NSF Aladdin Center

1
A Novel Approach to Labeling Images
Luis von Ahn
NSF Aladdin Center School of Computer
Science Carnegie Mellon University
2
Labeling Images With Words
3
Labeling Images With Words
4
Labeling Images With Words
Martha Stewart Flowers Super Evil
5
Labeling Images With Words
Martha Stewart Flowers Super Evil
Completely open problem
6
Image Search on The Web
7
Image Search on The Web
Uses filenames and surrounding text
Doesnt look at the actual image
8
Desiderata
A method for labeling images that

• Actually looks at the images
• For any image gives several keywords that make
  sense
• Is very fast (Google has 425,000,000 images)

9
Stealing Cycles From Humans
10
Stealing Cycles From Humans
Over 50 million people in the United States alone
play computer games on a regular basis!
(Interactive Digital Software Association)
11
The ESP Game will allow us to label all images on
the web in 30 days!
12
The ESP Game
Two-player online game
Partners dont know each other and cant
communicate
13
The ESP Game
Two-player online game
Partners dont know each other and cant
communicate
14
The ESP Game
The ES
Player 1
Player 2
15
The ESP Game
The ES
Player 1
Player 2
Guessing car
Guessing boy
16
The ESP Game
The ES
Player 1
Player 2
Guessing car
Guessing boy
Guessing hat
17
The ESP Game
The ES
Player 1
Player 2
Guessing car
Guessing boy
Guessing hat
Guessing kid
18
The ESP Game
The ES
Player 1
Player 2
Guessing car
Guessing boy
Guessing car
Guessing hat
Success! You both agree on car
Guessing kid
Success! You both agree on car
19
(No Transcript)
20
(No Transcript)
21
The ESP Game
Average labeling rate slightly over 4 images
per minute
5000 people playing the game throughout the day
would label all the images on Google in 30 days!
Individual games in Yahoo!, Pogo.com or MSN
average over 10,000 players throughout the day
22
Stealing Cycles From Humans is a More General
Idea
23
(No Transcript)