Slashdot Mirror
NVIDIA-Powered Neural Network Produces Freakishly Natural Fake Human Photos (hothardware.com)
MojoKid writes: NVIDIA released a paper recently detailing a new machine learning methodology for generating unique and realistic looking faces using a generative adversarial network (GAN). The result is the ability to artificially render photorealistic human faces of "unprecedented quality." NVIDIA achieves this by using an algorithm that pairs two neural networks -- a generator and a discriminator -- that compete against each other. The generator starts from a low resolution image and builds upon it, while the discriminator assesses the results, sort of like a constant critic, pointing out where things have gone wrong. The GAN is not a new technology, but where NVIDIA differentiates is through the progressive training method it developed. NVIDIA took a database of photographs of famous people and used that to train its system. By working together, the neural networks were able to produce fake images that are nearly indistinguishable from real human photographs, and a little creepy too.
A few of those example results are a little uncanny valley-ish, but the best are nearly good enough to serve as my dating profile picture. Google Image Search THIS!
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
Do they look creepy? They look like many or the retouched "real" photos you see in the media all the time to me!
This sounds like the standard idea of curriculum learning - you teach NNs via progressively more difficult tasks.
There can be no exceptions!
My ism, it's full of beliefs.
They should add the ability to recreate biometric parameters: Generate a fake picture that image recognition will attribute to a real person with the given biometrics. I'd buy that.
"two neural networks -- a generator and a discriminator"
IOW a democrat and a republican. :-)
Hollywood-Powered Neural Networks Produce Freakishly Fake Natural Human Photos
The rendered images look strikingly like actual human photographs, I'll bet they could fool nearly everyone -- you'd have to have a reason to think they were fake.
I'm wondering if their choice of celebrities as the training database somehow skews their results positive versus "ordinary" people. Celebrities almost seem too uniform in terms of facial features and general appearance. It makes me wonder if they tried with ordinary people if the algorithm woudln't produce freaks because it sees odd deviations among normal people.
https://www.youtube.com/watch?...
These faces are a lot less scary than previous neural-network human faces attempts. https://www.fastcodesign.com/3...
my opportunity to freely express myself with the potential persecution and hangings and such
Something about the faces still looks off
Since photographic evidence is commonly used to convict people of a crime, I can't but help wonder if our legal system will be able to keep up with technology in order to avoid the manipulation that may ultimately condemn an innocent person.
It's quite concerning when the term "indistinguishable" is used to describe technology, as 12 randomly selected citizens can be indistinguishable from a group of morons who are unable to tell the difference between real and fake.
Nah. Looks just as fake as the photoshopped crap in various magazines.
Maybe it looks "realistic" to people whose primary source for people's faces is said magazines. Get out of your mom's basement!
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
When seeing pictures like these in a random profile somewhere, the first idea coming to my mind is that they are fake. Even by ignoring some weird bits (some of them are surprisingly similar to various celebrities), the main issue is the lack of realism. A different story is being able to tell whether the fake picture was created by a quite-bad-at-faking person or a computer.
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
Now clothes catalogs and stock photos can be completely generated.
Fake celebrities will add a whole new dimension to fake news.
Now apply this to human voices; unlimited permutations in games, instead of fixed recorded lines.
but what we'll get is ads that call your name.
I'd clean up their act! They can be EASILY replaced
For how long it takes for a version that works for video, 5 to 30 seconds, and, of course for porn. Would also like to see the generator take a script for a scene as input along with physical descriptions of characters. Instant Hollywood!
S1m0ne! Is that you?
"A sad day in Hollywood... We say goodbye to one of the greats..." -OR- "Identify these celebrities... Only 7% Know the Answer..."
This could take catfishing to a whole new level.
Beware.
LK
"Hi. This is my friend, Jack Shit, and you don't know him." - Lord Kano
You can't get back detail that is missing from a low resolution image, so you can't go e.g. from an SD resolution movie to a 4K one, or at least the result won't look like a movie shot in 4K. Conventional upscaling is basically interpolate-and-sharpen, and it gives only a minor improvement. But while you can't get back the original missing detail, what you could in theory do is generate plausible synthetic detail.
Since this technique seems to involve building up the image through a series of increasing resolutions, I'm wondering if instead of generating a completely synthetic image, you could take a low resolution frame as the starting point, and use similar methods to add plausible synthetic detail. I would have thought that that would actually be a lot easier to generate a good result than if you're trarting from scratch to create a completely synthetic image.
Could it be that our Kazaa-era porn favourites will one day be viewable in 4K quality after all?
Oh no... it's the future.
Now there's a good oxymoron that would make a fitting name for a cosmetics line
"Everyone" vs "Every one"
They lost me on the first caption.
This is what Waifu2x does, for the limited case of anime-based images. It is a neural network based upscaler capable of doing some very good enlargements on comic-like and cartoon-like images.
http://waifu2x.udp.jp/
Who will be the guardians of "Real News" and how will these new technologies be used against an often gullible and deliberately misinformed public?
