Is an HD Super 8 transfer overkill?

[Uppsala BildTeknik]

Well if you don´t believe me, believe Roger. Or better yet, try it yourself. Get a SD transfer from whatever place you want and then compare to your downscaled HD transfer.

There really is no point in discussing this further. You are just being silly. Downscale to 1 pixel or whatever, it will still not look like a SD transfer.

The big difference here is that I am talking about comparing different transfers, and you are just talking about mathematical calculations, and about should, would or could.

Instead of just getting a SD transfer and using that SD transfer as your SD reference point. :roll:

[carllooper]

Well if you don´t believe me, believe Roger. Or better yet, try it yourself. Get a SD transfer from whatever place you want and then compare to your downscaled HD transfer.

I don't have to believe anyone - including Roger. I already know the answer. It's not an issue for me. I've got my own transfer rig and can scan the film at any sampling rate, from 0.00001K up to 3K.

So I don't need to believe in what someone says - I can do it myself - as I've done a number of times - and see it with my own eyes.

And I can say that there is no difference in spatial resolution between sampling the 3K copy at some specific lower rate, and sampling the original at that same specific lower rate.

There are very good reasons why this is so which have nothing to do with beliefs and everything to do with basic science, information theory and seeing what is so with your own eyes.

The only reason I posted resampled versions of my 3K scan rather than resampled versions of the original, is precisely because I already knew there was no difference - and assumed everone else understood this. Why go to the trouble of re-lacing the machine and doing yet another test - when the results of previous tests say that it is unnecessary to do so.

The only reason for doing so might be to help someone else see what you mean.

But if that someone else has already decided, because of some belief in what someone has said, that whatever you say, or show, or prove is going to be incorrect then yes, I guess this conversation must be over.

On the other hand, perhaps that someone else could have some humility and post their SD and HD chart. What's the worse that could happen?

I could be proved wrong.

I'm very happy to be proved wrong.

But they might be scared to do that - that it might be they who are proved incorrect. Perhaps it undermines their "authority" on the subject.

Carl

[MovieStuff]

I could be proved wrong.

I'm very happy to be proved wrong.

No disrespect intended, Carl, but science doesn't work that way. You have to prove that you're right in a "real world" setting. You seem to be struggling to formulate a mathmatical response to something that is more perceptive in nature and not quantifiable since it has to be experienced and not predicted. Hard to do on the internet but I appreciate your intent.

I'll resample your HD frame to resemble the SD frame and then you can tell me what the difference is.

Let's try to clarify something here:

For the whole HD>SD>HD versus SD>HD experiment as you describe, you seem to presume that the same HD scanner will be used to produce the original SD transfer so that there is a common starting point or a bench mark for comparison.

In theory, this makes sense.

However, in the real world, the cost of doing an SD transfer on an HD scanner is going to be the same as an HD scan. So the only real reason to originally transfer in SD -at all- is if you can save money by doing it on an older, dedicated SD scanner that might cost less and not a newer HD scanner that might costs more.

Will the quality of a dedicated SD scanner be the same as an SD transfer on a HD scanner? My guess is "no" but your mileage may vary.

Still, if it is your position that an SD transfer on a dedicated SD telecine upscaled to HD and viewed on a large HD monitor will look the same as an original HD transfer viewed on a large HD monitor, then the only valid way to prove that is to make an original SD transfer on a dedicated SD telecine and then upscale it to HD and watch it on a large HD monitor in the real world. Sorry, but anything short of that is just a guess and still frames on the internet isn't the same thing as watching full motion video on a large HD monitor where the overlapping of random grain patterns can be experienced.

Conversely, taking an HD clip produced on a new scanner, down-converting to SD and the up-converting back to HD proves nothing other than how terrific the HD scanner is and leaves out the variables inherent in the original SD transfer process which, in the real world, is hardly likely to happen on that same HD scanner unless one simply likes paying HD prices for an SD transfer!

And, finally, if it is your position that an SD transfer on an HD telecine upscaled to HD and viewed on a large HD monitor will look the same as an original HD transfer viewed on a large HD monitor then I have to respectfully ask "what's the point?" You're already paying HD rates so why limit yourself to an SD product?

In the end, if you can get an HD transfer on a Spirit for the price of an SD transfer on an old Rank, then the question about "overkill" is kind of moot. Go with the HD transfer. It will always produce the most dependable result, regardless of how it is viewed. But if the SD transfer is less expensive than the HD transfer, there's probably a reason. ;)

Roger

[carllooper]

Still, if it is your position that an SD transfer on a dedicated SD telecine upscaled to HD and viewed on a large HD monitor will look the same as an original HD transfer viewed on a large HD monitor

Good grief.

