Theoretically would it be possible to generate every single possible jpg that is 300x300 and go through them until we find one that shows something cool like taylor swift naked?
You'd probably have to make it greyscale only and do some sort of optimisation so you don't generate images that aren't anywhere close but would it be possible?
>>54346628
sure but you'll have to go through billions of images and have good randomness which computers are bad at.
You won't be able to achieve it on your home pc, easier would be to hack her phone or photoshop her onto a naked body.
But really Taylor Swift looks like an old hag.
you should ask the xkcd guy
>>54346628
Is it possible? Yes.
Is it feasible? No.
The algorithm as you're describing, if it took, say, 1 microsecond to evaluate if the image is interesting or not, would not be finished even if was started at creation of universe.
>>54346663
>billions
kek
But certain combinations of pixels are illegal and you would generate those as well.
Sure, if you have the time to look through trillions of pictures.
I guess, theoretically you could use face-recognition software to filter out all images that don't contain a recognizeable facial structure so you know you're only getting pictures with humans.
>>54346628
Theoretically -- yes. Practically -- no. Let's assume the picture in somewhat decent quality is 60 KB large. Then you'd have 60*8*1024 = 491520 bits. Each of those bits can either be 0 or 1, so that gives you 2^491520 combinations, which is about 10^148018, which is unspeakably much.
>>54346707
Not every combination of octets is a valid JPEG picture. Most of them are not. Your prediction is way off.
>>54346696
>(300*300)*256
...
>>54346628
>generate every single possible jpg that is 300x300
do u have any idea of the space needed to store just all the variations of a 2x2 image? each variation of a single pixel image is 16777216 bytes long, all the variations of a 2x1 image is 281474976710656 bytes long (thats 280 terabytes)
>i really hope this is a bait thread, cause the alternative is just sad
>>54346730
what
>*256
what
>>54346696
>^256
For what purpose
>>54346752
There was two stars, I swear failchan replaced them with 1 you can try it yourself.
But it's 256**(300*300).
>>54346763
300*300 would be one image all the same color.
I imagine he did that to account for a reasonable color distribution.
>>54346721
>Most of them are not.
Just assume that you generate 60 KB of data and then generate an appropriate header. That is always possible.
>>54346696
That's assuming it's only running once though, what if you have every device in the world all running simultaneously on different sections of the collection.
And what if you get everyone in the world to help out as well. Humans couldn't get down to 1 microsecond obviously but they could be pretty quick at determining if it's just random shit or not just by looking at it.
>>54346743
You are telling me that a single pixel image of 256 colours is that long? I think you've worked it out wrong.
>>54346794
now we are talking so theoretical you might as well just send her a tweet asking for nudes
>>54346763
To list all combinations of 300x300 greyscale picture.
I assume that each pixel can have 256 distinct values - from black to white.
With 1 pixel, there would be 256 variations, with two pixels - 256*256, with three pixels - 256*256*256. With N pixels it'd be N^256 and with (300*300) pixels, it would be (300*300)^256.
>>54346794
Great idea! Divide that number by, let's be generous, hundred billion. Hundred billion devices. check out the pic on the left.
>>54346809
please don't reply to bait
>>54346831
you don't need all those colors, simple black or white will do
>>54346831
>With 1 pixel, there would be 256 variations, with two pixels - 256*256, with three pixels - 256*256*256. With N pixels it'd be N^256 and with (300*300) pixels, it would be (300*300)^256.
No.
1 pixel = 256 possibilities.
2 pixels = 256*256.
(300*300) pixels = (300*300)^256.
By the (300*300)^256 logic, a 1x1 image would have 1^256 = 1 possible value.
>>54346831
>>54346696
That is 256^(300*300), retard.
>>54346850
It is greyscale.
>>54346743
You don't have to store them. Generate, evaluate, throw away.
>>54346786
I'm pretty sure you'd still be getting tons of junk data which will result in repetition.
>>54346858
>>54346831
Fuck, other way, 256^(300*300).
>>54346809
>24bpp image, 1 pixel, 3 colors ranges, each color range 1 byte
unless you are talking about a 8bpp image i got the math right
and going grayscale wont save you any much if you want all the iterations of an image
>>54346858
>>54346869
Okay I fucked that one up.
256^(300*300) is the right answer.
But that's a much, much greater number even.
>>54346831
Fuck
>>54346870
>I'm pretty sure you'd still be getting tons of junk data which will result in repetition.
I'm pretty sure that with a little work you can make sure that no junk data is generated, JPEG is not really that complicated, it's actually quite robust.
