AutoCollage: this paper makes me feel very dumb.
Tags: , ,

26 Responses:

  1. fo0bar says:

    I mentally read "Microsoft Research Cambridge, UK" as "Exterminate All Rational Thought"

  2. Don't. Remember that basically every equation in that paper probably represents one or two solid days of work on the part of one of the authors. And also that the paper isn't intended to be read by anyone who isn't already basically familiar with the crunchy bits of most of the contents, which they acquired knowledge of over the course of years.

  3. Where is the download?

  4. strspn says:

    In future work it is planned to elaborate the object sensitivity.

    Translation: Not ready for prime time.

    Since sky detection was admittedly 83%, faces and bodies probably got cut at least half the time, which makes me think the samples used for the glowing-results survey were probablly vetted instead of actually "random."

    If it was worth anything they would be describing actual alcorithms instead of integral calculus equations. You are supposed to feel stupid so you don't ask too many questions.

  5. evan says:

    It's a pretty awesome result technically. But I also find the resulting collages are unpleasant to look at.

  6. jaydubbee says:

    They didn't even cite you. Wimps.

  7. What has Microsoft Research Cambridge given the general public? Nothing.

    What has Jamie Zawinski given me? Good code, good laughs, and a bar I have gotten very drunk at.

    • otak says:

      What has Microsoft Research Cambridge given the general public?

      They employ a load of the core Haskell developers. Yeah I know -- not really something for the general public, but it certainly bothers me. My only love sprung from my only hate!

  8. runehog says:

    It would be kind of fun to work on a system that does the exact opposite -- collages images together in the worst possible way (ie. uses only bits of sky or pelvis.)

    I've had the same reaction to some of the google papers -- the GFS one is pretty sane, but the others are, um...

  9. buz says:

    Fuck math, let's dance!

  10. fantasygoat says:

    Yes, but who gets laid writing boring papers?

  11. remaker says:

    Let's see:

    Webcollage is shipping and has been for a long time. It is working code, not a pile of ideas, formulae, and experimental code. It takes virtually aribtrary input and produces interesting collages.

    Autocollage is a superengineered system which requires carefully hand-massaged input and I don't see a usable download or source. Just a few filtered results and a paper with minimal well-presented useful data to a collage application implementor.

    Some clever person once said: "I don't use any Microsoft products and neither should you."

    But maybe there are some useful ideas in there that can help you to exterminate all rational thought. Isn't that what Microsoft speciali[sz]es in?

    • rapier1 says:

      I think the thing to keep in mind, and what Jamie probably understands, is that while his code produces something functional the paper describes an elegant system where the resulting images are just a by product of the research. The point of pure research isn't to come up with a worthwhile product but to develop the theoretical underpinnings which can be used to advance science along a wide range. The implications of this paper actually extended well outside of the narrow focus of building a better collage. Microsoft Research, no matter what you think of the company as a whole, does damn good work and they aren't dummies by any means.

  12. rapier1 says:

    How much shit happens really depends on your context. In the community I work in (high performance networking and computation) this sort of thing happens all the goddamn time to everyone. If you take it personally then you're fucked.

  13. edouardp says:

    Well that's taking it to one extreme (maths-laden research papers), and I just saw someone taking it to the other extreme (arty wank-speak):

    The description is full of "exploring relationships" "recontextualization" etc. And, of course, referring to the PHP hack as "the peice".

    It should make you feel very smart.

    • jwz says:

      I should filter all of the xscreensaver descriptions through the arty wankspeak translator. Then I'd get to drink on NEA's dime.

      • joel says:

        you joke, but xscreensaver has already been used in performance art

      • edouardp says:

        Jamie Zawinski - Drunk On The NEAs Dime (2006): The installation features the artist lying in a pool of his own urine, surrounded by Jack Daniels bottles. Vomit, presumably the artists own, also features heavily in this mixed-media work.

        This piece challanges traditional societal views of art by recontectualizing the artists relationship with his audience in both the creative and consumptive spaces. It also explores the normally hidden modes of communication between the artist and funding bodies, which are themsleves the product of further, often unrealised, relationships involving the audience themselves, thus completing a cycle of in/action.

        • edouardp says:

          OK - so I'm not very good at that :-)

          I believe a major source of income for English Lit post graduate students is writing the arty wank-speak you find next the pieces in art galleries...