Where did I argue that an SD scan upscaled to HD would look the same as an HD scan?

Nowhere. Please read my posts. They are very exact and precisely expressed in both mathematical and emperical terms.

What I said is, and expressed in yet another way, is that both of the following pipleines produce the same result:

Film > SD
Film > HD > SD

But if it will make everyone happy then I'll use the former approach to argue the exact same thing as I've already been arguing - because it won't make any difference whatsoever.

Carl

[Uppsala BildTeknik]

Roger, this is actually almost the other way around... He believes that a HD transfer is better. Great. But he goes so far as to say that super8 has image details way up to 3k in resolution...

By upscaling, downscaling and whatnot. (Instead of getting native-transfers to whatever resolution you are about to compare.)

That would be like several times the resolution of 1080p. And in my ears (and eyes) this is just crazy. And one of the arguements is that he somehow created a mathematical difference-image that shows in a very weird image that "yes, here are lots of image details on a 3k scan that are not visible on a lower resolution scan".

3k resolution from super8 optics...

[carllooper]

That would be like several times the resolution of 1080p. And in my ears (and eyes) this is just crazy. And one of the arguements is that he somehow created a mathematical difference-image that shows in a very weird image that "yes, here are lots of image details on a 3k scan that are not visible on a lower resolution scan".

Thanks Uppsala - yes that is exactly what I'm saying. I think we're in agreement there. Although what I showed wasn't the 3K scan as such, but 2K and 1K versions of such.

I know it's crazy - but there it is. I just subtracted the 1K version ("SD") from the 2K version (HD) and there it is - the teapot still visible - albeit bordering on being lost.

But that's the thing. It is only at that point where the original signal (the teapot) vanishes that one can say with any confidence that increasing the scan resolution any further won't do anything more than sharpen the noise.

At the disappearance point is that point where one has found a reasonably objective threshold beyond which all else can be called "overkill".

But if you can still see something resembling the original signal (the teapot) then you are still seeing something of the original signal. The original signal hasn't yet disappeared into noise.

If it looks like the teapot then it is the teapot. it is not an hallucination. It is not your memory palying tricks on you. It is what remains of the original signal holding on to the very last vestages of it's existance in the data.

Carl

[S8 Booster]

Seems like there is going on a bypass complemental non-compliant discussion?

Carl wants to prove that a HD sampled image holds on better when doing scaling doing the procedure to prove so while Kent - from a transfer-guy point of view says: whatta fuck do you do that for? Do HD or do SD an shutta fuck up.

I will not mix into that discussion but in my view more is better - in terms of resolution - always.

But to throw in an analog easily provable for anyone with normal hearing - in the digital audio world: early in the fantastic 90 i got this fab AV mac fully video NLE edit capable with a reasonaby good sound system. There was a knife little software which could sample standard 16 bit 44.1 kHz CDs at 32 kHz, 44,1 kHz and 48 kHz switching quickly between the 3.
32 was a step way below 44. 48 was almost that much better.

I guarantee that everyone would agree on this. The difference was biggest with classical music.

These days i enjoy my 8 DAC 1 bit 368 times oversampling CD based system with the shortest and most direct signal path produced in any system. It turns ordinary 16 bit CDs even those i thought were shitty productions into mighty masterpieces. This system plays the fuck out of anything i hve ever heard regardless of price.

Recently i have downloaded some 192 kHz 24 bit totally linear original music pieces from HD tracks. I have played those directly on the HI-Fi system as well as burned them to CDs via AIFF transformatory.
As it came out it is difficult to tell the versions from the other. To compsre them makes no sense.

Now, the all new remasred 192 kHz 24 bit versions sounds much better than many original older CDs but that - in my view goes down to production fiddeling rather than limitation of the Compact Disc medium.

Shoot....

[MovieStuff]

Still, if it is your position that an SD transfer on a dedicated SD telecine upscaled to HD and viewed on a large HD monitor will look the same as an original HD transfer viewed on a large HD monitor

Good grief.

Where did I argue that an SD scan upscaled to HD would look the same as an HD scan?

Nowhere. Please read my posts. They are very exact and precisely expressed in both mathematical and emperical terms.

If I misunderstood what you meant, my apologies. No offense but your posts are not as clear and concise as you seem to think. I do agree with Kent that 2k - 3k resolution off of super 8 is a bit optimistic. That said, I'll let you guys duke it out by yourselves. ;)

Roger

[carllooper]

I should point out, before anyone draws the wrong conclusion, that the difference signal represents the resolution of the film. It does NOT tell us anything (yet) about the resolution of the lens.

As pointed out very early in my posts, the resolution of the signal is a function of the original signal (the one that enters the lens), the lens and the film.