It would be faster to just find the image in Pi. Every combination of digits is already there, you just have to scroll through them.
>>54346870
>It is greyscale.
not what i meant, you don't need the values inbetween. just use black or white.
>>54346879
>1 pixel would have 256 variations instead of 16777216
>wont save you any much
???
>>54346628
>Theoretically would it be possible to generate every single possible jpg that is 300x300 and go through them until we find one that shows something cool like taylor swift naked?
There was a Foxtrot comic way back in 2000 or so that had this as the conceit - that every file on the computer dude's website was randomly generated and just so happened to resemble copyrighted material.
>Every combination of digits is already there
only if you assume Pi is infinite.
>>54346926
Hey
>>54346696
To give a perspective 10^80 is ~ the amount of ATOMS in the observable universe
>>54346914
300x300 with just two colors is not very descriptive.
>>54346900
>JPEG is not really that complicated, it's actually quite robust
it is against noise, not against junk data, crappy data will either increase the filesize or create color aberrations on the output
>>54346914
1bpp? like in a fax?
this thread is sad, im out
>>54346628
Actually with deep learning you could generate a naked picture based on a non-naked one. You'd need a lot of clothes / no clothes people as input, but it would be decent.
>>54346953
it'll be alot faster than using greyscale
>>54346953
That's good enough for me, it gets the point across
>>54346946
now do the age of the universe in planck units.
>>54346663
hahaha why would you use randomness? If you want to generate every possible permutation you use some exhaustive algorithm, not RNG.
>>54346926
>mfw the offset takes more space to store than the image would have
>>54346967
Nobody cares about you.
>>54346917
do the math yourself for a 10*10 8bpp image
jesus christ...
>>54346984
randomness =/= RNG
So wouldn't it just be 2^(300x300) now?
>>54346975
>>54346979
Fine. But alas, since my previous formula was incorrect, it severely underestimated the difficulty of the task.
Here is how many variations exist of 2-color 300x300 picture. A number that won't even fir no matter how many times I click more digits.
Even a 50x50 b/w picture has 2^2500 = 3.758·10^752 possibilities which are impossible to search through.
>>54346981
8e60 which points out an interesting fact. if time is simply another dimension, the universe is wider than it is old.
>>54346953
you could store somesort of seed, when op saw an image he liked, he could generated it afterwards in higher and better resolutions/colors
>>54347008
Why would I? All I want is to show that using 8 bit color will result in much, much less combinations than using 24 bit color.
>>54347019
>27 093 decimal digits
that's not that much.
op needs something to predict the images in a range, that way he can cut out alot of junk images.
I guess we need quantum computing or a smarter method then.
Fuck.
>>54347041
>27 093 decimal digits
>that's not that much.
>27 093 decimal digits
>that's not that much.
>27 093 decimal digits
>that's not that much.
>27 093 decimal digits
>that's not that much.
anon seek medical assistance
>>54346663
google image search "old hag naked" then ctrl f "taylor swift" the the similar image results.
>>54347054
you literally have a better chance breaking into taytay's lair and taking a picture when she's stepping out of the shower
Okay what about we take an image that looks sort of like the image we want and then keep changing it randomly until we find what we want.
I was thinking about this a little bit.
Instead of checking pictures algorithmically, I was thinking about letting a human do it.
i assume that we'll order the pictures in the picture space like this:
where `n` is it's number, '*' is a black pixel, and '-' is a white pixel0
- - -
- - -
- - -
1
* - -
- - -
- - -
etc..
First of all, we know that the entropy of pictures that are interesting to humans lies within a certain range. Find out what this range is, and you can already remove many noise pictures.
Second, we'd need to imagine ways to meaningfully navigate the picture-space. If we can repeatedly and quickly jump over a chunk of pictures, then we might be able to close-in to our target, which could be anything!
One such a jumping mechanisms could be based on the property that pictures near each other, say 0 and 1, look a lot like each other.
You can skip over amountOfPixels2 pictures to move on to the next "set" of pictures.
Find meaningful jumps, and we might be able to do something here.
>>54347033
>Why would I?
/g/ - technology - millennial edition
good show old chap, perhaps next time you will do better
>>54347132
Nice work, you've got us closer to the goal
>>54346743
I guess you would compress an image before you safe it,
So generate, compress, save, repeat
>>54347132
>manually random generated indexing photos
are u a poo in loo looking for an awful job?
>>54346628
Would it be technically possible? Yes. Would it be feasible? No.
You're better off trying to date the girl herself than getting this to work.
>>54347134
You're missing the point. I did do the math for 300x300 greyscale picture. Because I did the math, I do know that you'd get way, way less iterations compared to 300x300 24-bit color picture.
