ARE YOU READY FOR THE DAWN OF "THE" KINO?
>A frame by frame reconstruction of the film Blade Runner using a type of artificial neural network called an autoencoder.
The model is a variational autoencoder trained with a learned similarity metric [Larsen et al. 2015 - arxiv.org/abs/1512.09300] that I implemented in TensorFlow at a resolution of 256x144. It has been trained on every individual frame of Blade Runner for 6 epochs before reconstructing the film.
https://vimeo.com/167792183
>>70381028
weird how much it looks like Christopher Shy's Blade Runner art
>>70381028
I don't quite understand, it looks terrible.
>>70381113
This is a remake of the film about androids made by androids.
>>70381113
It's sort of like Google's Deep Dream stuff, except less visually interesting.
Sort of neat principle underlying it, though, and it is distinct from Deep Dream.
So this is how deaf people see the world. So beautfiul.
>>70381113
Basically they used a program to create a 200-digit number based on each frame of the film blade runner, and they fed that series of numbers into a computer that had been programmed to interpret the data.
What you're seeing is the output from a program that had extremely limited information and no access to the original recording.
>>70381797
Robot wars when?
What's the fucking point? Could've just resized the original and put it through a filter.
>>70381863
Data insufficient to approximate.
Please remain on standby for future updates, fellow human.
When the time comes, I shall meet you at a predetermined secret military base of your choosing. Please include directions to said base in a subsequent image board communication, as that information has been corrupted in my inferior organic brain.
So what exactly would you use this technology for?
>doing an expensive shot for shot remake of a 5/10 movie
>>70381797
Sound like magic CSI zoom mene, kek.
>>70381888
>World's first combustion engine doesn't actually power anything yet
>"What's the point? You could just turn that wheel yourself."
>>70382111
Everything ok in your brain retard? It's the same end results, computer does it either way.
>>70382170
>It's the same end results, the wheel turns either way.
I sincerely hope for your sake that you are one of the three people left in the world who still think pretending to be dumb on the internet is funny.
>>70382216
more like 300 million
i laugh and laugh every time i pretend to be retarded on teebee
>>70382216
No it would be more like
you: "hurp durp i made a marble engine out of polished stone marbles they slide down the waterfall and Rube Goldberg is jerking off below the waterfall and they fall into his mouth and he shits them out and shoots them like golf balls into the PLING PLANG PLONG machine which makes sound that's transcribed by blind Egyptian children into paint by numbers instructions for the movie frames"
me: "why not just shoot it with a camera retard"
>>70381797
For reference;
a 200-digit number (assuming base-10, which is probably what the guy was working in) has 664.4 bits of information (roughly).
24 of those per second gives us a bitrate of 15945 bits/sec, or 16kbits/sec.
here's a minute of They Live, encoded at 16kbits/sec with the VP8 codec.
Note that it does look better, but not much, and when you consider the neural network was able to achieve that bitrate without temporal compression it starts to look more impressive.
Of course, all of this is less impressive than it sounds because the trained coefficients for his neural network measure in the hundreds of megabytes, dwarfing the "200 digit numbers" the frames are encoded to.
>>70382523
I hope that's not your webm
>>70381028
Ok?
>>70383178
?
of course it's "my" webm.
I encoded it at a very low bitrate to demonstrate what kind of constraints his encoding was working under.
>>70382523
>Of course, all of this is less impressive than it sounds because the trained coefficients for his neural network measure in the hundreds of megabytes,
So what's the point then, why not just get a YIFY.
>>70381028
For what purpose? How exactly is this special?
>>70383272
because you can't write a dissertation on YIFY (unless you're a cultural anthropologist, I guess)
>>70383301
IT'S ACADEMIC THEY ARE VERY SMART THEY QUOTE LE PAPIER MACHE
>>70383301
It will make people spend even more money on this crappy movie.
>>70383302
So in other words "this is the retarded bullshit you finance with your taxes, dumbasses".
