NVIDIA-Powered Neural Network Produces Freakishly Natural Fake Human Photos (hothardware.com) 140
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.
Not Bad (Score:5, Insightful)
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!
Re:Not Bad (Score:5, Interesting)
The real trick is when they are animated.
I remember a back in 2000 where they were showing screen shots of the upcoming final fantasy movie. The screen shots looks like real people without the uncanny valley. However when they started moving and talking then it came to light.
Granted graphics and animation have improved greatly in the past 18 years but I hold my doubts until I can see the rendered images move and interact.
Re: (Score:3)
yeah, not quite there yet - see tarkin and leia in that last star wars
Re: (Score:3)
I hope there comes a day when there are no more overpaid actors. Let THOSE jobs go to computers! We aren't there yet but this is a step in the right direction.
Re: (Score:1)
And few would go to see the movies. One thing that keeps people coming back is they can see a star they can relate to. Stars generate press for themselves, they come with a backstory, and some with stories we'd rather not hear. However, this is what the proles see as interesting.
Re: (Score:3)
Re: (Score:2)
While I go to the movies for a sense of escapism, there still needs to be a sense that what I am watching is fiction if it is fiction. I am OK with other actors playing the part of actors who are unavailable, dead, or have aged to a point where they will not fit in the movie. Sure they may look different and sound different, however If I am willing to suspend my belief for Space Wizards, Technology based loosely on shaky scientific hypotheses, I can suspend it to say this guy is the same guy that was pla
Re: Not Bad (Score:1)
"And few would go to see the movies."
For the most part I absolutely disagree. Some of the biggest movie/series ever ($$$$) were not about the actors: Avatar, Transformers, Star Wars, Star Trek, Batman, Lord of the Rings, etc. not to mention all of the animated works of Pixar and Disney. In my opinion the character matters, not the actor. Only established actors are a potential box office draw. If and when digital works replace established actors in a convincing and cost effective fashion, so long to real
Re: (Score:2, Interesting)
People went to see toy story and other animations. They went to see Avatar. They will see movies with synthetic actors too - when those get good enough. No more overpaid actors or problems with stunts. No body doubles, no issues with nakedness or "I won't play that sort of character". Instead of actors they pay a team of animators, but those are more replaceable and can't demand crazy pay.
After a while, some of the synthetic actors will become famous, and attract moviegoers just like a real star (or like m
Re: (Score:2, Informative)
Already happened in Japan - Hatsune Miku virtual singer
https://www.wmagazine.com/story/hatsune-miku-crypton-future
Re: (Score:3)
And few would go to see the movies. One thing that keeps people coming back is they can see a star they can relate to. Stars generate press for themselves, they come with a backstory, and some with stories we'd rather not hear. However, this is what the proles see as interesting.
Easy google search if you require proof, but the gaming industry does more business and makes more money than the movie or music industry. Stars generate press for themselves, they come with a backstory....that's *also* true in the gaming industry. Mario, Samus, Master Chief, Cloud, Zelda and Link, etc.
If you can computer generate a superstar in a franchise that is virtually indistinguishable from a real person, in a franchise or industry that makes more money / has more fans / gets more attention than a
Re: (Score:2)
BRING ON THE PROCEDURALLY GENERATED ALIEN TENTACLE PORN!!!!!!
(She said, breathlessly into her cell phone...Siri responded "I'm sorry, all I have is disturbing uncanny valley porn. You will need to upgrade to the iPhone xXx for that, now with genital recognition."
Re: (Score:2)
NOOOO!
Hearing Siri do a five minute belly laugh would just be too much......
Re: (Score:3)
Re: (Score:1)
I hope there comes a day when there are no more overpaid actors. Let THOSE jobs go to computers! We aren't there yet but this is a step in the right direction.
Even when computers can generate photo-realistic people that can move like real people you are still going to have actors.
Why?
Because the job of actors is not to simply be "photo-realistic people that can move like real people" (we call those people "models"), their job is to act. They bring life and nuance to the characters they portray. Consider Andy Serkis's Gollum. He is as realistic as he needs to be, but he would not exist as a convincing character without Serkis's motion capture acting.
Computers do n
Re: (Score:2)
A good film is made by throwing away >90% of the recorded footage, that's like practically the heart and soul of procedural generation.
To come back to a point you made that I think undercuts your argument, Andy Serkis's Gollum was an entirely digital character, aside from voice.
Andy Serkis and various animators input digital data to computers to make Gollum
Re:Not Bad (Score:4, Insightful)
Don't be deliberately stupid. The bulk of the profit from movies goes to studio owners and producers. You want them to get even more?
It's like Jim Bouton said of player salaries, "[the players] don't deserve the money, but the owners don't deserve it more."
Re: (Score:2)
Re: (Score:3)
Different problem maybe. Tarkin and Leia are generated from pictures of real people. The lighting and animation of them seems to make them look a little plastic and fake. Tarkin worked best, I think because the actor was playing such a stiff role that he hardly had to move his mouth.
This is more about generating fake people... animating them is still going to be a problem.