Maybe /g/ get together and make a program that does this. I have a great logo that we could use for it.
>>54347170
There is no need to save. Generate, evaluate, throw away.
>>54346628
They were just talking about this concept on a podcast and I can't remember which one. Getting Doug With High? All About Android? Pistol Shrimps?
>>54346628
Actually, you just need to find a bikini pic of her, cut out the interesting parts and iterate those. It would probably be a good idea to not go full random, but to approximate and then alter the image, i.e. you know that it's mainly skin color, so you can start from there. You could probably develop some AI engine that browses through porn to find the best matching pair of titties or something.
>>54347237
Yeah that was one of the mentioned potential solutions
>>54347237
You'd also have to develop another AI that decides if the resulting picture looks realistic. That's a much more reasonable approach - but of course nowhere near as simple to implement.
>>54346785
256 could be 8 bit colour
an 8 bit 300x300 image, thats pretty shit tho
>>54347270
Just crowdsource it, I'm sure people would do it for free
>>54347019
>2^(300*300)
lel you fucked up
>>54347277
OP was even going to settle for 300x300 1 bit color image.
>>54347283
At lease I can be sure they'd get the logo done.
>>54347219
Giant Bombcast did it in some old episode, I think.
Shouldn't a quantum computer be able to output the image before the program was even uploaded to it?
>>54347212
Fuck you, my logo is better.
>>54347295
I'm pretty sure I'm in the right though this time.
NOT ONE FAGGOT IN THIS THREAD HAS HEARD OF PHOTOSHOP
>>54347307
How about you both post them and then we will decide
>>54347132
That's not going to help with the size numbers we're dealing with. We're trying to bring 10^216741 down to a manageable number.
Even if it were brought down to 10^50 combinations to go through, and you could somehow solve 1 every clock cycle with a 3 ghz processor, you would bring it down 3.3*10^40 seconds to solve. Which is 1.046 * 10^33 years. Comparatively, the universe is 7.6*10^22 years old.
>>54347304
The problem is not generating the images, that's simple, the problem is deciding which one is the correct one. I doubt that quantum algorithms are anywhere close to that level of complexity.
>>54347304
Ah, no, a quantum computer would not be able to do that, because a quantum computer does not exist.
>>54347321
Mate it doesn't count unless it's what she really looks like and the only way to find out is to randomly stumble upon a series of bytes that represents what she looks like
>>54347321
You're completely missing the point, adobecuck.
>>54347333
I'm pretty sure I read last year that a quantum computer was indeed built.
Unless I'm confusing it with quantum memory, but I doubt it.
>>54346628
> run random image generator
> generates cp
> gets arrested
wew lad
>>54347316
what are you smoking? >>54347019 's 2-colour result was 26000 orders of magnitude larger than >>54346696 's 256-colour result!
granted, it wasnt a huge fuck up, just switch the exponent and the base around and you get a more reasonable answer
>nobody posted this yet
https://babelia.libraryofbabel.info/
>>54347189
>astronomical number is way, way less than astronomical number
thx f.a.m.
>>54347327
My point is that with the right jumps, you could skip chunks, that is *percentages*, of the space. That means it doesn't matter how big the state-space is. We're looking for algorithms that helps us navigate their insane sizes. I really have no idea what kind of jumps would help us out here, but I'm just going to try some stuff and see where I end up.
" a monkey hitting keys at random on a typewriter keyboard for an infinite amount of time will almost surely type a given text, such as the complete works of William Shakespeare."
>>54347362
>>54347406
It'll fuck up the accents.
>>54347342
>what she really looks like
>randomly generated bitmaps
>>54346697
I know what you mean.
>>54347380
It's pretty fun to read that post because every quote has (You) after it.
Yes, my evaluation in >>54346696 was completely off. It should be 256^(300*300), and it's a much larger number.
>>54347399
No problem pal. Less iterations is less iterations. You get an astonomical amount of iterations less.
>>54347406
> tfw a random image generator could create a picture of your children that you never had
> tfw a random audio generator could crate the dialog between you and your wife that you never met
> tfw
>>54347359
They demonstrated something that they claimed was a quantum computer. They could not achieve anything a normal computer couldn't with it.
>>54347472
> a random DNA generator could generate a perfect replica of my wife's son
>>54347448
wait, is it (colourcount)^res? I thought it was the other way round.
huh
>>54346628
Why not just ask her nicely for a nude pic?
Here's an idea: how about evolution?
The "natural selection" would select against "what picture is more likely to look like Taylor Swift?"