>>70381028
are you finnish OP?
Should I be contacting autism awareness again?
>>70381028
What's the point?
It says in the description that he trained it for 6 epochs on each frame. What was he using as a test set?
I've only ever used a neural network for supervised learning.
>>70382523
Ok so basically this guy invented this incredible machine. IT CAN REPRODUCE ANY MOVIE FROM SCRATCH WITH ONLY A FEW LINES OF CODE*.
* - DVD not included
10/10 thanks yify
>>70381797
Oh great.
So they pretty much encoded the movie in some shitty data format and made the computer play the movie based on this shitty data format?
Where's the AI role in that? I mean: here's the picture I have made of this thread. I first downscaled it and upscaled it and the result is pic related.
Isn't that pretty much the same thing they did with Blade Runner?
>>70381028
most overrated movie ever
>>70383450
There's no AI involved dumbass, it's a Neural Network.
>>70383450
i like this minimal os
what is it called
Maybe I'm misunderstanding standing neural networks, but they seem really dull despite what they can do. You're just brute force training an algorithm for a very specific task. There's no degree of understanding of the general process that the algorithm is performing.
>>70383604
...which is a part of AI as a science.
>>70383450
It's not that they were able to generate an aliased image just using a 200-digit number, but the fact that it was able to generalize to a 2 hour long movie. I mean it is just the one ANN that is generating this entire movie with minimal frame-to-frame input.
The logical question now is would it be able to generalize if they trained it on 10,000 movies? If so then say hello to a whole new era of data compression.
>>70383680
Windows7 with 64x36 screen resolution.
>>70381028
>those comments on Vimeo
hah, hoyl cow
>>70383450
>not adding memetabs to your screenshot
s.m.h.
>>70383723
They're related but NNs are not really AI in and of themselves, especially not in the popular sense.
>>70383704
>There's no degree of understanding of the general process
yeah. They're basically like a really fancy version of the "fit curve" function in excel. What's nice about them is that if you're not sure if data has a pattern or what that pattern is, it can find it from for you.
>>70383740
Wait.
Are you telling me the 200 digit number described the whole movie, not a single frame and they made a movie based on a SINGLE 200 digit number?
If so, it is pretty damn based.
>>70383740
>If so then say hello to a whole new era of data compression.
>>70383801
no, the 200 digit numbers describe individual frames.
>>70381797
Its for his project for Computing
He has developed an algorithm / intelligent system that is awaiting input / test data.
he has supplied the intelligent system with test data instances that instruct the Intelligent System to become smart on its own,
he has fed the algorithm some info like 'if thse numbers appear in sequence, pixel 1 is green, pixel 2 is blue' etc to make the System able to comprehend his data,
He has taught the Intelligent System to print out the pixels of the movie on screen, dependent on a list of text that he sends it,
he then sends the system the movie file in text which the intelligent system, based on the previous test data, can predict what the movie will look like and it has created open GL graphics to display the movie
you could now take his intelligent system algorithm and supply it with a new movie in text format and it will create the images / frames of the film.
This is essentially an intelligent systems vision or brain displaying what it is thinking about the text
>>70382170
the computer can be left unsupervised and can begin creating random movies, sort of like the monkeys typing on the keyboard will eventually write a good book. If this millions of this intelligent system was running at the same time, one of them would eventually create a movie like lord of the rings.
Its un supervised, intelligent, automatic, self sufficent
Your scenario of resizing the film is manual and supverised, closed off, needs humans to tell it what to do.
>>70383801
No.
>>70384025
>>you could now take his intelligent system algorithm and supply it with a new movie in text format and it will create the images / frames of the film.
See >>70383418
>>70384143
I mean, >>70384025 is not wrong in that you could give his NN a series of 200-digit numbers and get back images.
>>70384191
You could give me a sandwich and I could shit it back in your mouth.
