Despite increasing computing power, high quality media still presents a storage problem. If the raw file is compressed to piss poor quality with an algorithm, the same algorithm could be used to faithfully (or with a margin of error outside our sensorial capabilities) live upscale the media file back while you are playing it. So you could have a full 8k video library resting on 720p files, lossless audio coming from 192kbps files, and stuff like that, a quality good enough to embed for piss poor connections and processing power, but that can be locally live upscaled by the algorithm on better devices.
ps: Upscaling media is already real, it’s just bad at the moment. Funny to see people believing it won’t get better. How easily are you still telling AI videos apart from the real thing without the aid of professional tools? Just a year ago it wasn’t so difficult.
If the raw file is compressed to piss poor quality with an algorithm, the same algorithm could be used to faithfully (or with a margin of error outside our sensorial capabilities) live upscale the media file back while you are playing it
No it can’t. Almost all video compression formats are lossy meaning that data is permanently lost. You can’t recover it
Conceptual Reading Recomendation - lossy compression vs lossless compression.
So low quality lossy upscaled to high quality lossy on the go, that any difference would only be found by meticulously inspecting the spectrogram. Even the most snob FLAC enthusiasts on hyper high end hardware already fail A/B tests with 320kbps mp3 anyway
MP3 is very high quality lossy compression because of the psychoacoustic modelling to know what our brains don’t process when hearing, and dropping only that. This includes frequencies outside of our hearing range and effects like masking where loud noises of a certain frequency mask adjacent frequencies. The compression only loses the parts no one hears.
Your post, while not specific, was talking about video. Lossy video compression loses quality that we can perceive. Upscaling can help with some of it, but you can only interpolate so much.
I think there is a fundamental-misunderstanding in that…
You can fake resolution, but you can’t create it, once you’ve thrown it away, with lossy-compression.
The brilliantest example ever, was given like this ( had to do with “universal file compression” scammery, last-century )
Say you’ve got 4 files, that are 2-bits long:
- 11
- 10
- 01
- 00
& you compress them all down to 1-bit, so now you’ve got 2 compressed files:
- 1
- 0
… how could you de-compress them back out to 4 separate files?
The information’s GONE.
You’ve only got 2 left, right?
The same is true for compressing media-files.
I wish all the quote-images were done in SVG, all the audio was in OPUS, & all photos were in JXL, simply because that seems to be the optimal solution.
I wish the best video-format ( that isn’t patent-encumbered ), AV1 or something, were normal, too.
It isn’t that there is some specific video-resolution which is required, for the archived quality, rather, it is that there is some specific information retention that is required, & some of that information may have to do with resolution, other may have to do with color, other may have to do with motion, etc…
so it isn’t specific-resolution, rather, it is specific-percentage-of-information-discarding that is what’s being done…
_ /\ _
If the raw file is compressed to piss poor quality with an algorithm, the same algorithm could be used to faithfully live upscale the media file back while you are playing it.
Not really, compression takes an average over an area of video, you can’t really reconstruct the missing values by reversing the algorithm.
AI upscale is already real, and if compression behaves always the same way it’s easier to recreate what was lost.
Let’s try to explain compression with a mathematical example:
You have the numbers 10, 35, 56 and 59. You form an average out of those values, which is 40. Now you give that average number to me. You tell me it’s made up of 4 numbers and ask me what those numbers were. I can form a guess, based on other examples you send me before and told me afterwards what numbers were in there. But I will at most randomly guess the right values by chance.
Compression is not different from this example, were the mathematical numbers represent colors.
No
Once detail is removed it can never be restored faithfully. Ai can generate details and an individual may not detect much difference but in direct comparison it will be different.
If you somehow include the details as data all combined in a compressed format you save space but the result is not a lower res, but a lossless compressed native res in which case upscaling is irrelevant.
