Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

github.com

Post reply on HN
Print view View original

31–40 of 129 posts

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#31

Post by flurdy »

satori99 wrote: post #18
iforgotpassword wrote: post #7

I'm probably being captain obvious here, but if this is what's being released for free, I wonder how much better a polished commercial version does, and when we reach the point where we can't trust anything we see anymore. It doesn't even have to be super perfect, even reaching the point where it takes experts about two weeks to determine if something's real or not might already be long enough to do great damage. Fro…

It could also lead to crimes like Blackmail becoming extinct. It would be hard to hold incriminating recordings of anyone over them if near-perfect audio and video synthesis was common. Especially for public figures with lots training data available.

Maybe but only after extensive proliferation.

Initially, every Tom, Dick & Harry script kiddie will try to blackmail people whilst the public and justice system is not really aware of the technology possible.

When eventually enough of the public is made aware to not trust photos and videos anymore then these types of blackmail attempts would be less effective. But there will still be less informed gullible targets.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#32

Post by pbhjpbhj »

flurdy wrote: post #31
satori99 wrote: post #18

Earlier quoted context omitted.

It could also lead to crimes like Blackmail becoming extinct. It would be hard to hold incriminating recordings of anyone over them if near-perfect audio and video synthesis was common. Especially for public figures with lots training data available.

Maybe but only after extensive proliferation. Initially, every Tom, Dick & Harry script kiddie will try to blackmail people whilst the public and justice system is not really aware of the technology possible. When eventually enough of the public is made aware to not trust photos and videos anymore then these types of blackmail attempts would be less effective. But there will still be less informed gullible targets.

Surely lawyers will know about the tech soon enough to simply show the same video evidence with the judge in place of the defendant -- reasonable doubt right there.

The blackmail angle I expect will work for a while. People falsely accused of sexual crimes can have their lives ruined by the accusation.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#33

Post by pbhjpbhj »

TomMarius wrote: post #9
iforgotpassword wrote: post #7

I'm probably being captain obvious here, but if this is what's being released for free, I wonder how much better a polished commercial version does, and when we reach the point where we can't trust anything we see anymore. It doesn't even have to be super perfect, even reaching the point where it takes experts about two weeks to determine if something's real or not might already be long enough to do great damage. Fro…

I'm sure the point is already crossed. People on 4chan are making fake porn that is nearly indistinguishable - and they are complete amateurs.

Are you sure they're amateurs, not professionals don't it for fun (/pro boner/!?!).

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#34

Post by mcintyre1994 »

SeriousM wrote: post #21

I'm scared. I can't trust anything I haven't took myself. The problem is that other people not even know that this technology exists and if you tell them about it, you're a lier.

I don't think there's going to be a long period where this technology is being used and isn't widely known about. It'll be used extremely quickly to abuse people using their Facebook pictures, and once there's a Facebook angle the media will be able to run with it and everyone will get it.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#35

Post by albertgoeswoof »

jjeaff wrote: post #29
iforgotpassword wrote: post #19

Earlier quoted context omitted.

On average I agree they are on par. Sometimes OSS is better (ffmpeg), sometimes commercial stuff (adobe after effects). In this specific case I think it might be beneficial if you can spend a lot of money on gathering training material and tweaking the network. But then again someone here mentioned the quality 4chans fake porn has reached, so maybe I'm wrong after all.

This might actually be one of those edge cases where commercial is better, but the government's secret version is by far the best.

Like what though? Historically what tech does did governments have that was unknown to commercial domains?

People are motivated by money, governments pay little and there’s no way to get rich working there so it doesn’t make that much sense.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#37

Post by pbhjpbhj »

pjeide wrote: post #23

What media would someone collect now to be used in the future to reproduce the likeness of loved ones? Video clips of them moving? Talking? Different poses of pictures? Reading the dictionary out loud to get vocal patterns? Heck with impersonating the POTUS. What about a lost friend, sibling or parent?

Black Mirror did this I think - if we can make video, then why not VR (down the line if processing catches up), Second Life iterated.

Heaven on Earth?

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#38

Post by homarp »

sjwright wrote: post #25
have_faith wrote: post #15

Earlier quoted context omitted.

Adobe have definitely bested most oss competitors in their space. Although with the amount of man power at their disposal it would be hard to beat.

The vast bulk of Adobe's advantage is in UX, not technical algorithms. Which makes perfect sense because that tends to be the case with most F/OSS software—technically brilliant but with an face only a programmer could love. Yes, Adobe do have some remarkable algorithms that would be difficult to replicate (e.g. heal brush and content aware fill) but these are a small minority of Adobe's software advantage. The one t…

for content aware fill and heal brush both GMIC ( https://patdavid.net/2014/02/getting-around-in-gimp-gmic-inp...) and resynthetizer ( http://www.logarithmic.net/pfh/resynthesizer ) are working quite OK.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#39

Post by Retric »

albertgoeswoof wrote: post #35
jjeaff wrote: post #29

Earlier quoted context omitted.

This might actually be one of those edge cases where commercial is better, but the government's secret version is by far the best.

Like what though? Historically what tech does did governments have that was unknown to commercial domains? People are motivated by money, governments pay little and there’s no way to get rich working there so it doesn’t make that much sense.

Around WWII Computers, Radar, Nuclear power, crypto, jet aircraft, rockets.

Around 1960 space flight, supersonic flight.

More recently hypersonic flight, and we don’t really know becase people back in WWII had no idea.

Re: Nvidia Vid2vid: High-resolution photorealistic video-to-video translation

#40

Post by magnat »

Findeton wrote: post #22

Some applications for this kind of tech: - Porn, yeah, first application you can think of, there are already some startups doing it. - Doubling actors, and applied to sound, maybe you could translate from one language to another but kind of keeping the accent and tone. - Propaganda and misinformation. Now you can get your enemy to say and do whatever you want, on video. - Photo-realistic games. Create a rough 3D mode…

> Photo-realistic games. Create a rough 3D model of an scenario and train the AI for it. Instead of photo-realistic rendering with math, render it with the AI based on a rough render, in real-time.

According to last month's nvidia rtx presentation/launch event [1], they are going to do something similar quite soon. Games will ship with DNN pre-trained offline on extremely high quality renderings. Game itself renders at lower resolution (limited by performance needed for proper raytracing) and uses DNN to upscale it.

[1] https://youtu.be/Mrixi27G9yM?t=51m15s

Post reply on HN