Thanks Bano,

Neat! This is pretty fun too. For everyone else, the idea of peekaboom
is that you play with a partner again, and your partner sees a label that  
has been assigned to a picture, and then your partner slowly reveals the  
image to you, such that you can guess what the label is. So, if the label  
is "Hair", then your partner presumably reveals to you the hair found in  
an image, so that you can guess hair. And you take turns, one being the  
revealer, and the other the guesser. I must admit my success rate with  
this game was horrible, and I am sure that my partner (anonymous thank  
goodness) was ready to kill me. I think part of the problem is that  
images are so multifaceted that it is hard to assign a single simple label  
to them, and then of course if you have a partner who is not serious and 
reveals irrelevant parts of the image, then that makes it tough too. But,  
maybe with a little practice... 

I had an interesting experience with Google images, btw, where it points
to a conflict between labeler incompetence with wanting to do well on the
game. One of the images shown to me and my partner was clearly a plot. So
I put labels of plot, line plot, graph, graphical plot...but nothing 
was matching with my partner. I knew what the problem was, probably my  
partner had called it a chart, which was wrong, but I went ahead and  
called it a chart and got my 100 points. :) So there is an interesting  
issue there about knowing not so much what the image is, but what your 
partner thinks it is. 

Enjoy,
Ted

On Sun, 10 Sep 2006, Satanjeev Banerjee wrote:

> There's a follow-up game from Luis's group that produces bounding boxes
> around objects inside an image. Additionally, it attaches meaning to the
> labels: Is this a noun? A verb? A piece of text in the image?
> 
> The paper: http://www.cs.cmu.edu/~biglou/Peekaboom.pdf
> The game: http://www.peekaboom.org
> 
> -Bano
> 
> On 9/10/06, ted pedersen <[EMAIL PROTECTED]> wrote:
> >
> >
> > You may recall some time ago I mentioned something called the ESP
> > Game. This is where two people who don't know each other are partnered
> > up over the Internet, and they assign labels to images together,
> > and when they agree, the label is used.
> >
> > http://groups.yahoo.com/group/nlpatumd/message/655
> >
> > This is a reasonably good way to collect data from volunteers in a
> > somewhat reliable way, and if you can make it fun, so much the better,
> > you'll get more data!
> >
> > Well, apparently this has caught Google's eye, and you can now play the
> > ESP Game with Google Images.
> >
> > http://images.google.com/imagelabeler/
> >
> > Give it a try, it's sort of fun. One heuristic I have noticed is that
> > anytime there is text in the image, if you use that as the label it will
> > match your partner. For example, I labeled one image of a group of people,
> >
> > one of whom had a NASA bag in their hand as NASA. This matched and became
> > our label. I don't know if this is the best label or not, but it's a
> > remarkably effective heuristic (if your goal is to simply match). Now, if
> > your goal is to do meaningful labels, you probably shouldn't just look for
> > text in the image and assign that!
> >
> > Enjoy,
> > Ted
> >
> > --
> > Ted Pedersen
> > http://www.d.umn.edu/~tpederse <http://www.d.umn.edu/%7Etpederse>
> >
> >  
> >
> 

-- 
--
Ted Pedersen
http://www.d.umn.edu/~tpederse



 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/nlpatumd/

<*> Your email settings:
    Individual Email | Traditional

<*> To change settings online go to:
    http://groups.yahoo.com/group/nlpatumd/join
    (Yahoo! ID required)

<*> To change settings via email:
    mailto:[EMAIL PROTECTED] 
    mailto:[EMAIL PROTECTED]

<*> To unsubscribe from this group, send an email to:
    [EMAIL PROTECTED]

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/
 


Reply via email to