"and even less of what you hear" I think that old saying has gone way down in value.. So can you now get caught doing anything and say it was "fixed"? Has truth in pictures gone way of the Dodo and dinosaurs?
...dammit.
This all looks quite impressive, but to me it looks like it is 'just' morphing between preexisting images, not generating new ones. With the faces you can clearly spot plenty of well known Hollywood celebrities. And with the objects you pretty clearly see how it's morphing between regular photos, as the in-between images often don't make any sense.
This might of course just be the result of the limited training set, not the algorithm itself, but it looks like this algorithm would need a lot bigger training set to generate something that could truly be considered new and not just be an obvious recombination of existing photos.
It's like the old joke with the neural network that stated "all humans are celebrities" because all it got to see were photos of celebrities.
Normal humans don't look like those photos. You and I don't look like that.
Many of them seem to have some weirdness at the top of the head. Pointedness, baldness that kind of thing, I wonder what that says about the algorithm.
Nullius in verba
This is our legal system:
-- Law: Absolute rules of right and wrong, even though in reality everything is relative. And everybody must conform. Including other cultures and actually sovereign countries/states.
-- Bureaucracy: Separation of rules and intention and observed patterns. Rules must be obeyed, even if it directly contradicts the original intention, and nobody knows anymore for which real cases they were actually created.
-- Guilt: Making the last link in the chain the scapegoat for the entire graph/net of causality. Always imply that everyone has complete control over himself, or at least over his loss of control. And if not... what the heck! Imply free will, even though there is no scientific basis for the concept, and we're only clinging to it out of a schizophrenically-motivated messed-up obsessive-compulsive illusion of our own superiority/specialness.
-- "Proof": Argumentation, based on unverified, not scientifically solid, anecdotal claims, from sources that are by nature relative, distorting and erroneous. (Only psychological/mental damage "doesn't count", because one "can't see" it. --.--)
-- Punishment: Harm is absolutely OK, if you have an excuse that is good enough. E.g. that the other one did it too, or that he is "evil". That grants absolution to all crimes committed against him. Harm is always just when others do it. It's always only the others, who are at fault, and therefore fair game. We harm people, who harm people, because harming people is wrong.
-- Prohibitions: Discrimination on the basis of punishing many for the actions of a few, through (universal) circumcision of freedom.
Is that the system where you're wondering if it can keep up and not harm the innocent?
If after reading my title you didn't guess: naked people. All those photorealistic painting artists will be out of a job.
"Everybody's naked underneath" -- The Doctor
The uncanny valley effect is probably due mostly to the massive amount of digital editing in photos today.
That would go both towards tainting the data set making more blemish free smooth faces and the normalisation of those computer doctored photos in our minds.
If we were less used to looking at, essentially, computer generated faces and the neural net was trying to reproduce something that wasn't, essentially, a computer generated face then the difference would be more apparent
I don't get it. Based on the summary, it sounds like they are taking a picture and then tracing it, and another program is constantly saying "you traced wrong, go back and do it right!". So... they can take a picture, and from it render the same picture? Haven't computers been able to do that since... oh.... forever? Without AI involved at all.
This is actually just blending real photos from a database.
It's not generating anything from scratch, it's progressively layering bits of various photos together and blending. They start low res and then add in details at higher res. Some of the results show this as the general shape looks fucked up or kind of fuzzy at a high contrast edge (jawline or hair line), but the lips and especially the eyes look perfect.
NVIDIA took a database of photographs of famous people and used that to train its system. By working together, the neural networks were able to produce fake images that are nearly indistinguishable from real human photographs.
In the samples you can clearly recognize some celebrities such as Adam Sandler's and Zoey Deschanel's contributing large chunks of an image.
It's still putting out mostly great results, but it's not generating these from scratch. It's playing Mr. Potato Head.
This + flexible screen tech means this cloaking hood shouldn't be that far away.
I find it fascinating how all of us have seen the same movie, but came to a different conclusion regarding how well those two were animated. Some people think Tarkin looked better than Leia, some think Leia looked better than Tarkin, and some think that both sucked. I haven't yet met someone who claimed that both were good though.
Yeah, they both looked wrong to me, very wrong and I didn't know anything about the CGI people being in the movie when I went in. But Leia and Tarkin each looked wrong in different ways. Leia's skin looked fake as hell and was too obviously CGI. Tarkin's body language/facial micro-expressions were all fucked up. Some people aren't very sensitive to or aware of micro-expressions whereas some of us are actually consciously aware of them. To me Tarkin was just deeply wrong and obviously not human.
But eventually studios will hook up with the university psych dept somewhere that has identified many of these micro-expressions and been able to teach AI to recognize and interpret them. I would like to think that data could also be used to help make more convincing CGI human avatars and such.
4K please