<<In a message dated 3/31/2009 9:58:16 A.M.
Pacific Daylight Time,
(In image search, Google and all other search engines still suck.
Here's to tagging coming to Commons.)>>
Somewhere in my hazy memory I remember two projects I'd read about.
There was a guy or a bunch of guys maybe who were trying to teach a computer
how to recognize "large expanses of flesh" so it could find nude pictures I
It sort of worked under some conditions but that is only one type of image.
Then there was a human-tagging project where people ranked up points as they
tagged images with whatever. I'm not clear how it worked, and I haven't
heard of it for some years.
There are some tasks that humans will always be better at than machines.
That's probably why Google's image search sucks. But if I search for
Pitt Nude" I still get several decent hits anyway.
Problem is neither humans nor machines are very good at image tagging.
Machines are obvious but lets look at humans. First problem is that
your average human isn't an expert on everything. Unless you are
interested in the early history of submarines you are going to have a
hard time working out what this photo is of:
and unless you
are a real submarine nerd you will likely only know that because it is
the pic everyone uses.
Now you would hope that the uploaded would tag it. Problem there is
that the unloader is only thinking about it in one context when it
might usefully be tagged with additional things to over other