When is AI going to make everything else obsolete?

[plehoediv]

Hi,

I just started to "import" all my old VHS tapes to a little box, and a Blaupunkt RTV 936 VCR, and so far, according to everyone over in another thread of mine, the footage sucks, ha ha. I think it is OK, but pretty much everyone else says it sucks. I am not disagreeing, but I am disappointed, because I thought I had some pretty "good" hardware. . Turns out, not so much, ha ha. That said. I see people talk about using software to make the footage look better, so my question is:

When is AI going to be so good that nothing else matters!? You can throw some pretty bad footage at it, and the AI software will turn it in the crisp 4k footage. 5 years? 10 years? Or maybe only in 2? What do you think?

Thanks.

P.S. In my case, I have time. I am showing my footage, the way it is, to family and friends and they are losing their minds. After that, I can wait another 10 years and make it look perfect then. : )

[Selur]

I doubt that this will be available (or possible) in the next 10 years.
(I also think what you aim for would be more like software splitting the source video into video objects, generating 3d representations of those and trying to recreate a scene with upscaled and refined 3d representations of those models and objects. So no restoration, with these models&co you could create whatever scene you wanted.)


Attached an example where I upscale suzie 176x144 (suzie_lossless) to 1920x1436 using BasicVSR++ two times (suzie_1920x1436), which looks nice in my book, but is far from perfect and nowhere near 4k.
(used script: https://pastebin.com/NfmamfyJ; encoding took ~2min for 150frames)

Cu Selur

[plehoediv]

I doubt that this will be available (or possible) in the next 10 years.

I don't know man, 10 years is a long time when it comes to this stuff. I am hoping for the best.

(I also think what you aim for would be more like software splitting the source video into video objects, generating 3d representations of those and trying to recreate a scene with upscaled and refined 3d representations of those models and objects. So no restoration, with these models&co you could create whatever scene you wanted.)

If understood correctly what you are saying, No, I was not looking for what you described... I guess.

Attached an example where I upscale suzie 176x144 (suzie_lossless) to 1920x1436 using BasicVSR++ two times (suzie_1920x1436), which looks nice in my book, but is far from perfect and nowhere near 4k.
(used script: https://pastebin.com/NfmamfyJ; encoding took ~2min for 150frames)

I think that looks really nice. But not for me. I am not into coping and all that stuff. Sure, if I can copy and paste like I do sometimes with ffmpeg, I do it, but anything more than that needs to be automated, or I am not touching it, ha ha.

Thanks.

[Selur]

10 years is a long time ...

but "When is AI going to be so good that nothing else matters!? " is a really high bar.

I am hoping for the best.

me too

[plehoediv]

10 years is a long time ...

but "When is AI going to be so good that nothing else matters!? " is a really high bar.

I am hoping for the best.

me too

Well, I should have worded it a little bit different, I guess. What I meant with "nothing else matters" is instead of having to use really high end vcr's, capture cards and what not to create a good copy from your old VHS tapes, you could just take the "crappy" footage I have now by not using super high-end hardware and have the AI do the rest. Does that make more sense?

Thanks.

P.S. I knew something wasn't "right" when I got a reply at this time a day... In the USA it is in the middle of the might, but not here in Germany...

[lollo]

What I meant with "nothing else matters" is instead of having to use really high end vcr's, capture cards and what not to create a good copy from your old VHS tapes, you could just take the "crappy" footage I have now by not using super high-end hardware and have the AI do the rest. Does that make more sense?

Not really. Starting from the best possible input is always the optimal option. The same (tuned) AI algorithm applied to a better source will produce better results and less artifacts.

[lordsmurf]

I just started to "import" all my old VHS tapes to a little box,
the footage sucks, ha ha.

No amount of AI can ever overcome the massive amount of image damage done by that ridiculously overpriced low-quality ClearClick (aka ClickCrap, a deserved nickname).

When is AI going to be so good that nothing else matters!?

Probably never. Too many variables, and algorithms truly suck at determining visual issues.

P.S. In my case, I have time. I am showing my footage, the way it is, to family and friends and they are losing their minds.

Remember, many people are just too nice. If it looks like crap, they won't tell you.

[DVDLover00]

well, AI is a good helper, when it comes to render/restore images !
I dont have an good pc to run AI with programs, but maybe in near future, with better system, possible.

[johnmeyer]

Artificial Intelligence is the most over-hyped concept in the history of computer software. The term has been around for decades. Remember the Spielberg movie of the same name? That was almost a quarter century ago.

The short answer is that there is no horse someone is going to ride into town which will fix poorly captured video. As others have suggested, you should absolutely re-capture your tapes while you still have them, and this time use better hardware.

The is -- and there never will be -- any substitute for a proper capture.