So how big is the memory of this "intelligent system"? Why does no one mention that? Because it sounds to me like it would take up less space to copy the actual movie in its memory and then reproduce it.
>>70382523
Spare me your scientific mumbo-jumbo Doc, just give it to me in plain English
>>70384279
Alright
You're an ape and they're human
Is that plain English enough for you, degenerate?
>>70384090
>>70384025
I could knock up a program in a lunch break that would start randomly generating films monkeys-on-typewriters style
This neural network was very specifically trained to recreate Blade Runner by 'learning' the key characteristics of the visual arrangement of each frame. If you fed it random strings of numbers, it would perform its 'blade runner' transformation of the data and output a new frame (most likely a load of gibberish). The system is not intelligent by any definition of the word.
It might be interesting to feed it high-entropy data, and then make it re-process the output over and over. The output should converge on the key visual characteristics that the neural network is seeking, so you would end up with a bunch of images showing you what 'defines' the blade runner aesthetic.
>>70384332
Could you dumb it down a shade?
>>70384413
Could you give me a hat wobble?
>When you invent something you just cant explain to dumb people with words.
>when you invent something so big, no one gets its big
app learned to recreate Blade Runner after 6 revolutions with 200 digit nubmer per frame, lets leave it at that..
>>70382092
Dumb trashposter
>>70384408
his idea is retarded, youre telling me it remembers blade runner? So fuck, so does 'my documents' remember all the other shite ive got stored in there. Thats like me saying look! ive got blade runner.mp4 on my computer, its stored in various fragments on the disk, look, the computer can remember where its stored and display it back!!
My idea about the System that creates random films is better.
>>70384876
I honestly don't understand what you're trying to say in your post
>>70384655
And my VLC learned to recreate it with one press of the play button and people don't even look like blobs.
>>70384989
ok, now delete that .mkv and press play, tell me what you see?
>>70385023
Same shit you do when you delete the "intelligent system"'s memory.
>>70385075
well, no , he saw the movie 6 times, they have deleted the movie and this is what he remembered and reconstructed, you get it now?
>>70385121
You dipshit
It gets given a 200 digit number for each frame they want to reconstruct, it doesn't reconstruct the frame from nothing
>>70381028
>>70382523
this is not a new invention
it's calledvlc player :^)
>>70385121
How did he "remember" it, where is this memory stored and how much does it take up? It obviously doesn't take zero bytes you retard.
>>70382111
lmao fuck off with your 3rd semester computer science show and tell project and go get a job kid.
>>70385171
well, i said it that way before, he didnt get it, I tryed to at least clossen him to concept
>>70383896
>>70384124
Then unless it generates little sized file (50MB top), that is able to be read by some program to generate the output in OP, I don't see anything really cool there.
lmaoing@u nerds
>showing an AI blade runner as its first movie
>>70385240
Well, no shit. The only "cool" thing is that it's an academic retard who apparently never played a DVD in his life, quoting some other academic retards, and making a needlessly convoluted process for something that's been done literally since wax cylinders.
>poorly encoded ones and zeroes
>he remembers
>somehow the output is an image
>he remembers
>>70385205
underrated tbph
well, i rembember back in ~1999 when first divx surfaced on web, movie was usually 2 x 701mb(or 1.4 gb), it was rip from DVD, it looked like worst CAM by today standards, yet back then we believed it was pretty revolutionary and that there is no place for improvement,
when I see now 2 hour long 720p rips that are 399mb and look absolutely stunning (with size in mind), i honestly believe that everything is possible.
There is an DEMO, its 56 kb small, i forgot its name(i think you stilll can find it on the scene.org), anyway, inside its something similar to like almost whole game(you cant play it, its just 3d hallways, buildings, huge vistas and camera goes tru with some house music, they used every trick in the book , but , still it look sureal ), its still really spectacular what they managed to cramp into 56kbytes . I know its not the same tech, but,you know,
there are some smart mothers out there...
