Live data from Hacker News

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

github.com

111–120 of 129 posts

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

#111
post #109
post #103

Earlier quoted context omitted.

I suspect that a combination of the two is where it's at, ie, store a lossy classical compressed version, then remove the artifacts/dream up details with deep learning

or pretrain the network on the movie you want to compress, then ship cartoon compressed version + trained network.

Right that was exactly what i was thinking

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

#113
post #102

Kind of surprised this stuff hasn't totally blown away traditional video compression techniques yet.

This got me thinking: in future we will probably stop streaming video and instead just send simple vectors. The data then gets rendered in real time into actual video. The customers will even be able to customize the movie: pick their favorite actors, the environment, etc.

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

#114

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…

We reached that point of not being able to trust any media a while ago. It's just getting easier and more widely distributed.

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

#115
post #75

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…

AI can also be used to identify these fake videos, and they are probably better than human identification. There will be a rise of AI forensics. The type of neural network they use (GAN) works by having two networks battling each other, one tries to generate fake, and another (discriminator) tries to identify fake, it's a constant arms race. As the generator gets better, the discriminator also gets better. Which mean…

The whole GAN thing is useless though if neither of the models have a concept of symmetry or that those pixels do represent actual solid objects in space. Modern neural nets can't represent abstract spatial and semantical concepts or reason, so that videos are full of glaring perspective inconsistencies and large-scale artifacts. So nope, neither AI can help to "identify fake videos", nor do we need AI for that.

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

#116

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.

Someone's gonna come up with a simple way to sign pics and videos with public key encryption, after which signed media will be the only stuff that is trusted. Of course people will still make sex tapes and sign them when they are drunk because the signing software will be automatically integrated into the camera app, so blackmail is still a possibility.

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

#117

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.

Someone's gonna come up with a simple way to sign pics and videos with public key encryption, after which signed media will be the only stuff that is trusted. Of course people will still make sex tapes and sign them when they are drunk because the signing software will be automatically integrated into the camera app, so blackmail is still a possibility.

But.. if the camera app automatically signs pics/vids, that would mean that a private key is available to the app without any passphrase (or one embedded in the app :O).. So why not just extract they key and sign your fake vid?

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

#118

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…

Probably Nvidia is not interesting in this type of tools and this is more like "hey, look what you can do with a lot of Nvidia cards, buy us a lot of them and you will do this and much more"

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

#119
post #117

Earlier quoted context omitted.

Someone's gonna come up with a simple way to sign pics and videos with public key encryption, after which signed media will be the only stuff that is trusted. Of course people will still make sex tapes and sign them when they are drunk because the signing software will be automatically integrated into the camera app, so blackmail is still a possibility.

But.. if the camera app automatically signs pics/vids, that would mean that a private key is available to the app without any passphrase (or one embedded in the app :O).. So why not just extract they key and sign your fake vid?

Well, for something like this the user's signing key itself would probably be managed by the OS, so not extractable. When the app is done editing, it asks the user via the OS to approve the result and the OS performs the signature. You could even embed a downsample of the original that came signed from the camera hardware.

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

#120
post #75

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…

AI can also be used to identify these fake videos, and they are probably better than human identification. There will be a rise of AI forensics. The type of neural network they use (GAN) works by having two networks battling each other, one tries to generate fake, and another (discriminator) tries to identify fake, it's a constant arms race. As the generator gets better, the discriminator also gets better. Which mean…

The problem is that it's a game of superhuman cat and mouse.

We'll have systems arguing with each other and no way to tell which is correct. If, for example, someone is able to get a copy of the forensic AI model, they can train their decoder/generator to work against it until the results pass as legitimate. With no human ability to argue with the results of the forensics AI, we'll just trust it and pass it off as truth.

If you can clone the jury and conduct a quadrillion private trials, your chance of success in court is going to increase substantially.

Things gonna get creepy.

Post reply on HN