Re: (Score:1)
Re: (Score:2)
I like Star Wars, but I am not a nitpicking fan. So when I saw Tarkin and Leia I didn't feel that anything was odd on the first impression. After it was pointed out, then I could see the imperfections. I expect a lot of the fans before they saw the movie realized these characters would be shown in CGI. Me I didn't do all my research about how the movie was made before I watched it, so I wasn't displeased from it.
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Check this out: https://www.youtube.com/watch?... [youtube.com]
Re: (Score:2, Interesting)
I want an algo that I can feed my own pictures to and it will produce a picture that resembles me enough to be recognizable by a human who knows me in person, but won't match my actual face using facial recognition (as in it would subtly change the biometrics of my face like distance between eyes, between mouth and nose, etc). That would make for a good dating profile pic - it looks like me in person so its nobody is surprised if we meet in person, but the dating site can't easily link all my data based on
Re: (Score:2)
Re: (Score:2)
Girlfriend in GANada?
Re: (Score:2)
Ok, that was good. Kudos.
Fake News! (Score:1)
Fake celebrities will add a whole new dimension to fake news.
Re: (Score:2)
I think the entire article is misleading. All of those faces and images are not generated from scratch, but are in fact based on graphical templates that I guess were fed to this neural network. And all that neural network seems to do is to apply some morphs to the existing images by interpolating with other images.
I was sceptical at first, but then the Petronas Towers showed up during the morphs of the "Towers" category. Also there is so much background detail in the images, no way all of that information
Re: (Score:2)
Ah that explains why several of the photos look very similar to existing actors & actresses
Odd Standards (Score:2)
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
Re: (Score:2)
Do they look creepy? (Score:5, Interesting)
Re: (Score:2)
From the static picture they seem as creepy as people with overdone makeup. I was hoping to see a bunch of pictures of the normal people. Not a bunch of models. It is like early 3D static realistic however impossibly clean and perfect.
Re: (Score:2)
I was hoping to see a bunch of pictures of the normal people. Not a bunch of models.
You might want to retract that criticism. The article says they trained the network with celebrity photos, most of which benefit from professional makeup, lighting, and retouching.
If they trained it with regular family photos instead, I assume the output would be more in line with your expectations.
Re: (Score:2)
Perhaps, but with the professionals with a lot of the imperfections glossed away it may mean that regular photos may not work as well, because it would be too complex because of all our imperfection, skin blotches, crooked teeth, and less then symmetric faces. Or the algorithm will have a hard time figuring out what is normal, Thus make a lot of people with exaggerated imperfections, so what would had been a normal looking person, ends up being quite ugly.
Re: (Score:1)
Some of them are very creepy. The most common case is a pair of eyes that are too different. Different sizes, different slanting. Each eye may be perfect alone, but they do not form a pair - yet they are in the same face.
Then there are all the weird eyeglasses, when morphing from one kind to another.
See the movie about the training of the neural nets. Some of the intermediate images are quite yucky - hair looking like a heap of veins or something. More training and it got better.
Now, will they be able to tr
Re: (Score:2)
Re: Do they look creepy? (Score:1)
Curriculum learning (Score:4, Informative)
This sounds like the standard idea of curriculum learning - you teach NNs via progressively more difficult tasks.
Re:Curriculum learning (Score:5, Informative)
It isn't in GANs one model, the generative, tries to fool the other, classifier one, by giving it images itself has generated. There's no incremental task switching from easier problem domains to more difficult ones.
I'm familiar with GANs - what it sounded like (and what they did) is add curriculum learning, but they also did it layerwise as is done with autoencoders, (Also they had some other interesting ideas, but that was the crucial bit). In this case the easy is the lower resolution images and the hard is the higher dimensional images.
From their paper
The idea of growing GANs progressively is related to curriculum GANs (Anonymous), where the idea is to attach multiple discriminators that operate on different spatial resolutions to a single generator, and furthermore adjust the balance between resolutions as a function of training time. That work in turn is motivated by Durugkar et al. (2016) who use one generator and multiple discriminators concurrently, and Ghosh et al. (2017) who do the opposite with multiple generators and one discriminator. In contrast to early work on adaptively growing networks, e.g., growing neural gas (Fritzke, 1995) and neuro evolution of augmenting topologies (Stanley & Miikkulainen, 2002) that grow networks greedily, we simply defer the introduction of pre-configured layers. In that sense our approach resembles layer-wise training of autoencoders (Bengio et al., 2007).
Rule 34 (Score:2)
There can be no exceptions!
Build-to-order (Score:1)
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.
Re: (Score:2)
Really? (Score:5, Funny)
"two neural networks -- a generator and a discriminator"
IOW a democrat and a republican. :-)
Re: (Score:2)
Ok i laughed :)
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
That's excusable, as an European I can't even tell them apart.
In other news... (Score:2)
Hollywood-Powered Neural Networks Produce Freakishly Fake Natural Human Photos
Training database seems skewed (Score:5, Interesting)
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.
Re:Training database seems skewed (Score:5, Informative)
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.
