1 / 40

3D vs. 2D Debate in Object Recognition Theories

Recap of human visual system stages, depth cues, and object recognition theories. Explore the debate between 3D and 2D representations in visual perception. Learn about Tufte's insights on data visualization techniques.

amastin
Download Presentation

3D vs. 2D Debate in Object Recognition Theories

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Lecture 4 • Human Visual System • Recap • 3D vs 2D Debate • Object Recognition Theories • Tufte – Envisioning Information

  2. Human Visual System – Recap • Sensory Representations Effectivebecause well matched to early stages of neural processing • Physical World Structured • Stages of Visual Processing 1 Rapid Parallel Processing • Slow Serial Goal-Directed Processing • Visual System Detects CHANGES + PATTERNS • LuminanceChannel More Important than Color • Pre-Attentive Features • Position • Color • Simple Shape = orientation, size • Motion • Depth

  3. Proximity Similarity Continuity Symmetry Closure Relative Size Figure and Ground Gestalt Laws – Recap

  4. Space Perception – Recap • Depth Cues • Shape-from-Shading • Shape-from-Contour • Shape-from-Texture • Shape-from-Motion

  5. DiffuseLambertian Specular AmbientShadows Simple Lighting Model – Recap Light from above and at infinity Diffuse, Specular and Ambient Reflection Depth Cues

  6. Motion parallax 0.001 Occlusion 0.01 Relative size Depth Contrast 0.1 Binocular disparity Convergence accommodation 1.0 Aerial 1 10 100 Depth (meters) Depth Cues – Relative Importance – Recap

  7. 3D vs 2D Debate - Display Abstract Data in 3D? • Depth Cue Theory • Depth cues are environmental information about space • Occlusion most important Depth Cue • Perspective may not add anything by itself • Stereo important for Close Interaction • Motion important for 3D layout • Surface Perception • Shape-from-Shading • Shape-from-Texture

  8. Relative Position Judgment • Fine Judgments - threading a needle • Stereo is important • Shadows • Occlusion • Large Scale Judgments • Perspective • Motion parallax • Stereo is not important

  9. Image + Object Recognition • Properties of Image Recognition • Remarkable image recognition memory • Up to 5 images per second • Applications in image searching interfaces • Easier to Recognize than to Recall • Image Based Theories • Template theories based on 2D image processing • Structural 3D Theories • Extract structure of a scene in terms of 3D primitives

  10. Template Theories Template with simple morphing operations

  11. Template Theories – Scale Matters Visual degrees = 4optimal for object perception

  12. Geon Theory

  13. Geon Theory (cont.) 3D Primitives “Geons” Structural skeleton Shape from shading is also primitive

  14. Canonical Silhouettes

  15. Recognition – Processing Stages

  16. 11.4% errors 21% errors 20% memory errors 34% memory errors Pattern Finding & Recognition – 3D vs 2D

  17. Edward Tufte • Books • The Visual Display of Quantitative Information • Envisioning Information • Visual Explanations

  18. Tufte - Minard's Napoleon's March to Moscow

  19. Enforce Visual Comparisons Width of tan and black lines gives you an immediate comparison of the size of Napoleon's army at different times during march. Show Causality Map shows temperature records and some geographic locations that shows that weather and terrain defeated Napoleon as much as his opponents. Show Multivariate data Napoleon's March shows six: army size, location (in 2 dimensions), direction, time, and temperature. Use Direct LabelingIntegrate words, numbers & images Don't make user work to learn your "system.” Legends or keys usually force the reader to learn a system instead of studying the information they need. Design Content-Driven Tufte - Escape Flatland: Napoleon's March

  20. Tufte – Challenger Data: Launch? Graph obscures important variables of interest: temperature is shown textually and graphically; degree of damage is not mapped onto a nominal scale

  21. Tufte – Challenger Data: Launch? • Diagrams can lead to great insight, but also to lack of it

  22. Cause of cholera epidemic in London in 1854? John Snow’s deduction that a cholera epidemic was caused by a bad water pump Modified in Visual Explanations by Edward Tufte, Graphics Press, 1997

  23. Maximize data-ink ratio Data ink Data ink ratio = Total ink used in graphic Maximize data density Number entries in data matrix Data density of graphic = Area of data graphic Measuring Misrepresentation  close to 1 Size of effect shown in graphic Size of effect in data Lie factor = Tufte’s Measures

  24. Show Data Focus on Content instead of graphic production Avoid Distorting what Data has to say Make Large Data Sets Coherent Encourage Eye to Compare Different Pieces of Data Reveal Data at several Levels of Detail Closely integrate Statistical and Verbal Descriptions Tufte - Graphical Displays Should

  25. Example Stock market crash? 500 475 450 1998 1999 2000 2001 2002

  26. Example 500 250 Show entire scale 0 1998 1999 2000 2001 2002

  27. Example 500 250 Show in context 0 1960 1970 1980 1990 2000

  28. Tufte - How to Exaggerate with Graphs “Lie factor” = 2.8

  29. Tufte - How to Exaggerate with Graphs “Lie factor” = 2.8 Error: Shrinking along both dimensions

  30. When to use which type? • Line Graph • x-axis requires quantitative variable • Variables have contiguous values • familiar/conventional ordering among ordinals • Bar Graph • comparison of relative point values • Scatter Plot • convey overall impression of relationship between two variables • Pie Chart • Emphasizing differences in proportion among a few numbers

  31. Tufte - Graph & Chart Tips • Avoid Separate Legends and Keys • Make Grids, labeling, etc., Very Faint so that they recede into background • Graphical Integrity • Where’s baseline? • What’s scale? • What’s context? • Watch Size Coding: Height/width vs. area vs. volume • Using Color Effectively • To label • To measure • To represent or imitate reality • To enliven or decorate

  32. Tufte – Hierarchy of Visual Effects

  33. Tufte – Hierarchy of Visual Effects

  34. Tufte – Hierarchy of Visual Effects in Maps

  35. Tufte – Be aware of visual artifacts

  36. Tufte – Leverage Illusionary Contours

  37. Tufte – Narratives of Space & Time

  38. Axonometric Projection Tufte – Micro / Macro Readings - 2½ Displays To Clarify, Add Detail

  39. Tufte – Micro / Macro Readings - 2½ Displays

  40. Tufte’s Principles – Summary • Good Information Design = Clear Thinking Made Visible • Greatestnumber ofIdeasin Shortest Timewith Least Inkin theSmallest Space • Principles • Enforce Visual Comparisons Show Comparisons Adjacent in Space • Show Causality • Show Multivariate Data • Use Direct Labeling • Use Small Multiples • Avoid “Chart Junk”: Not needed extras to be cute

More Related