[Cornucopia]

IMO, belief in the universal panacea of AI falls into the category of "magical thinking", which should be avoided at all cost.

Scott

[TreeTops]

Scott and John Meyer. Both very smart persons. I studied AI 60 years ago and learned then that it would be a very long time for AI perfection. Something approaching a human brain may someday succeed.

[Bwaak]

I am showing my footage, the way it is, to family and friends and they are losing their minds.

I could give you half a dozen links to YouTube videos right now that are stretched, have horrible combing, blown out highlights, not even mentioning broken motion, yet they have hundreds of thousands of views and thousands of likes.

https://www.youtube.com/watch?v=oYMszuMyGG4 - 1.1M views, 20K likes, 1.7K commendatory comments. Watch and weep.

Visit any sports bar and observe TVs showing stretched or squished picture, or showing an SD channel when HD is available, and no one cares.

[plehoediv]

Not really. Starting from the best possible input is always the optimal option.

I agree. However, that would be like saying that if you are not born rich (Starting from the best possible input is always the optimal option), you will never be rich or as rich.

The same (tuned) AI algorithm applied to a better source will produce better results and less artifacts.

That is true. However, that is like saying that 8 billion is "better" than 6 billion, but I am sure everyone would agree, that it does not really matter, or does it!? I know I am a bit of a d*ck here, but I am just trying to make a point.

Having said that. I am 100% sure that in 10 years all this "fuzzing" around will no longer be necessary. You perhaps might be even able to "feed" the software photos of whoever is in the video footage and the AI (or whatever you would like to call it) will "reconstruct" it in a way that the output will be ridiculously good! Just like it is not with Deepfakes.

But hey, this is just my opinion.

Thanks.

[plehoediv]

No amount of AI can ever overcome the massive amount of image damage done by that ridiculously overpriced low-quality ClearClick (aka ClickCrap, a deserved nickname).

I believe it will be so powerful that you can just give it still images of yourself and it will "create" a brand-new video of whatever you want.

Probably never. Too many variables, and algorithms truly suck at determining visual issues.

Maybe. Still, I believe it will create its own ways of dealing with the issue of whatever it encounters. That is the whole point of it being "AI".

Remember, many people are just too nice. If it looks like crap, they won't tell you.

I can see that. However, it depends what culture you are from, and where I am from, they will ask your wife if she gained some weight, ha ha.

Thanks.

[plehoediv]

well, AI is a good helper, when it comes to render/restore images !
I dont have an good pc to run AI with programs, but maybe in near future, with better system, possible.

I agree.

[plehoediv]

Artificial Intelligence is the most over-hyped concept in the history of computer software.

Maybe. Probably.

The term has been around for decades. Remember the Spielberg movie of the same name? That was almost a quarter century ago.

And it sucked. from what I remember.

The short answer is that there is no horse someone is going to ride into town which will fix poorly captured video.

I hope you are wrong.

As others have suggested, you should absolutely re-capture your tapes while you still have them, and this time use better hardware.

I should and I will! I already gave of the three VCR's I just bought, to my parents. Next I will get rid of the other two and then get one with TBC. After that, not sure yet. Still trying to figure out what route to take.

The is -- and there never will be -- any substitute for a proper capture.

Confidence works both ways. I am as convinced that there will be, as you are that there will not be.

Thanks.

[plehoediv]

IMO, belief in the universal panacea of AI falls into the category of "magical thinking", which should be avoided at all cost.

Scott

Thanks for your input.

[plehoediv]

Scott and John Meyer. Both very smart persons. I studied AI 60 years ago and learned then that it would be a very long time for AI perfection. Something approaching a human brain may someday succeed.

Interesting. Thanks.

[El Heggunte]

Scott and John Meyer. Both very smart persons.
I studied AI 60 years ago and learned then that it would be a very long time for AI perfection.
Something approaching a human brain may someday succeed.

Not even an octopus brain in the foreseeable future.
And "artificial intelligence" is a huge misnomer, to begin with.

[lollo]

but I am just trying to make a point.

The point is that doing a very good / proper capture is not that difficult nor that expensive. You have been given indications in the other thread

[plehoediv]

I am showing my footage, the way it is, to family and friends and they are losing their minds.

I could give you half a dozen links to YouTube videos right now that are stretched, have horrible combing, blown out highlights, not even mentioning broken motion, yet they have hundreds of thousands of views and thousands of likes.

https://www.youtube.com/watch?v=oYMszuMyGG4 - 1.1M views, 20K likes, 1.7K commendatory comments. Watch and weep.

Visit any sports bar and observe TVs showing stretched or squished picture, or showing an SD channel when HD is available, and no one cares.

I agree. However, the fact that people are like that is a totally different story. Actually, a sad one, if you ask me.

