Tumblelog by Soup.io
Newer posts are loading.
You are at the newest post.
Click here to check if anything new just came in.

September 07 2014

kallaballa
Way faster and precise approximation of a sobel filter
The video shows the approximation process of a sobel edge detection filter using a genetic algorithm. The first row shows the current search result and the seconds row shows the target filter with the picture on the left, the dft in the middle and the dft in binarized form on the right.

http://github.com/kallaballa/Mimikry

September 04 2014

kallaballa
I'm using a genetic algorithm to approximate a series of convolutions which try to mimic a given effect. In this example it tries to learn a 8x8 blur effect generated with ImageMagick (convert -blur 8x8).
The upper images in the video show the current approximation and the magnitude spectrogram of its discrete fourier transform (DFT).
The lower images show the target image and its DFT.
Only results of the best candidate of each generation is shown in the video.

https://github.com/kallaballa/Mimikry  

Reposted bypumucklcliffordnexelpollodiablo
kallaballa
I'm using a genetic algorithm to approximate a series of convolutions which try to mimic a given effect. In this example it tries to learn a sobel edge detection convolution generated with ImageMagick (convert -define convolve:scale='!' -morphology Convolve Sobel).
The upper images in the video show the current approximation and the magnitude spectrogram of its discrete fourier transform (DFT). The lower images show the target image and its DFT. Only results of the best candidate of each generation is shown in the video.

https://github.com/kallaballa/Mimikry  

Reposted bymetalab metalab

July 30 2014

kallaballa
kallaballa
3547 3608 500
What i did
Reposted byjuliwieaugustmynnia
kallaballa

July 13 2014

kallaballa
5141 6584 500
Lindabrunn Stadtflucht 7
Reposted fromlydschi lydschi viastadtflucht stadtflucht
kallaballa
2893 ea7e
Pyromantics Part I - Close-up
Reposted fromstadtflucht stadtflucht
kallaballa
6659 dbf1
Pyromantics Part I
Reposted fromstadtflucht stadtflucht

June 16 2014

kallaballa
6608 f719 500
  built an IR midi paddle @metalab. we are using the IR proximity sensors for guitar/blues harp effects and the buttons for looper control.  
Reposted bypumucklmazzosydnormetalab

June 12 2014

kallaballa

June 10 2014

kallaballa
0074 d065 500
The stepper driver board wirehead and me built @metalab
Reposted bymetalabsydnorAndi

April 23 2014

kallaballa
This scenario is the continuation of the previous "Deliver the ammo 1". It's sole purpose is to check if they actually seek the facilities and enemies because sometimes they are actually reacting to rather strange circumstances. Changing scenario dimensions is a good check for that.
Reposted bypumuckl pumuckl
kallaballa
Last week i changed the rules fundamentally. Now the space ships have finite fuel and ammo and they die when they are out of fuel. So i added "facilities" which they must use to recharge themselves.
To make them learn to recharge i gave them only a very small amount of fuel and no ammo in the beginning - so either they find the facility quickly or they die.
Additionally enemy facilities may be captured if the amount of enemy ships in range is greater by 3 than the amount of friendly ships in range but neither have i adjusted the fitness function nor did i train them yet to do so.
Reposted bypumuckl pumuckl

April 18 2014

kallaballa
Lately i spend a lot of my time working on optimizing game/learning performance. In this video is it only takes 10 generations to learn the difference between an enemy and a friend and the basic movement required to attend this scenario. Also the simulation runs quiet fast.
Reposted bycliffordcygenb0ckmetalab

April 09 2014

kallaballa
Another video of the neurocid simulation. this time the populations had a whole night of training.

April 05 2014

kallaballa

Lesson 2 of my neural network space battle simulation. In this lesson they should learn to hold their fire until they are in range.

http://github.com/kallaballa/Neurocid  

Reposted bypumucklclifford
kallaballa

Lesson 1 of the training for my neural network space battle simulation.
At this point the are completely untrained.

http://github.com/kallaballa/Neurocid  

Reposted bypumucklclifford

April 04 2014

kallaballa

A video of  a lightly trained Neurocid population. Neurocid is a space ship battle simulation controlled by neural networks, evolved through a genetic algorithm.

I've trained them for about 15 minutes with smaller scenarios. I'll post videos of the training soon.


http://github.com/kallaballa/Neurocid  
Reposted bypumucklm68kmetalabnexpumuckljascha

December 18 2012

kallaballa

Yes, the folks at @metalabvienna are serious about Blinkendingens.
--MacLemon | @kallaballa
Reposted fromalphabet alphabet
Older posts are this way If this message doesn't go away, click anywhere on the page to continue loading posts.
Could not load more posts
Maybe Soup is currently being updated? I'll try again automatically in a few seconds...
Just a second, loading more posts...
You've reached the end.