Then you select from those any that look like she is naked. Or do another selection.
Actually, I just got reminded of someone who actually succeeded at solving a similar problem, I have *NO* idea how he did it however.
https://youtu.be/Qg3XOfioapI?t=1725
Tldr: He did what we are trying to do but in audio-space, he's iterating over all possible songs. The huge number you see skips to a particular part in the infinite-sequence which sort-of represents "blurred lines". He somehow managed to do this.
Though, I suspect he created the song himself, encoded it as an index for that infinite-sequence, and just stored the number.
Regardless, it's interesting.
>>54347440
One of the sequences is what she actually looks like logically so yes
>>54347479
Because all a "computer" needs to called that is be *something* complete.
Sorry, I forget the term.
Any real computer should be able to do what any other real computer can.
>>54347403
You could skip 99.99999% of the space and not be meaningfully closer to shrinking that number down to something that could be calculated in the human lifespan.
>>54347578
I thought of that. It's a possibility but I have no idea how a computer could do that, it would have to be humans selecting
goddamnit people
>>54347384
>>54347384
>>54347384
>>54347384
>>54347619
I'm not talking about skipping once, I'm talking about skipping repeatedly. Sorry for the confusion.
>>54346628
You are late senpai
https://babelia.libraryofbabel.info/
>>54347635
What? No. There is a field of evolution on computation. You could do it by face recognition software.
>>54347578
Okay, let's try this.
First iteration.
>>54347593
if weve never seen her naked, how the fuck would you know if the sequence in question is what she really looks like?
>>54347676
And how long is that going to take? It's not going to be a quick algorithm
>>54347684
Start with a picture that is almost the one you want and then go from there
>>54347619
The trick is going deeper.
go into the first dream, then thew second, then , the third, and then make your computer in the fourth dream. Since one minute in real life is like 50years down there. You should be able to get it to output the correct image of a naked tailor swift. Now you just need to remember the correct string and order of the 1&0s or hex if you want to memorize it easier because of less numbers to remember. And remember it all the way up to real life.
Remember to make a totem first.
>>54347684
I love you
>>54346628
http://norvig.com/atoms.html
3x4 pixels is more combinations than all the atoms in the universe. Good luck cycling through those.
>>54347684
>>54347694
But you don't really need her face to be there. You just want her body.
>>54347656
That approach is memoization; the problem being the memory space would quickly become intractable.
>>54347768
>Not reading any of the discussion before posting
>>54347684
I'm pretty sure those two images are the same but inverted both horizontally and vertically.
>>54347776
But the body has to match the unclothed parts, that is, the face, the arms, the legs, etc.
Otherwise we could generalize the search from "taylor swift naked" to "thin woman naked"
Say we start with this image.
It's almost what we want. We know which pixels need to be changed and we also know the range of colours (skin tones) that could fit.
Then we simply generate a ton of possible images and get humans to compare them 2 at a time until we find the best one.
Then we mutate this one and find the best and so on until we get the true solution.
>>54347866
Or we could give her a raptor head instead of taking off the bikini.
>well to go past the mountain you need to follow a steep road
>nonono, to go past the mountain you need to create ALL of the mountains, roads, seas, plains, cities, animals, microbes, planets, moons, galaxies... and then follow the steep road
>>54347866
Ok sounds good i'm down who has a neural network we can use
>>54346628
do people not understand powers or something?
there was a story from the way back times that pretty well explained powers, it was something like two guys playing chess and one guy was like if you beat me what do you want as a prize and the other guy was like ok, just some rice like this, one grain of rice on the first square then double it for every square so two on the second square of the chessboard and 4 on the 3rd square of the chessboard.
so the trollface grain guy won the match and then was like dude where is my grain so the other guy was like ok lets work out how much grains of rice i owe you and so he started counting it out
but if you do the maths the amount of rice in total would require two whole planet earths covered in rice paddy fields to produce the amount of rice the guy was due.
so like a fucking ridiculous amount of rice
the point is that like its way too easy to underestimate things, like you can't fucking imagine how many grains of rice it was.
a 1x1 pixel image of 1 bit has 2 possible combinations
2x2 with 2 bits is 16 combinations, so 8x
3x3 with 3 bits is like 72
look at the growth rate its incredible.
>>54347866
That's not a bad idea. I would still use "natural" selection though.
Is this easily doable? I mean, take the part of the bikini near the neck. Since the string is thin, the skin under it could be an average of the 2 regions at the left and right side of string.
If we want to simulate a tan, make that lighter.
>>54347937
That's why we abandoned that plan and are now working on finding a method that works in polynomial time
>>54347955
This. Keep up >>54347937, senpai.
