Game Development Reference
In-Depth Information
Fig. 3.2 Screenshot of the google image labeler implementation of the ESP game
mechanics. The task for the peek is to guess the correct word, which is known to the
boom , who is also given an image on which the object described by that word is. The
peek does not see the image, but during game, boom may reveal him small circular
areas of the image (as seen on screenshot in the Fig. 3.3 )togivethe peek hints on
what word to guess. The less circles they players use to correctly guess the word, the
more point they receive. Nevertheless, to be successful, the boom has to effectively
reveal areas that contain part of the object the peek has to guess. According to Ahn,
using the same image and tag in more games with different players and aggregating
the circular areas, the obtained contours were of high precision.
Generally, we can say that the image description acquisition SAGs are capable
of delivering quality descriptions and retain the quantity potential as they can be
played at the same time by many players. Also the image description SAGs are in
general easily transformable for video streams [ 17 , 24 ]. The Ahn's ESP Game, has
eventually become the best known SAG with hundreds of players every day, with
claim of more than million annotated images in 3 months, which is interesting even
in web scale [ 24 ], especially when the automated image categorization approaches
do not work effectively.
3.2 Audio Track Semantics Acquisition Games
Another task, where semantics acquisition games are utilized, is metadata acqui-
sition for audio resources. Similarly to graphics, sounds and music are resources
represented on a sub-symbolic level which is hard to be automatically interpreted
 
Search WWH ::




Custom Search