moyogo's little blog

Blogging about Open Source, fonts, language technology, maps and random stuff happening wherever I am.


Label it all!

I have no idea how long this have been around but Google has an image labeler. It's interesting because it allows you to just label pictures for Google. They just give you a picture and you're supposed to give as many labels as possible. Since you're teamed up with someone else, the first label that match theirs is recorded and another pictures is given for you two to label. The fun part is that there's a score board. You have 1 minute to match as many pictures with labels.

Google definitely found a way to use smarter pigeons or monkeys ;-) and the results surely are better than AI... at least for now.

Unfortunately, I don't think this data is available to the public. What are people waiting to start an open project with similar goals?


Searching pictures

I came across this interesting little website the other day: Retrievr. It's an experimental webservice that allows you to search for patterns on Flickr, by drawing a rough sketch of what you're looking for. The search is based on an algorithm that will only match larger shapes so don't expect to get exactly what you draw, unless you draw a really simple shape (like a vertical line or something simpler ;-) But the poor results are probably due to the size of the database.

Xcavator offers another way to search through pictures. It's rather complex but the results are much better than Retrievr. A video explains how to use the tool.

Unfortunately both are still at a very experimental level and only have samples of pictures from Flickr.