>>54347937
You forgot to mention the 256^3 possible colour combinations for each pixel.
>>54347866
Why does it have to be taylor?
Why not genderbent moot?
>>54347866
too much work for meh results, if someone takes some minutes of work on photoshop anyone could produce pictures of naked taylor that's 99% accurate.
>>54347983
It doesn't have to be taylor, it could be hitler in a dress fucking a dwarf.
>>54347955
>>54347983
>>54347989
>>54348001
>go to a guy good at drawing
>ask for tay nudes
>ask for hitler in a dress fucking a dwarf
To solve op problem you need a human level ai.
>>54347674
Anybody else have it open in case an actual image does come up?
>>54347955
its an excercise in futility since you don't have an image to validate against
here is a method that works in O(1) time along these rules: cut and paste boobs using photoshop
>>54348001
>>54347983
I think it's easier to proof. Say the algorithm is perfect, and produces a picture of a naked taylor.
How do we make sure it's her actual body?
Just proof it on clothed/unclothed people. My guess is "naked girls" has the most amount of pictures in the internet.
>>54348049
I've seen her in my dreams, I can tell
>>54348046
>>54348049
Consider the following.
There are people that look a lot like other people. Like that lady who will do porn because she looks like Ted Cruz.
Therefore, although the number of possibilities is theoretically infinite, it is limited by the chromosomes, right?
Similarly, boobs should follow the same rule. So instead of generating random images, just grab boobs from the entire internet and try to match up using an algorithm.
Again, proofing would be essential and easily doable, given the large amount of pictures of naked ladies.
>>54346628
this is basically the infinite monkey theorem with a motherfuckton more permutations. Technically yes you could, but as said it would most likely take longer than the lifespan of the universe
>>54348099
i wonder at what resolution/color depth it would become computationally less expensive to simulate every DNA combination and model boobs than to simulate the image
I don't know if i agree about proofing, are people interested in the image if they know its not actually real (or a computer generated but essentially perfect copy) but a 'best guess'
>>54348194
I think it would be way more efficient to generate the images in B&W, use facial recognition software to basically just constantly search google for close matches, and then just add colour to the good ones after.
Wasn't there a webpage that was already doing this? Anyone posted that yet?
>>54348259
imma need a sauce on that
>>54348194
The human genome has 3 billion base pairs with a bit depth of 4, BUT much of that is required for basic life functions and many sections are passed on in their entirety. So let's say human DNA is 99.99% static, that leaves us about 300,000 base pairs to shuffle about, but that's at a bit depth of 4. So we're looking at 4^300,000.
>>54348036
> autism
enjoy the eternal noises
>>54348259
are you talking about this
>>54347674
>this thread
http://www71.zippyshare.com/v/Qga2PTdX/file.html
copy the number in that file
https://babelia.libraryofbabel.info/imagesearch.html
paste into the text bar
>yfw if only /g/ had the gpu resources we could train a tay nudes neural network right now
>>54348411
Can I do, I has r9 290x
how do
>>54348411
We could infiltrate Nvidia and hide a botnet within a driver update
>>54348377
>>54348352
Yeah that. >>54348342
This is obviously bait, but I'll bite:
The problem with this is that even if you were to filter through the images that are completely trash or white noise, you would find yourself with other images where yeah, you might find a nude of taylor swift, but she's got double d's. And then there's another taylor swift, but she's got a dick and a vagina. After another random sequence, you'd find a taylor swift where she has the head of a human, but body of a horse. And even if you were to get a similar body shape as hers, after so many combinations, how do you really know that's her? How would you know that the cup size, labia shape, amount of pubic hair, or anything else we don't know of her's already really is her? I'm sorry OP, but the odds are definitely against you. You're more likely to find a nude of someone else than anyone in specific, especially taylor swift. And again, that is if you were to even find a way to filter through the corrupt images, images that are just noise, and others that are just complete garbage.
>>54348377
>>54348472
better yet, paste this
https://ghostbin.com/paste/mz3er
>>54348496
>mfw that website is probably full of cp
>>54348485
>where she has the head of a human, but body of a horse
I'm totally fine with this.
>>54348549
You could get better results with photoshop if you wanted a picture like that so bad
>>54346628
It is possible. There is a site with every possible book on it
This "randomly generated image" meme is in the same bullshit league as bible code fucks
>>54346628
Well it's pretty easy to see that you couldn't do it in your lifetime even with a billion home computers so... this thread is useless.
