Forgot your password?
typodupeerror
AI Graphics Japan News

A Program Learns Oriental Ink Painting 53

Posted by Unknown Lamer
from the midi-killed-art dept.
mikejuk writes about a neat use of machine learning. From the article: "Using reinforcement learning to make a computer paint like an oriental Sumi-e artist isn't just a matter of shouting 'well done' — and yet, when you look at the results, that's what you want to do. ... Three researchers at the Tokyo Institute of Technology have attempted to teach a computer how to do it [paint] using standard reinforcement learning. When the program used the brush to create a smooth stroke, it was rewarded. After it had learned to use the brush, it was set to rendering some photos and the results look very good."
This discussion has been archived. No new comments can be posted.

A Program Learns Oriental Ink Painting

Comments Filter:
  • but we just haven't made it fast enough, shrunk it into a form factor that can fit in our pockets, or integrated it into a multifunctional device that can talk, walk, feed us, and entertain us in kinky ways. (Apologies to William Gibson, who famously said, "The future is already here -- it's just not very evenly distributed.")
  • by Ryanrule (1657199) on Friday June 29, 2012 @10:39PM (#40501537)

    ...what was the reward? Human flesh?

    • by Anonymous Coward on Friday June 29, 2012 @10:49PM (#40501569)

      ...what was the reward? Human flesh?

      Nah.
      It got a byte to eat.

    • by guttentag (313541)
      I know, my first thought was, "what would a computer do with a lifetime supply of chocolate [imdb.com]?" So I RTFM.

      My understanding is that they basically told it your goal is to earn yourself the highest score possible. You get 0 points for painting off the edge of the page, lots of points for smooth brush strokes, etc. I wouldn't really call it a reward system, any more than WOW's grinding is a reward system.

      Obligatory: "I am now telling the computer exactly what it could do with a lifetime supply of chocolate!
      • The computer is rewarded with less uncertainty as the trials approach perfection. Surely, for deterministic calculating machines, there is no finer goal.
      • by drinkypoo (153816)

        I wouldn't really call it a reward system, any more than WOW's grinding is a reward system.

        When you do something that you find rewarding there is a chemical change in the brain which can be represented with a number. The computer doesn't actually have a brain, it's just a collection of state machines (real and virtual) so you just increase a number and it doesn't feel any way about it. Does that make it not a reward when you're talking about a scoring system? Splitting hairs, I say, if the result is the same; you "reward" the desired behavior with more points, and that leasts to more of the desir

    • by Anonymous Coward

      They gave it a partial copyright claim and a 1% take of all commercial net profit in region 2, paid once every 20 years.

  • by Anonymous Coward on Friday June 29, 2012 @10:57PM (#40501599)

    Once the brush agent was trained it was used to create ink paintings of photos. The contours that the brush follows were generated manually, so the artistic effect isn't quite as autonomous as it might appear.

    Basically, it only learned the basic movements. A person manually told it where to apply them.

    • by kermidge (2221646)

      Still, not too shabby. Check out the photo conversions on page 7.

    • by Cow Jones (615566) on Saturday June 30, 2012 @07:15AM (#40503547)

      Right. It would be interesting if we could use something like this to train Photoshop filters to get closer to the result we want...

      On a side note, one of the example photo conversions on page 7 of the PDF (or here [i-programmer.info] from the third link) has the i-programmer writer commenting "I can't help but think that the bird looks a lot like something from Angry Birds...". That's not an accident: the original source image is this photo of a red cardinal [betterphoto.com] bird, which was photoshopped by DeviantArt user mohamedraoof to look like a "Natural Angry Bird [deviantart.com]". All three images, the original photo, the deviation, and the sumi-e version look very nice in their own way.

  • Why can't they just call it what it is?
  • by TheLink (130905) on Friday June 29, 2012 @11:24PM (#40501711) Journal

    The bird pic is not from a real bird.

    Some guy did actually draw some birds based on the angry birds:
    http://funnyzela.com/real-life-angry-birds/ [funnyzela.com]

    So that oriental ink bird is a more abstract drawing of a pseudo-realistic version of an imaginary bird... Something like that - did I miss a step? :).

    • by retchdog (1319261)

      and the drawing contours were manually specified. the robot just learned how to make particular sumi-e strokes.

      is all japanese ai/robotics like this? here's a hint, you guys: most of the time, the motor control is the easy part; planning and feature detection are what is interesting. seriously, what's up? the japanese certainly aren't stupid, i guess they just have an obsession with building toys.

  • You just install FrontPage.
  • by Anonymous Coward

    We call it ASIAN these days, you racists.

    • by Ignacio (1465)

      The *people* are Asian. The *products* and *styles* are oriental.

      • by Zanadou (1043400)

        The *people* are Asian. [Therefore] the *products* and *styles* are Japanese.

        FTFY - from the point-of-view of most Western people.

  • Now turn it loose with a few Grand Strategy games, Real-time Strategy games, and Counter Strike for good measure. And don't forget a copy of The Art of War. This should turn out well.
  • What does a computer consider a positive reward? And, if you tell it what it expects as a positive reward, how does it realize it?
  • by khipu (2511498) on Saturday June 30, 2012 @02:29AM (#40502585)

    If you read the paper, it attempted to reproduce the pressure profile during the strokes.

    See those nice renderings of photographs as brush strokes? The path of the strokes was generated by hand, only the learned pressure profile was used. And the pressure profile it actually used seems pretty poor.

    The paper is pretty much faking its results (although it's at least honest about doing so).

  • by kelemvor4 (1980226) on Saturday June 30, 2012 @10:10AM (#40504317)
    Adobe Photoshop has been able to do this for years. At least since CS3 (the first version I used). Might be of some value as an exercise in a programming class, I guess. The finished product, however, is just a copy of other software that already exists.
  • Omori Sogen (first ``o'' in each name should have macrons over them) and Terayama Katsujo (last ``o'' should have a macrons over it) in _Zen and the Art of Calligraphy: The essence of sho_ examined brush stokes under high magnification to show how the strokes were infused w/ ki (the spirit of the artist) --- compare Plate 2 `` the ink particles are lacklustre and weak'' w/ Plate 5 ``a dramatic transformation has taken place --- the bokki (infusion of ki into the ink) is sold, dynamic, alive.''

    What do these

  • Oh great, our art skills are being outsourced to computers. :P

We are experiencing system trouble -- do not adjust your terminal.

Working...