If you look at the full paper, this is capable of so much more than faces. There are dozens of pages of every-day objects they generated, from bedrooms, to wine bottles, to boats, and bicycles. A few of them of some pretty obvious warping and distortions, but the ones that don't look like real objects. It's mind blowing.
Re: (Score:2)
I expect we will start seeing this used for TV shows pretty soon. Sets are expensive, and it's really obvious when shows like Suits re-use the same corner office slightly redressed for 20 different companies. Soon they will just film against bluescreen and click a button in Premier to auto-generate a single-use office set.
Re: (Score:2)
We can cashier pretty much all actors and actresses. Already synthesized voice singer robot is grammy award winner quality. Saw the post about the singing robots of Japan? We can just use CGI actors.
Finally people who were lucky enough to have some good looking features will stop hogging the limelight, yup limelight. The real creative people, the script writers, directors and cinematographers will get their due share. While these CGI characters who work 24/7 for a pittance
Re: (Score:2)
Re: (Score:2)
Yes, they can. Haven't you heard of VR porn and its accessory devices?
The difference here is that no bargaining need be involved. You don't need to say "Blow me and I'll make you a star!" you just say "Blow me, then go away, I'm tired."
Re: (Score:2)
Mindblowingly overfitted, probably.
Re: (Score:2)
Yeah, let's try this with "People of Walmart".
Re: (Score:2)
They do (more or less) but I wonder what would you do with them? I.e. what is the use for a single frame of a non-existing person, or the ability to generate single frames of many non-existing people?
Looks like a remake of that Godley and Creme video (Score:2)
https://www.youtube.com/watch?... [youtube.com]
not so scary now (Score:1)
Re: (Score:2)
This guy has a great future in the horror genre of movies. He just redefined "creepy".
Re: (Score:2)
They forgot to reverse-photoshop them. My guess is they trained the algorithm using various fashion- and lifestyle magazines that consist of nothing but photoshopped pictures of people stripped of any and all personality.
Can the criminal system keep up? (Score:5, Interesting)
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.
Re: (Score:3)
Rarely is photographic evidence alone used for a conviction. You'd be amazed how unreliable cameras, etc. actually are.
However, they are often used as PART OF a conviction. Especially if they have come from multiple independent sources (nearby shops as well as the one burgled, street cams, some random person's dashcam, etc.).
No court would convict on the basis of one photo alone - even if it was dated and had GPS EXIF info. Precisely because it's too easy to forge. That's why some cameras have cryptogra
Re:Can the criminal system keep up? (Score:5, Informative)
Speaking as a lawyer, I'm afraid you have far too much confidence in the judicial system. People have been convicted based on a lot less than a seemingly perfect photograph and few criminal defendants have the financial wherewithal to hire an expert to contest the veracity of a spoofed photo.
Re: (Score:2)
Speaking as a lawyer, I'm afraid you have far too much confidence in the judicial system. People have been convicted based on a lot less than a seemingly perfect photograph and few criminal defendants have the financial wherewithal to hire an expert to contest the veracity of a spoofed photo.
Exactly. The one with the most money wins is often more truth than hyperbole.
Re: (Score:2)
Realistic looking? (Score:1)
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!
Re: (Score:2)
Speech generation. (Score:3)
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.
Re: (Score:2)
Maybe; but maybe enough GFLOPS can manage to do it someday.
If I were Hollywood "actors" (Score:2)
Perfect FB Clickbait (Score:2)
Catfishing (Score:2)
This could take catfishing to a whole new level.
Beware.
LK
Upscaling application? (Score:5, Interesting)
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?
Waifu2x (Score:2)
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/ [waifu2x.udp.jp]
Re: (Score:3)
Here's a demonstration of a neural net doing temporal interpolation (increase framerate of source video by adding realistic intermediate frames)
https://www.youtube.com/watch?... [youtube.com]
Re: (Score:3)
Re: (Score:2)
Wow, I should have known someone would already be on the case, that stuff is impressive. And I like that those scientists actually mention the image enhancement tech from Blade Runner in their paper.
"Believe half of what you see" (Score:2)
Re: (Score:2)
Pretty much. Before, faking images and video was time consuming, but this is the next step in being able to produce convincing fake imagery on a mass scale outside of the realm of Hollywood. I foresee this tech being useful for disinformation experts, being able to construct complete profiles for sock puppet accounts to include family vacation photos, graduations, etc.
We won't be able to believe our lying eyes when it comes to anything on a display. Thankfully, we can always rely on those in a position of
Re: (Score:2)
Even the fake ones won't swipe right for mine (Score:2)
...dammit.
bad hair day (Score:2)
Re: (Score:2)
There's just a huge amount of variation in photos of the top of the head because of hats, hairstyles, etc. so it probably takes much more training to get that part right. The face is much more consistent among beautiful actors.
Re: (Score:2)
Know what the Internet will do with this tech? (Score:2)
If after reading my title you didn't guess: naked people. All those photorealistic painting artists will be out of a job.
I don't get it (Score:2)
Mr. Potato Head (Score:2)
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 Desc
A Scanner Darkly? (Score:1)
Re: (Score:2)