>>54348658
>bin laden
>not the jews
kek
>>54348695
Image is dating from when he was the big topic. Get a big enough book and set a computer to find what you want and it will, just ignore the fact that the character leap between them is 5K or more or whatever. They did the same shit for Moby Dick but nobody cared. My favorite part about this you might hear one day
>THEY (who?) DIDN'T WANT US TO HAVE THIS INFORMATION UNTIL NOW, THEY KNEW WE WOULD HAVE COMPUTERS TO DECODE IT ONLY NOW
>>54346663
>Randomness
No, you'd need to go through every possible iteration.
>>54346696
You could probably shorten that considerably by dropping a few colours but yeah, it'll still be unfeasible.
>>54346628
>something cool like taylor swift naked?
>>54347022
Your mom has that same problem.
>>54348695
>the jews
>not the U.S. government
O I am laffin'
>>54346901
Is there a program like this out there? Like that loads the digits of pi into a buffer and renders images in real time while you scroll through pi calculated images?
>>54349044
>implying that the US government isn't run by the kikes
eh you won't like the result OP.
> http://www.nerdcore.de/2015/12/04/neural-network-dreams-of-manga-trump/
that data set was a reverse generation from the learned images submitted to
> http://illustration2vec.net/
which was anime and hentai images, the tagging is secondary.
theoretically you train an ANN with pictures of Swift's face and nude females with faces blurred slightly. you might get it to 'think' all faces should should look like tailor and then draw a body with her face. but srsly photo shop would be easier. i'm sure this has been done.
> http://www.hecklerspray.com/taylor-swift-nude
>>54349337
>>54346972
now that is an idea i can get behind
>>54346901
Isn't that true of any irrational number?
Fucking creeps, stay away from my fair lady
>>54346696
That's a lot of numbers.
>>54346663
How dare you insult our aryan queen and future savior of the white race
>>54347554
Nah.
The max size of any integer on a 64-bit processer is 2^64. That's because it's <options>^<spaces>
I wonder if you could train a neural network on a shit load of taylor swift pictures and then throw in some nudes of someone else and somehow make it spit out a taytay nude. Anyone with ML experience think this is possible?
>>54346628
i've always sort of wondered the opposite: like is anyone autistic enough to sit down and write out an image in pure ones and zeroes?
like this>>54346953
is it even theoretically possible for someone to command such knowledge about image file formatting to generate something from raw binary data?
>>54347866
Evolving a neural network to produce nude pictures of Taylor Swift ?
I love it.
>>54347132
>Find out what this range is, and you can already remove many noise pictures.
literally google reverse image searching: reinventing the wheel edition
>>54347866
this is the first suggestion ITT that is actually easier than simply getting real nude pics
I'll be waiting
>>54348695
it's literally written in hebrew
>>54348549
These are already posted regularly on /b/
it doesn't matter if you had the storage space (you don't) or the computing power (you don't) to generate every possible 300x300 image, because you would die long before ever sorting through an infinitesimal fraction of them, much less finding the one that contains taylor swift
>>54346628
How would you even know its her boobs?
>>54348377
why is this running so slow? Is it deleting system 32?
>>54346628
you know what time it is, OP?
its time to find a real gf
>>54346628
Generating the images? Algorithmically trivial, though it would take a fucklong time.
Checking the image for being something you want? Would require computer vision so advanced that it would be more efficient to have a program paint a super-realistic picture of naked Taylor Swift.
>>54346628
Yes, there's actually a function that you can generate all possible pixel images on a 19x70 grid, I think that's the size. Probably not a trivial task to generalise it tho
>tfw i ransomly generated a picture of Taylor Swift naked the other day by accident.
>tfw I'm not fucking sharing with plebs
>>54347866
would it mutate accurately though?
you can't just pixel swap the bikini because then her boobs would just be suspended in perpetual impossible perkiness.
the network would have to know to find realistic boobs with realistic physics pertaining to the scene
though I suppose that's what the human element is for
>>54347026
thats not how it works you little shit. At _best_ you'd end up with a nightmare deep dream image.
>>54346628
Yes. And the program to do so isn't hard to make. The problem is that even with a 300x300 square using 32bit color, you've got more images than you could ever look at in your lifetime. Not to mention that the majority of it will be a mash of random colors. I've tried it before. Fun messing around project for random permutation but don't expect anything real.
Similar concept is the library of babel which is pretty neat since you can query it. Maybe if you could query images from your program it would be worth something. Basically give it a section of image and ask it to bring back all pages that have that somewhere in it. Only problem is it's essentially just placing those pixels in a random spot on the page and then everything around it is also random.
>>54347513
It could even do it without using your DNA. Oddly enough, you don't even need your DNA to generate a replica of your wife's son by conventional means.
