, Fatahalian
Thank You!
Download