Sweetcode

January 5, 2003 12:34 PM

Sweetcode has some sweet code: learning language by watching movies; describing images for searching by doodles; automated haiku searching; etc. New entries are sometimes far apart so use the mailing list.
disclaimer: they listed something of mine a couple of months ago (not one of their best finds) and it has been mentioned here once as an attribution, so i considered posting just the sub-links. but why give you a fish?
posted by andrew cooke (7 comments total)

The evilfinder is simply brilliant.
posted by bshort at 1:30 PM on January 5, 2003

So many people I'd always suspected of evil, and it's all true.
posted by devon at 2:03 PM on January 5, 2003

i check sweetcode periodically, and i think that this is far and away my favorite project, and this one is my favorite example from that project.
(if you want to get super geeked out, check out " Paketto Keiretsu" about a third of the way down the sweetcode page)
posted by dolface at 2:36 PM on January 5, 2003

Ah, Image Analogies. One of the better works to come out of SIGGRAPH 2001. I need to go back and try out the tools again, now that systems are ~3x as fast...they're all open source, conveniently.

I suspect interesting things are possible with image compression -- forget 8x8 blocks, select a chunk and clone the rest.

Dolface -- glad ya like Paketto :-)

--Dan
posted by effugas at 5:20 PM on January 5, 2003

sweetcode is awesome.. i thought about posting it here before, but figured it wouldn't go over to well..

kudos on the size of your gonads, buddy
posted by shadow45 at 9:45 PM on January 5, 2003

effugas - isn't that what fractal compression is all about (i'm guessing, as i don't know what the algorithm is behind these things - maybe it's more empirical)?
that image analogy example is amazing.
posted by andrew cooke at 2:44 AM on January 6, 2003

Fractal compression seemed to be all about trial and error, if I remember right. Try to find parameters that look like existing output; when you zoom it, it should continue to look like it even smaller.

Yeah, didn't work too great.

IA tries to match shapes and borders in a way that matches the human mind's tendancy to split the visual field into textures and edges. Does a hell of a job, too :-) The link to compression hasn't, I think, been explored too deeply -- certainly, one can imagine the source image as a "dictionary" against which the destination is forged. Reconstruction from a dictionary, even in a lossy manner, is at the heart of many compression algorithms.

Heh -- just something I'm playing with :-)

--Dan
posted by effugas at 5:12 AM on January 6, 2003