How to Lie with Visualization

1
How to Liewith Visualization

• Zoran Constantinescu

2
references

• Al Globus et. al. 14 Ways to Say Nothing with
  Scientific Visualization IEEE Computer, vol.
  27, 1994
• Nahum Gershon Presenting Visual Information
  Responsibly ACM Computer Graphics, vol. 33,
  1999
• Nahum Gershon How to Lie and Confuse with
  Visualization (VisLies) special sessions at
  Siggraph and Visualization conferences

3
outline

• problem definition
• 14 ways to say nothing with SciViz
• examples
• conclusions

4
problem

• Seeing is believing.
• in synthetic imagery this is not always true
• (anybody can vis. anything in any shape/form)
• sources of imperfection
• imperfect presentation

5
imperfect presentation

• can prevent getting the information or reduce the
  rate of absorption and understanding
• or the user get perceive it wrongly
• data can be too complicated to comprehend
• visualization can misrepresent the information

6
14 ways to say nothing with SciViz

• it can be used to produce beautiful pictures
• usually we fail to appreciate the artistic
  qualities of these images
• scientists will use it to understand the data
• techniques to confound (confuse) such activities

7
1. never include a color legend

• many visualization techniques involve assigning
  colors to scalar values
• spoils the beauty of an image
• the viewer may be diverted into contemplation of
  the reality

8
example
3D global view of Mars
9
example
3D global view of Mars
elevation
10
example
MRI scan of human head
11
2. avoid annotation

• used for pointing out features of interest
• used in combination with explanatory text
• promotes clarity of understanding
• undermines the sense of awe and confusion the
  best scientific visualization engenders

12
example
H2O molecule
13
example
H2O molecule
electron density
14
3. never mention error

• visualization techniques might introduce error
• scientists might not be properly impressed if
  mentioning error characteristics
• never imply by word or deed that the technique
  introduces any error
• if the picture looks good, it must be correct

15
4. when in doubt, smooth

• smooth surfaces look much better than numerous
  ugly facets
• can also obscure errors and
• allow users to publish their results earlier
• always strive for the smoothest possible surface
• choose lighting normals to hide sharp edges

16
example
17
5. avoid providing performancedata

• it is completely irrelevant the time it took to
  calculate the picture
• e.g.. hours for ray-casting a isosurface, or
  seconds using marching cubes
• even if it takes longer, it is much smoother

18
6. use stop-frame animation

• each frame of a scientific video usually takes
  from seconds to hours to produce
• ? generate video frames one at a time
• then play back at high frame rates
• can dramatically improve perceived s/w perf.

19
example
each frame about 36 sec
20
7. never learn anything about data

• debugging is more difficult if worried about
  producing correct results
• complex accurate interpolation techniques
• ad-hoc techniques produce prettier pictures
• programming bugs can produce stunning images

21
8. never compare results

• with other visualization techniques
• may detect bugs to be fixed
• other techniques may produce prettier pictures

22
example1
view dependent rendering of terrain data set
23
example2
flat map view of Mars
3D global view of Mars
24
9. avoid visualization systems

• provide mechanisms to add new modules
• users may violate rule 8 (never compare)
• usually not invented here
• (so we dont use them )

25
10. never cite references for data

• dont cite reference describing the data used
• someone may read the paper
• and discover the visualization bears no
  relationship to the original experiment
• will divert attention from the pictures appeal

26
11. claim generality

• but show results from a single data set
• difficult to write vis. algorithms to work
  properly on a variety of data
• much effort can be saved, if
• run on one data, then make the image look
  different, as if from other datasets
• and use rule 10 (never cite refs)

27
12. use viewing angle to hide imperfections

• many vis. algorithms produce 3D objects
  containing unpleasant imperfections
• avoid viewing angles exposing them
• if not, try another data set
• or

28
13. use specularity or shadows

• specular reflectionreflection from a smooth
  surface (mirror) maintaining the incident wave
• use shadows or brilliant highlights to hide the
  unpleasant 3D imperfections

29
14. this is easily extended to 3D

• 3D algorithms much more difficult than 2D
• the effort of generalizing a 2D alg. can detract
  from producing pretty pictures
• simply claim that the algorithm is easily
  extended to three ore more dimensions

30
conclusions

• details some techniques to divert attention away
  from data and towards beauty,
• (audiences love color graphics and animations)
• and to avoid tedious debugging of software
• to lie or not to lie?
• variations in the method ? influence the users
  perception and interpretation of data

31
- end -

• http//www.idi.ntnu.no/zoran