Imagine a game which looks like this. Shaking the view around disrupts the image and you need to keep it still if you want more detail. And if something passes by quickly, you wouldn't be able to make out what it is.
Thank you Kakss for replying, because for some reason I was not notified of the earlier reply. Now I watched the linked video and it's amazing! I loved the part where you walk on the feet of your mirror-self ... and the same applies to basically everything else as well. I appreciate all the effort that was put into this. This is the kind of stuff I want to do myself some day. I do, however, think the filter makes it rather trippy and hard to watch. It's not what I expected. Now I'm thinking, maybe a game where most elements are rendered normally and only some of them get the genetic algorithm filter? But that might go together poorly. I can't really say anything without seeing it. I think it'd make a great way to display ghosts, though.
A very impressive little project. Out of interest, is it a project as part of a course, a pet project, a proof of concept, or is there some commercial application you're working towards?
I am a visual effects artist and motion designer and I am really interested in this. Is there a way to play with your code, or can you give me a rough overview, on how you structured this project. (Not detailed, but how to approach it?). I am fascinated by deep learning and genetic algorythms for a while now, and I think there is alot of potential to use it in animation.
What do you mean by visual interpolation? Do you generate individuals and then, in the next generation you have animation between past one and the current best individual?
Not really. Here's a possible algorithm of the loss function (just off the top of my head): Take image, run edge detection on it. Turn it into B/W. (white or black, not grey). Take a screenshot of the lines, perhaps with increased width. Run edge detection on that too, maybe do a blur beforehand. Compare the pixels. See what % of them are correct, which are false positives and false negatives. I'm sure this could be improved further if given more than 5 mins of thought, but you get the point.
well TheFlynCow's way of thinking wasn't bad, you could try: Edge detection AB is one of the vector trying to reproduce an edge do AB.EdgeN * (distance between the two vectors), get the highest value and make a fitness out of it
I literally started lauging out loud while watching this video because I was way too much impressed, especially when you waved your hand over the screen! Incredible.
is there any learning here or the thing just moves the lines to be in closest coordinates to the edge vectors? whats the point, i mean, tell me and ill get fired up about it, really, whats it do besides look interesting?
Two questions: - Don't you love people who upload videos without any description and without giving any context to their RU-vid videos? - Don't you also love people who waste your time by causing you to read needless comments on half-assed RU-vid videos?
If there is a purely mathematical formula for this, then this might be the most important intellectual work to ever be made in mathematics. What comes of the mandelbox? OP Frederic Marquer is on the verge of answering...
i think he used C++ and OpenCV(CV = Computer Vision) to capture and process the webcam image. its very fast. there are tutorials and examples for it.there are wrappers for any language. then he used a genetic algorithm that produce mutations of a random line image with lets say 1000 lines (adjusting the x,y coords of the line points) and compared it with original image. the best image will be mutated again and it starts over again ( iterations on the left side) he just interpolates between the best mutation of every generation of images. so u get a nice line flowing animation. well done.
Great answer. Do you think the fitness function just used some kind of edge detection? Then you could score the best image based on how closely the lines match with the edges. Then the question becomes how to detect edges...
how can the last one work? is that a real time camera? how can genetic algorithm can work with real time camera? it can possible if it mix with neural network but i dont understand how can it work with only genetic algorithm? can you explain please the logic of your algorithm?
Durro Yep, if you could read you would know he used two crossed image of that tower, but since you are not into getting knowledge, you can stay in your flat earth iluminati reptalian world where everything is fake
Took a look at PiCompressor. Just so you know, any n-bit sequence is expected to appear about once in every 2^n bits of pi, so your compressor would have an average compression rate of exactly 0%
This is really interesting i'm amusing each genome represents a lines x,y and rotation? I'm struggle to imagine how the fitness function works if this is the case so maybe i'm going about this the wrong way i don't suppose you have a paper or a source so cover the evolutionary algorithm? Much appreciated.
This is pretty impressive. The style of this is awesome. So what purpose is this for you... I don't mean to sound rude, I just want to know what you are doing with this. Shame you won't let us download this :D.