A Program Learns Oriental Ink Painting

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."

Comments

Re:It was rewarded...

[Anonymous Coward]

...what was the reward? Human flesh?

Nah.
It got a byte to eat.

It only did the strokes, not the art

[Anonymous Coward]

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.

Re:It only did the strokes, not the art

[Cow Jones]

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 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 bird, which was photoshopped by DeviantArt user mohamedraoof to look like a "Natural Angry Bird". All three images, the original photo, the deviation, and the sumi-e version look very nice in their own way.