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

Extreme video compression with prediction using pre-trainded diffusion models

github.com

Post reply on HN
Print view View original

71–80 of 89 posts

Re: Extreme video compression with prediction using pre-trainded diffusion models

#71

Post by didntcheck »

CamperBob2 wrote: post #30
ToJans wrote: post #19

Ahhh, Sloot's digital coding system [1] is finally here ;). [1] https://en.m.wikipedia.org/wiki/Sloot_Digital_Coding_System

In the [Sloot Digital Coding System], it is claimed that no movies are stored, only basic building blocks of movies, such as colours and sounds. So, when a number is presented to the SDCS, it uses the number to fetch colours and sounds, and constructs a movie out of them. Any movie. No two different movies can have the same number, otherwise they would be the same movie. Every possible movie gets its own unique numbe…

It sounds almost like someone explained content-addressed storage to him and he misunderstood (where you can uniquely identify a movie by number, down to some hopefully a negligible collision likelihood, but you're merely indexing known data)

Re: Extreme video compression with prediction using pre-trainded diffusion models

#72

Post by djmips »

fasa99 wrote: post #59
holoduke wrote: post #54

Back in 2005 there was a collegue at my first job writing video format converters software. He was considered a genius and the stereo type of an introvert software developer. He claimed that one day an entire movie could be compressesed on a single floppydisk. Everybody laughed and thought he was weird. He might be right after all.

I used to work with a guy like that in 1997, during the bubble, Higgins was his name. He'd claim you could fit every movie ever onto a CD-ROM, at least one day in the future it would be possible. Higgins was weird. I can still recall old Higgins getting out every morning and nailing a fresh load of tadpoles to that old board of his. Then he'd spin it round and round, like a wheel of fortune, and no matter where it st…

A deep thought.

Re: Extreme video compression with prediction using pre-trainded diffusion models

#75

Post by HarHarVeryFunny »

holoduke wrote: post #54

Back in 2005 there was a collegue at my first job writing video format converters software. He was considered a genius and the stereo type of an introvert software developer. He claimed that one day an entire movie could be compressesed on a single floppydisk. Everybody laughed and thought he was weird. He might be right after all.

Well, as a reality check, even the soundtrack of a 1hr movie would be 50x floppy size (~50MB vs 1MB) if MP3 compressed.

I guess where this sort of generative video "compression" is headed is that the video would be the prompt, and you'd need a 100GB decoder (model) to render it.

No doubt one could fit a prompt to generate a movie similar to something specific in a floppy size ("dude gets stuck on mars, grows potatoes in his own shit"). However, 1MB is only enough to hold the words of a book, and one could imagine 100's of movie adaptations (i.e. visualizing the "prompt") of any given book that would all be radically different, so it seems a prompt of this size would only be enough to generate one of these "prompt movie adaptations".

Re: Extreme video compression with prediction using pre-trainded diffusion models

#77

Post by david-gpu »

leoff wrote: post #70
Rygian wrote: post #68

Earlier quoted context omitted.

How does that make sense? Compression is deterministic (for same prompt, same output is algorithmically guaranteed). AI is only deterministic in corner cases.

AI is always deterministic. We add noise to the models to get "non-deterministic" results, but if the noise and input is the same, the output is also the same.

It's a bit more nuanced than that. Floating point arithmetic is not associative: "(A+B)+C" is not always equal to "A+(B+C)". Because of that, certain mathematical operations used in neural networks, such as parallel reductions, will yield slightly different results if you run them multiple times with the same arguments.

There are some people working hard to provide the means to perform deterministic AI computations like these, but that will come with some performance losses, so I would guess that most AIs will continue to be (slightly) non-deterministic.

Re: Extreme video compression with prediction using pre-trainded diffusion models

#78

Post by 5kg »

aqme28 wrote: post #21
Animats wrote: post #6

Extreme compression will be when you put in a movie and get a SORA prompt back that regenerates something close enough to the movie.

Where’s that quote? Something like “AI is just compression, and compression is indistinguishable from AI”

Language Modeling Is Compression: https://arxiv.org/abs/2309.10668

Re: Extreme video compression with prediction using pre-trainded diffusion models

#79

Post by kimixa »

visarga wrote: post #56
aqme28 wrote: post #21

Earlier quoted context omitted.

Where’s that quote? Something like “AI is just compression, and compression is indistinguishable from AI”

> AI is just compression, and compression is indistinguishable from AI Almost. Compression and AI both revolve around information processing, but their core objectives diverge. Compression is focused on efficient representation, while AI is built for flexibility and the ability to navigate the unpredictable aspects of real-world data. Compression learns a representation from the same data it encodes, like "testing on…

A lot of current compression techniques use prediction followed by some set of correction data to fix mis-predictions. If the prediction is more accurate, you can have a smaller correction set.

But you're right the predictor does need to be reproducible - the output must be exactly the same to match encoder and decoder behavior. While I don't think this is big focus right now for many, I don't think there's a fundamental reason why it couldn't be, though probably at the cost of some performance.

Re: Extreme video compression with prediction using pre-trainded diffusion models

#80

Post by jazzyjackson »

visarga wrote: post #56
aqme28 wrote: post #21

Earlier quoted context omitted.

Where’s that quote? Something like “AI is just compression, and compression is indistinguishable from AI”

> AI is just compression, and compression is indistinguishable from AI Almost. Compression and AI both revolve around information processing, but their core objectives diverge. Compression is focused on efficient representation, while AI is built for flexibility and the ability to navigate the unpredictable aspects of real-world data. Compression learns a representation from the same data it encodes, like "testing on…

representations' efficiency increases the more flexibility you give it
Post reply on HN