>>54351057
No gf can ever replace TS.
>>54346663
>have good randomness which computers are bad at.
When will this meme die?
>>54346972
this is a far more suitable solution
>>54347041
it's so much that you literally aren't even comprehending how much
why would someone want to see taylor swift naked?
>>
there are much cooler examples of a similar setups, here are some experimentation from 2011
>http://www.pouet.net/topic.php?which=8357&page=1
>http://wurstcaptures.untergrund.net/music/?oneliner=
eg:t&(sin(t&t&3)*t>>3)/(t>>3&t>>6)|((t>>((( t>>6)&19)+11)&3)<<6)@ 11khz
>>54351945
this was meant to be a reply to
>>54347588
but i fucked it up
anyway here is another one((sin(2000/(t&0xfff))*127)*0.2 + (( (t<<1)*(1+0.333*((t&0xffff)>0x7fff)+0.177*((t&0xffff)>0xbfff )) )&0xff)*((t&0xfff)/0x1fff)*0.4 + (( ( (t>>4^t>>6|t>>10))|(t*3*(1+0.333*((t&0xffff)>0x7fff)+0.177*((t&0xffff)>0xbfff))))&0xff)*0.25)@ 11khz
>>54351847
>"I had a great time today, anon."
>>54346901
https://github.com/philipl/pifs
>>54346953
I could fap to this.
taylor is an actual /b/tard, i wouldn't be suprised if someone screencaps this thread and posts it around to laugh at you spergs and she see this :^)
lol
>>54346972
you know theres already a lot of clothes/noclothes data out there... similar poses and all, just hit up /b/
>>54347010
he is still right though, no need to waste processing time on duplicates.
>>54347127
yeah this is good, we train a neural net to change parts of an image and validate fitness using something like isitporn, though inb4 it just photoshops dicks on everything
>>54347595
Turing Complete, bro. After the greatest gayfag the earth was blessed with.
>>54347593
Well if the picture is the same as if we had actually taken the picture.
All of this is making me feel uneasy about the validity of recorded data in general. A picture doesn't really have a link to a person. The person isn't 'in' the picture. The picture just happens to look like what the person would look like given the scene, lighting, etc. There's no reason to believe that even a film camera might not have generated a picture that 'looks like the person really does' without the person present. Maybe you pointed it at some mist and it randomly reflected the same set of photons.
As you can see, Judge, you can't convict me because I was /really/ fapping to an actual child.
If you take every person that has existed, exists and will exist; and make them look at the images in a rate about a million images per second from the beginning of time, the universe would die out before they could find it.
>>54352449
Just means that taylor swift naked images don't exists.
>>54347694
If all sets are generated, then the logical conclusion is that at least one of them is "what she really looks like", and several magnitudes more sets are "what she probably looks like, and an algorithm would verify with 99% certainty that the image is not doctored"
>>54346628
>Theoretically would it be possible
No, it wouldn't. And "theoretically" doesn't mean "with disregard for all laws of nature".
Unless that is what you meant, in which case yes and the moon is also made of cheese.
>>54349035
underrated fucking kek
>>54353421
>No
Yes, theoretical you can most definitely create every image of a set pixel count.
There being 16,777,216 different colors for a pixel might prove difficult though.
>>54353579
>Yes, theoretical you can most definitely create every image of a set pixel count.
I don't know what universe you live in, but in the one we're currently communicating in, there are theoretical upper bounds as to how much information you could actually process given the amount of energy in the observable universe.
>>54347406
It was the best of time, it was the BLURST of times?!
>>54347448
Wouldn't it be easier to not go about this totally randomly but add restrictions like that not all neighbouring pixels can have different colours?
Couldn't we even sample Taylor Swift's hair color, her skin colour and tell the algorithm to create clusters that are comprised of mostly that?
>>54347020
that is some hot bentover :)
almost calling mods on this one
>>54348377
>The image you have tried to link was not accessible to our server. Try following the link yourself, downloading it, and using the upload feature instead.ERROR: Couldn't connect to server ::::: FUNCTION: perform
wew lad
Since we are on the topic of neural networks I have a question:
How flexible are they? Say I want to make an image recognizer that has the task to only recognize a certain model of car.
So I train it on pictures taken of the car, and other cars, from all kinds of different angles. Would this work? How big would the training set have to be for a 80%<= reliability? Would the network be able to recognize the car from a slightly different angle that was not in the training set?
Would the background of the image throw the algorithm off? Would the background need to be made in a single color first?
Also: what about the resolution of the training set? Wouldn't scaling the images down to something like 128x128 turn them into pixel mush that can't recongize the fine details of the car?
