Slashdot Mirror

← Back to Stories (view on slashdot.org)

NVIDIA-Powered Neural Network Produces Freakishly Natural Fake Human Photos (hothardware.com)

Posted by BeauHD on from the machine-rendered dept.

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.

27 of 140 comments (clear)

  1. Not Bad by mentil · · Score: 4, 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!

    --
    Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
    1. Re:Not Bad by jellomizer · · Score: 4, 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.

      --
      If something is so important that you feel the need to post it on the internet... It probably isn't that important.
    2. Re:Not Bad by Anonymous Coward · · 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 facial recognition.

    3. Re:Not Bad by Hal_Porter · · Score: 2

      Girlfriend in GANada?

      --
      echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
    4. Re:Not Bad by NoZart · · Score: 2

      yeah, not quite there yet - see tarkin and leia in that last star wars

    5. Re:Not Bad by dreamchaser · · Score: 2

      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.

    6. Re:Not Bad by Austerity+Empowers · · Score: 2

      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.

    7. Re:Not Bad by Gilgaron · · Score: 2

      Rachel looked pretty good in BR2049, though... much better than the efforts in Star Wars. The way the scene was lit may have helped, though.

    8. Re:Not Bad by mark-t · · Score: 2

      I disagree.... while yes, you are right that people want to see the human actors that they know and recognize, I can definitely see no small appeal to virtual computer actors, and a demographic of people that they would appeal to that is more than large enough to service.

      --
      File under 'M' for 'Manic ranting'
    9. Re:Not Bad by Anonymous Coward · · 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 mickey mouse) . But without a real star's price tag.

    10. Re:Not Bad by Notabadguy · · Score: 2

      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 movie franchise, why not? This might automate / reduce demand for real people in Hollywood, but I think the real applications are going to be in pornography and video game franchise character development.

      When they can perfect this beyond faces and into contextual surroundings and bodies, along with animation...then you can do all sorts of things that have a following but push legal and ethical grey areas in erotica, create custom scenarios for people, and start exploring virtual reality in a fashion people might care about.

      Makes me want to re-read Killobyte by Piers Anthony. https://en.wikipedia.org/wiki/...

    11. Re:Not Bad by Anonymous Coward · · Score: 2, Informative

      Already happened in Japan - Hatsune Miku virtual singer

      https://www.wmagazine.com/story/hatsune-miku-crypton-future

    12. Re:Not Bad by cellocgw · · Score: 3, 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."

      --
      https://app.box.com/WitthoftResume Code: https://github.com/cellocgw
    13. Re:Not Bad by morethanapapercert · · Score: 2
      Virtual actors might not be all that attractive to the consumers, at least, not as long as they can tell the difference. BUT; I'm sure they'll be hugely appealing to the folks who produce the media content. They'll be a hell of a lot cheaper, easier to work with, scandal free and readily disposable if the director or the fickle public decide a given character is yesterdays news. With that kind of motivation, I'm sure enough money will be thrown at the problem until the studios get characters the audience can't possibly distinguish from the real thing.

      William Gibson touched on this stuff in his novel Idoru. A sophisticated enough synthetic person who is so convincing that a major music star wanted to marry her. I think that it is only a matter of time before we see the re-birth of the old studio system, only with wholly fictional characters. There was a time when almost everything visible about a Hollywood star was the product of cigar smoking men in back rooms, plotting out the lives of the actors to maximise their box office draw.

      --
      I need a wheelchair van for my son. Help me get the word out. https://www.gofundme.com/wheelchair-van-for-jj
  2. Do they look creepy? by Chrisq · · Score: 4, Interesting

    Do they look creepy? They look like many or the retouched "real" photos you see in the media all the time to me!

  3. Curriculum learning by LetterRip · · Score: 3, Informative

    This sounds like the standard idea of curriculum learning - you teach NNs via progressively more difficult tasks.

    1. Re:Curriculum learning by LetterRip · · Score: 4, 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).

  4. Really? by Anonymous Coward · · Score: 5, Funny

    "two neural networks -- a generator and a discriminator"

    IOW a democrat and a republican. :-)

  5. Training database seems skewed by swb · · Score: 4, 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.

    1. Re:Training database seems skewed by BradleyUffner · · 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.

  6. Can the criminal system keep up? by geekmux · · Score: 4, 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.

    1. Re:Can the criminal system keep up? by ledow · · Score: 2

      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 cryptographic signatures that write hashes into the image information, etc.

      Legal tests have been in place for hundreds of years because, applied appropriately (which they aren't always, but that's why you have lawyers and appeal courts), they don't care about the particular technology involved (whether that's digital cameras or some guy with an angle-measurer). What matters is how that correlates to other evidence, how independent it is, how individually reliable it is (that's what you're talking about) and how "reasonable" it is that actually it depicts what the prosecution/defence claim it does.

      There's a reason that defence lawyers are highly paid, and they'd find holes in almost everything like this if it got their client off.

      To be honest, I've provided CCTV footage to police on dozens of occasions over the years as part of my job, and not once has it been the sole evidence, and not once did it actually get used as anything other than "corroborating" a story that could be corroborated any number of other ways too.

    2. Re:Can the criminal system keep up? by nealric · · Score: 4, 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.

  7. Speech generation. by Kaenneth · · Score: 2

    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.

  8. Upscaling application? by Tx · · Score: 4, 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?

    --
    Oh no... it's the future.
    1. Re:Upscaling application? by religionofpeas · · Score: 2

      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?...

    2. Re:Upscaling application? by barbariccow · · Score: 2

      You've obviously never seen CSI. Didn't you know they can take a 12DPI resolution image from a shitty security camera and turn it into 1080P? It actually uses its own assembly language, which they managed to trim down to a single instruction: ENHCE -- Enhance. Instead of typing in that program and using loops, you literally just use voice-to-text, stare at a screen and say "Enhance! Enhance! Enhance!"