Harmut Neven's Presentation at Emerging Communications Conference & Awards 2010 America

1,196 views

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,196
On SlideShare
0
From Embeds
0
Number of Embeds
50
Actions
Shares
0
Downloads
21
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Harmut Neven's Presentation at Emerging Communications Conference & Awards 2010 America

  1. 1. Searches originating inside and outside of your head Hartwig Adam, Ulrich Buddemeier, David Petrou, Ted Power, Andrew Harp, Mihai Badoiu, Avi Flamholz, Andrew Harp, Anthony Sciola, Gabriel Taubman, Matt Bridges, Matt Casey, Sharvil Nanavati, Max Braun Johannes Steffens, Jiayong Zhang, Lijia Jin, Alessandro Bissacco, Fernando Brucher, John Flynn, Anand Pillai, Yuan Li, Andrew Rabinovich, Henry Rowley, Rafael Spring, Andrew Hogue, Shailesh Nalawadi, Hartmut Neven
  2. 2. Sananga, eye drop entheogen used by the Yawanawa tribe
  3. 3. Enables hunters to better see monkeys
  4. 4. Enables hunters to better see monkeys
  5. 5. Definition Augmented Reality View of a physical real-world environment augmented by computer- generated imagery R. Azuma 1997, P. Milgram and A. F. Kishino 1994 Generalization Presentation of information as a function of the environmental context Not: Visual Search Using an image as a search query Not: Wearing a head mounted display
  6. 6. Will AR cause a paradigm shift comparable to the GUI? Some deconstruction … Pixels are from men, photons are from god.
  7. 7. If this is our future then maybe AR is not all that important
  8. 8. Augmented Reality requires a physical reality one is interested in Kaiser Family Foundation: Today, 8-18 year-olds devote an average of 7 hours and 38 minutes (7:38) to using entertainment media across a typical day (more than 53 hours a week). And because they spend so much of that time 'media multitasking' (using more than one medium at a time), they actually manage to pack a total of 10 hours and 45 minutes (10:45) worth of media content into those 7½ hours.
  9. 9. Augmented Reality requires a physical reality one is interested in • Prerequisite is a novel object in my environment or the availability of novel information about an object nearby. • Hence AR may not meet the bar of daily engagement. • Same holds for visual search.
  10. 10. At least a 10:1 ratio between externally and internally triggered searches • Less opportunity for externally triggered searches • But if performed it feels magical and is often of high utility • Most valuable when there is no voice or text substitute: o Faces, do a background check on new nanny Disclaimer: Face recognition will only be offered once accepatable privacy models have been established o Restaurant in Tokyo for person who does not speak Japanese • High convenience o Barcode o Text for translation
  11. 11. Sources of Augmentation From Memory • Conventional • Examples: • Navigation information • User generated updates and themes: Layar, Tonchidot
  12. 12. Sources of Augmentation From Memory • More examples • AR ads • AR in complex visualizations
  13. 13. Sources of Augmentation From Sensors • Unusual • Multispectral camera o infrared: notice a person lying, find your pet o ultraviolet: determine quality of food • Far out: Lockheed's recent patent on quantum radar
  14. 14. Google Goggles Design principles behind Goggles 1. Universal 1. Makes it more difficult compared to vertical solutions 2. Needs very low false positive rate 2. To the degree possible do not force the user to select modes 3. Specificity 1. Object instance more important than object class recognition 4. Put best foot forward 5. Recall, precision, scale and speed
  15. 15. Recognition disciplines that work and do not work
  16. 16. Summary of the state of the art (for product managers:) Product Faces Unpackaged Rigid Packaging Products Articulate Textured Logos Contour Defined Landmarks Cars Pets Easy Hard
  17. 17. Accuracy of place recognition 130K Place Models automatically mined from Photo Collections
  18. 18. TOTO The Other Textured Objects 100 10-1 > 50M images in database 10-2 10-3
  19. 19. Current accuracy in face recognition
  20. 20. Word level recall/precision of OCR Dense Text Sparse Text
  21. 21. Grand challenge vision problems 1. Massive Class Recognition 2. Recognition of weakly textured, contour defined objects
  22. 22. Demo: Recognition triggered virtual objects
  23. 23. Roadmap • Combine external and internal searches within one framework • Audio-visual search • Audio feedback to guide the user • Make the search engine your personal companion • Offer AR when it is the superior UI paradigm
  24. 24. Thank you! Hartwig Adam, Ulrich Buddemeier, David Petrou, Ted Power, Andrew Harp, Mihai Badoiu, Avi Flamholz, Andrew Harp, Anthony Sciola, Gabriel Taubman, Matt Bridges, Matt Casey, Sharvil Nanavati, Max Braun Johannes Steffens, Jiayong Zhang, Lijia Jin, Alessandro Bissacco, Fernando Brucher, John Flynn, Anand Pillai, Yuan Li, Andrew Rabinovich, Henry Rowley, Rafael Spring, Andrew Hogue, Shailesh Nalawadi, Hartmut Neven

×