Starting up this thread because i thought it was fun.
- original web version:
- new web version:
- qt version (multiplatform, multithreaded):
- qt version compiled for Windows:
- C version for GNU/linux:
- Original Windows version:
- In progress opencl versions:
- forked mona with preliminary gif support
- New (to us) builds with various code tweaks, for testing and comparison
- share pics here:
- converting the xml output to svg:
- something similar for audio
its not mine, it was done by a guy named robert alsing, basically you put in an image and it creates random vectors till it matches the image
currently working on this
I guess there are more efficient ways of this concept. Like just having n points with a color and varying the color and position of these points. I wanted to do this once, but then I was too lazy to figure out the interpolation of these points.
A few weeks back, I wrote a python script that would use least squares to "solve" the color dimension and a genetic algorithm to evolve the spatial dimension.
It was buggy as shit, only worked on linux, and required a prodigious amount of ram to form the matrices, but it got better results faster.
That wasn't what I was doing. I was converting the image into a 3-column vector of the form:
I = [R,G,B]
Then, I would make masks of the polygons, m, and form a matrix of the form:
P = [m1, m2, m3,...]
Then, to get the colors, I would pseudo-invert the P matrix, multiply the I vector, and get the least-squares optimal colors. The problem was that the P matrix is #polygons x #numberpixels in size, and if the image is too large, it'd run out of RAM.
This skips any evolution of color and just gets it exactly right on the first pass though, which usually meant the first guess would be good enough for me and I wouldn't need to continue optimization.
final image, i got up to 98.25% and decided to do something new
>believing sick fucks dont fuck deformed women
its like you havent seen all those images of some guy breeding some downsyndrom woman and producing 5 more with the same deformed face
It's been ages since I looked at this but I did have a copy of the fitness function on CUDA. The cost of copying was too high to make it worth it, though: It probably makes more sense to make a more informed mutation algo and then just parallelize the (mutation+fitness) step
Oh. My image was from a website that knew the basic of a face, nose, eyes, mouth, ect, and would generate a bunch of polygons to try and create a face. It starts with a blank space. No original image.
Also, the qt version give much better results
I ported the C version to cuda last year when this was a thing on /g/, wonder if that code is still floating around. iirc the gpu wasn't that much faster. a multicpu implementation would probably be better since you don't have to deal with the slow pcie bus
Well the Tux3 version, while multithreaded, really only puts one core's worth of load on the entire system.
Also the updated version began to lag like fuck for no reason after 4 million gens, but that may have been the high poly limit.
I never finished mine, I can't even find the code anymore...
I'll start it again tomorrow.
Whould this be a good proj for the resume?
It looks as thought I missed the windows version in the OP.
So here's OSX from last year till I get this year's working:
Ι'd јust like tο evolp fοr a mοment. What yοu're referring tο as GNU plus Linux, is in fact, GΝU/Linux, οr as Ι've recently taken tο calling it, Linux.
Here you go, v1.3 (latest) for OSX.
Built with gcc 4.9 on 10.7 in case anyone was wondering.
If anyone can give me a better image that I can use as the logo it would be awesome.
runtime: ~10 minutes
but the size of the original image is only 30KB
I'm sure you don't complain about the Macbook / keyboard / Linux vs Windows / Firefox vs chrome threads
Evolve has really improved.
I made them the last time this threads where around here.
Okay never mind. It just pauses every 10k generations or so.
Humans literally breed for freaks of nature. We want to have genetically superior and aesthetically beautiful looking people, which freaks of nature in a world filled with ugly people.
What is lusus naturae?
It's from last year. I'll just be uploading some I downloaded.
This is the current one made in like 5 minutes.
I guess frame by frame processing through command line and a bit of bash with
FFMPEG & Mona would make that happen - albeit very slowly unless it's optimised
or faked somehow for realtime feedback.
Try adding yourself to a picture of your waifu and see how it turns out. Could be interesting.