Thanks.

[plehoediv]

but I am just trying to make a point.

The point is that doing a very good / proper capture is not that difficult nor that expensive. You have been given indications in the other thread

I am not sure why some, including you, act as if me asking means I am not going to try to do "very good / proper capture". I will. I was just wondering what people think about AI and stuff like that. Thats all...

Thanks.

[lollo]

I am not sure why some, including you, act as if me asking means I am not going to try to do "very good / proper capture". I will.

It was not an attack from my side, just trying to explain what can be done at capture level, before going into software level. Both will help to achieve good results.

[plehoediv]

I am not sure why some, including you, act as if me asking means I am not going to try to do "very good / proper capture". I will.

It was not an attack from my side, just trying to explain what can be done at capture level, before going into software level. Both will help to achieve good results.

It's all good and I appreciate your input. . Like i said, I was just curious to see what people think about the whole AI thing, and now I know.

That said, I just figured out how to copy from VHs to DVD, on this combo machine I borrowed from a friend. Cannot wait to see how the footage will look like compared to my sh*t box video "grabber". I have to go back and check what someone wrote because I do not remember if it is supposed to be better or worse, than using my grabber box.

Thanks.

[johnmeyer]

Confidence works both ways. I am as convinced that there will be, as you are that there will not be.

The reason I am confident that AI will not soon produce anything that will help improve video in ways that are remarkably better than what can be done today is the same reason I am confident that nuclear fusion will not be providing a huge amount of power within the next 100 years. My statement is not based on belief: anyone can believe anything they want, but that belief won't sway anyone nor will it make something magically happen. Instead my statement is based on learning curve theory. With nuclear fusion, Lawrence Livermore Labs (and others) have been trying to create a net positive energy result from fusion for almost forty years. They finally achieved that goal for a small fraction of one second, but there is little in that experiment that would suggest they will be able to sustain that result for even a second, much less continuously. Therefore, given that it took forty years to get here, and given that there has been no new branch to the effort (like "cold fusion"), it will be at least 100 years before anything practical gets built.

Same for AI. The concepts have been applied to multiple video problems and so far, the results have not broken any new ground or provided results that wouldn't have happened anyway. More to the point, the AI efforts to improve video have not yielded anything of note, while other more traditional software approaches have given us stabilization, dirt removal, denoising, sharpening, and a host of other technologies that can make fantastic improvements and are used everyday by lots of people.

If AI video enhancement were imminent (about to happen) we'd be seeing precursors and real-world results.

[_Al_]

In this regard, a breakthrough is inevitable, pretty soon, computing power is a limit. But it is going to be reconstructed. Like imagine hair. It needs to be reconstructed. Video will be enhanced, beautiful, except not really. We have a saying, if there is nothing, not even devil can take. Details are going to be made up. It is pretty safe guess to think like that, but folks do not realize that. They almost think that detail is going to be back on. To reconstruct a shirt with a pattern, perhaps very close to original, but eyes, skin, details, grass, trees, other random patterns etc... will be made up, to sort of match what was there, but made up. Because we are talking video, not images. When something is flickering (grass) or skin changes while something is moving enhanced image must be continuous, and because discrepancies cannot be smoothed out across frames, it will be reconstructed.

Look at pictures that are being upscaled, enhanced on web, like Photoshop and other even more sophisticated things. I saw one video where software made small woman person instead of of small kid. They fix these obvious things, but really, who is to tell AI while enhancing a face in a distance how old that person is? Those are subtle details for us humans to know, it is called culture.

[DVDLover00]

I am showing my footage, the way it is, to family and friends and they are losing their minds.

I could give you half a dozen links to YouTube videos right now that are stretched, have horrible combing, blown out highlights, not even mentioning broken motion, yet they have hundreds of thousands of views and thousands of likes.

https://www.youtube.com/watch?v=oYMszuMyGG4 - 1.1M views, 20K likes, 1.7K commendatory comments. Watch and weep.

Visit any sports bar and observe TVs showing stretched or squished picture, or showing an SD channel when HD is available, and no one cares.

happens a lot, really. I cant think people watching those crappy tv's, is an pain on their eyes
https://www.youtube.com/watch?v=oYMszuMyGG4 anyway, Really nice video !!

[lingyi]

GIGO (Garbage In Garbage Out), Can't make a silk purse out of a sow's ear

[zing269]

GIGO (Garbage In Garbage Out), Can't make a silk purse out of a sow's ear

Been able to do that for over 100 years. Of course, just like making gold out of lead, it isn't cost effective.
"On the Making of Silk Purses from Sows' Ears," 1921

[rrats]

No Ai tinkering can fix the losses of detail present in the original transfer to the pc.