>>54347684
I consider making a page for that, and then use humans to get the best tailor swift image.
>>54354566
Pretty damn flexible, but a bitch to program. I recommend starting with Python and Theano. It's not very flexible, but it's a good introduction. You can program shit like an letter/number recognizer and eventually, facial recogntion. After a while, you should looking into making one yourself. That way, you have full flexibility and control.
>>54352313
we're talking millions, not hundreds
>>54354445
fucking kekking
>>54354476
Make a powerful AI that will limit to gender and age as well.
Or make a more powerful AI that will just draw Taylor Swift like one of your French girls. You'll finish the AI before the other anons with the random 2 color image get their pic.
The random anons will have a monkey typing Hamlet out on a type writer though as well as Taylor Swift nude.. So I think they win in the end
>>54348240
This could work. A 2 step process, first zero in on general shape etc of image in black and white, then color in later. Kinda like how artists do their work, they generally start with blobs to get the proportions looking right, then add detail later.
>>54350191
No. It is not proven true for any irrational number (except a irrational number composed of all natural numbers (0.12345678910111213...)). Pi may have infinitely many digits, but does not necessarily contain all possible combination of digits, it could resolve in a number of unending but numerically finite sequences for all we know (3.14159...101101110111101111101111110...)
>>54347588
that's fucking brilliant
>>54354566
>How flexible are they? Say I want to make an image recognizer that has the task to only recognize a certain model of car.
Something similar to this has actually been done: arxiv.org/pdf/1506.08959v2.pdf
>So I train it on pictures taken of the car, and other cars, from all kinds of different angles. Would this work? How big would the training set have to be for a 80%<= reliability?
That sounds doable. They amount of images you need for good results varies hugely for different tasks - but on the low end of the spectrum, people have gotten 91% accuracy on a dataset with only 30 images for each of 101 categories(caltech101). On the high end, imagenet has 14 million labelled images.
As a wild guess, 300-1500 labelled images of the car should get good results. 80% isn't very high.
>Would the network be able to recognize the car from a slightly different angle that was not in the training set?
Quite likely, as long as the rotation isn't huge.
>Would the background of the image throw the algorithm off? Would the background need to be made in a single color first?
I imagine this is a problem only if the training set has many images where the car takes up a small(35ish% or less) area of the image. Convnets tend to be good at learning which parts of the image to focus on.
If this happens it might work better to train 2 nets: one to localise the centerpoint of the car, and another to recognise the car based on only the pixels close to the centerpoint. Making the image a single color would might lose valuable information.
>Also: what about the resolution of the training set? Wouldn't scaling the images down to something like 128x128 turn them into pixel mush that can't recongize the fine details of the car?
You do lose some information scaling down images, but not as much as you'd think. You don't actually need very fine-grain images to get good results for most classification tasks.
>>54350191
>Isn't that true of any irrational number?
Nope. Simple counterexample:
0.10100100010001000...
>>54346901
that's not how infinite sequences work retard
listing every odd number is an infinite sequence that will never have the number 2 in it
Why not just use black and white images? 2*(300*300) = 180,000 different possibilities.
>>54356941
Pi has the property of not having any repeating sequences. Theoretically, pi contains every possible number as a string of digits.
>>54357017
sugee na
>>54347588
That is amazing stuff, thanks for sharing that video, opened up a lot of ideas for me.
>>54357017
It's like you didn't even read the post you were responding to
>>54347588
Well done.
Amazing
>>54346628
What you want is a library of Babel but for images.
https://en.m.wikipedia.org/wiki/The_Library_of_Babel
You'll get nowhere.
>>54354506
same for me
>>54350515
From the results I've seen Google produce, that's very possible.
On the other hand, there's a fat chance the neural network will cover the nude bodies with approximations of the clothing Taylor was wearing in the images you fed the network, if it recognises the human body shape.
>>54357013
It's 2^(300x300)
>run algorithm
>go at the speed of light so time slows down
>You can now run the algorithm and have no worries about dying of old age
Have fun watching you're naked mum
>>54346972
We owe it to the world to make this a thing.
If you make an algorithm that can generate all possible jpgs that is 300x300 pixels.
Could you get charged for possession of CP?
Or is it a grey area?
>>54360908
I'd say yes and probably production too because fuck you
>>54360908
You'd also get images of the judge and prosecutor doing illegal stuff with which to blackmail them.
Hell, you'd also get images of your mom getting rammed by 3bbc in every hole of when she was nursing you when you were a baby.
>>54347684
Definitely A
>>54349035
