[cc-community] Semantic API for Creative Commons Images
Paul Houle
paul at ontology2.com
Fri Aug 20 15:24:40 EDT 2010
On 8/20/2010 12:08 PM, Nathan Yergler wrote:
> Very cool, Paul! Are you just indexing Wikimedia Commons right now,
> or are you discovering CC licensed images in the web at large?
> Regardless, it's a great resource and looks like an interesting linked
> data application.
The current image acquisition pipeline is based on Wikimedia
Commons because I think it's a good way to get a "seed" data set that
will help us solve some problems we had with
http://ny-pictures.com/
which used a different kind of pipeline. In particular, the
biological neural networks we were using were easily fooled by the
American flag in
http://ny-pictures.com/nyc/photo/picture/39985/looking_intimidating_flag_columns_stock_exchange
to classify the target as the "American Stock Exchange". If we
can get high-accuracy reference images of targets, we can feed these
into the other pipeline and have fewer errors of that sort... On the
other hand, the Wikimedia-based pipeline makes different mistakes but
I think those can be cleaned up more easily. The immediate plan is to
capture as much as we can from Wikimedia, do a round of cleanup, and
then use an improved version of the old pipeline to grow it out to at
least 100 times the size it is now.
So far as other image sources go, I'm not so interested in
crawling the web at large as I am in getting content from large
collections where there's some standardization of the metadata &
providence information. Certainly there are sites like Flickr and
Picassa, but I'm definitely interested in the more formal kinds of
"digital libraries" that are weak in the usability and discoverability
departments.
If anybody's got a collection of pictures that they'd like to see
indexed in Ookaboo, I'd be glad to talk about it. I'm not against
providing manual interfaces for uploading content, but I'm more
interested in working on a wholesale rather than retail basis for the
time being. When I've actually got an audience, then I might think
about adding the more conventional "community" features -- until then
I've got a lot of other things to keep me busy.
More information about the cc-community
mailing list