tagging gone mad

ROTFL.. well I just let tagcow at my flickr archive. It's an interesting concept of semantically tagging images. Presumably they have some object/facial recognition software behind this - although they don't specify what/how.

I was allowed to upload a couple of photos and identify the people in them (I was in a hurry to see the result, so I only did two: syber and frobozz).

What I ended up with is hilarious, sometimes obvious and more than occasionally wrong.. my brother-in-law is tagged as frobozz multiple times.

Who wants to be tagged as a lady?? "lady syber" (which includes one of my german housemate - does she look like me?)

Especially this classic, me holding some pig crackling and checking out my freely lymphing scraped knee.

A brilliant summary here; two of my friends drinking becomes "two men bar".

This one is by far my favorite: frobozz faster easier together accerating hpc man sitting front servers floor mouth wire keyboard lap. Not only did it pick up *all* the text in the image (a tiny OCR error on accelerating), and identify the objects correctly, but it noticed the "mouth wire"!!! I didn't even see that!

So, the verdict on tagcow. Good idea. "A" for effort. Fail for tagging every freaking photo with useless information like "guy", "woman", "sitting".

And what the hell is up with the typing errors? Is this really an automated system, or is it a bunch of low-wage outsourced workers typing like mad?