The lens acts as a low pass filter, ie. removing high frequencys from the original signal. Prior research suggests the limit is around 400 px/mm (a 2K scan). But it would be good to find out where a lens limit might really be - whether lower or higher - if only in a particular experimental case study.

Now if we want to use a scan of film to locate this limit we need to know where the film itself has it's limit. The film needs to be at least capable of resolving detail up to where we think the limit might be.

That is the purpose of the 3K scan, and the 2K - 1K difference signal, to show that the film is, at least, capable of discerning the lens signal for all frequencys up to 2K/5.69mm.

Whether the lens signal itself has those frequencys is the second part of the overkill question. And this has got lost (or conflated) in arguments over the difference between resampling film (film transfers) and resampling digital images (so called "rescaling").

Now contrary to how some have interpreted what I'm suggesting, I personally expect the system overkill threshold to be found somewhere between a 0K scan and a 2K scan.

As per original proposition.

And if not, then I too will be surprised.

Carl Looper

[carllooper]

If I misunderstood what you meant, my apologies. No offense but your posts are not as clear and concise as you seem to think.

Yeah - there were a number of things being discussed simultaneously. Easy to get confused.

I was arguing that there was a difference between a 2K scan and a 1K scan, and showing an exact image of it, and showing it was a real signal - not just noise.

And if there was this difference, then it implied that Super8 should be scanned at somewhere between 1K and 2K, if not more.

But I haven't explained why this conclusion can be incorrect.

The difference signal between two scanning resolutions is a simple form of high pass filtering. But a high pass filter applied to a clean digital signal differs very much from a high pass filter applied to a film signal such as Super8.

I've put the following paper up to explain it (I hope).

http://members.iinet.net.au/~carllooper ... lution.pdf

In the paper I identify that, in fact, despite the caution regarding what conclusions you should draw from a high pass filter, that there still appear to be high frequency components of the signal (rather than the noise) emerging in the range between a 1K scan and a 2K scan. However where in that range I can't place it ... and it may still be an illusion ...

I need to shoot an MTF chart - like Kent did - rather than an ugly teapot.

Carl

[Uppsala BildTeknik]

Oh, the teapot is all nice. I was talking about your "difference-image" as the "ugly teapot". Because... well it is quite ugly to look at all day (the difference-image, not your original-image).

8)

[mattias]

No disrespect intended, Carl, but science doesn't work that way. You have to prove that you're right

not really. in science the common procedure is to have a theory and as long as it's not proven wrong, which you should of course constantly try to do yourself, you stick to it. it's very hard to scientifically prove that you're right, so if that's what you want you have to prove that the alternatives are wrong instead.

[MovieStuff]

No disrespect intended, Carl, but science doesn't work that way. You have to prove that you're right

not really.

Yes really.

in science the common procedure is to have a theory

But you only fall back on a theory if what you believe is impossible or impractical to prove or disprove. While some believed the earth was flat, many people had theories that the earth was round long before anyone had data to support it. Upscaling an SD transfer to HD isn't quite as formidable as sailing ships around the world to prove that it's round. ;)

... and as long as it's not proven wrong, which you should of course constantly try to do yourself, you stick to it.

Semantics. You prove it wrong only by trying to prove that it's right.

Look, if you awoke tied to a chair with a blindfold over your eyes with no way to remove the blindfold, you would be limited to theorizing whether the room lights were on or off. But if your hands are free to remove the blindfold, then continuing to theorize is rather pointless. More to the point, if your hands are not tied and you keep insisting to others that the lights are on, then it is incumbent upon you to prove your theory since there is nothing stopping you. But to insist that others prove your theory wrong is just lazy since there is no need to fall back on theory to begin with.

it's very hard to scientifically prove that you're right,

Sometimes, but not always and certainly not in this particular case.

so if that's what you want you have to prove that the alternatives are wrong instead.

Again, semantics. You prove one and you disprove the other or vice versa. But the bottom line is that if a theory can be easily tested, then there is no point in continuing to theorize.

Roger

[carllooper]

My challenge to Kent was in relation to whether there was any difference between resampling an HD scan at SD definition, vs resampling the original film at the same SD definition.

My position was that there was no difference. Kent's position was that there was.

So when I challenged Kent to prove me wrong, I was simutaneously challenging Kent to prove himself right. Neither of us offered any evidence of our respective positions in terms of much explanation, or experimental results.

Anyway, I'll be following Kent's approach this week (when my camera comes back from New Zealand) to do some rigorous experimental testing with an MTF chart.

Carl

[Uppsala BildTeknik]

Anyway, I'll be following Kent's approach this week (when my camera comes back from New Zealand) to do some rigorous experimental testing with an MTF chart.

Make sure you send your super 8 test film to someone who transfers film to SD, for your SD-version.