New Super 8 Home Scanner

[Randy J Tomlinson]

nice idea but 3000$ for this unit? not so for me. not unless they have a full HD sensor in it.

[Ernie Zahn]

I'd like extendable arms so the same form factor unit can hold larger spool sizes.

[Anthony Schilling]

nice idea but 3000$ for this unit? not so for me. not unless they have a full HD sensor in it.

They tested a full HD sensor, but with the current sensor you can uprez to full HD. According to the tests, there was no detectable difference of a full HD capture from an uprez. And a full HD sensor would have driven the cost up more.

[Carl Looper]

They tested a full HD sensor, but with the current sensor you can uprez to full HD. According to the tests, there was no detectable difference of a full HD capture from an uprez. And a full HD sensor would have driven the cost up more.

 

You can up-rez data from any sensor. You can up-rez a single pixel to a 1000 K data file if you want.

 

As for undetectable differences, there is no such thing in the first place. Has anyone ever seen an undetectable difference? No. Because if you could see it then it wouldn't be undetectable, would it? Either there is a difference or there isn't. And if there is a difference then it is detectable.

 

C

[Carl Looper]

The only reason for not using a larger sensor would be economics. It makes the unit cheaper to make, and therefore makes it cheaper to buy.

 

The idea of "no detectable difference" is purely promotional. It's like selling a bicycle on the basis that it's wheels are not squares. Or a car as being a "horseless carriage". But even these examples are not quite analogous, because there are such things as square wheels or horses, the absence of which would be a useful feature.

 

But there is no such thing as an undetectable difference. But if you can sell it, I guess why not.

 

C

[Carl Looper]

What I'd like to see manufactured in the digital sensor universe, is an alternative approach to sensor design, one which takes some ideas from film and implements them in a chip. The first thing I'd recommend is making sensor cells of different sizes and in a random configuration rather than a raster layout. This would immediately eliminate moire issues without having to do any filtering. The second rather counter-intuitve idea is to decrease the number of bits per sensor cell - the purpose of which is not to degrade the image but to provide a relaxation of the constraints otherwise limiting the manufacture of smaller area sensor cells.

 

Consider for a moment, a 2 bit sensor cell that was 1 mm x 1 mm square (for the sake of argument). Such a cell would be capable of capturing 4 light levels (0,1,2,3). Now lets replace this with 4 x 1 bit sensor cells, but each of which was 0.5mm by 0.5 mm.

 

For the same area, the amount of light levels captured is slightly increased. Each of the smaller cells can register two light levels (0,1) which gives a total value for the 1mm x 1mm area, between 0 (0,0,0,0) and 4 (1,1,1,1). Not only that but one also increases the frequency response (sharpness).

 

By using cells of varying sizes one can improve the dynamic range. Indeed cameras such as the Alexa, as I understand it, have already caught onto this idea, using two or more different sized cells.

 

Now one of the problems with a random configuration is that it can introduce grain and it would be a fixed pattern. However insofar as the pattern can be calibrated (unlike film) it can also be cancelled out. If the position, size and shape of each sensor cell is known (as it would be or can be) a custom filter can be used to cancel out the grain it would otherwise produce as a side effect. However it could also reduce the definition.

 

A way to eliminate a fixed grain structure (rather than grain itself) would be to change the pattern on each frame. One could have a number of sensors, each of which is used in alternation. However this wouldn't completely solve the problem as the pattern of each sensor would emerge as a statistical observable. A better idea is to have a unique sensor pattern on each frame.

 

This could be done by manufacturing a ribbon of sensors (much like film).

 

One way to create a ribbon of low bit sensors is to use nano-technology. For example one could use light sensitive silver bromide particles, perhaps suspended in cellulose acetate.

 

C

 

 

 

 

C

[Anthony Schilling]

You can find all the bells and whistles mentioned in all the posts above in the next best option, the Kinetta desktop scanner, at a cost of $179,995. Or a high end pro telecine at $450 per 8 rolls of neg that you have not even seen yet, For my dollar, I could not ask for anything better than the Retro 8 scans burned to bluray played on a flat screen. The alternatives just don't make any sense.

[Carl Looper]

You can find all the bells and whistles mentioned in all the posts above in the next best option, the Kinetta desktop scanner, at a cost of $179,995. Or a high end pro telecine at $450 per 8 rolls of neg that you have not even seen yet, For my dollar, I could not ask for anything better than the Retro 8 scans burned to bluray played on a flat screen. The alternatives just don't make any sense.

 

Indeed. The Retro8 is extremely cool. It is really well made and really affordable. And produces excellent results (from what I've seen so far). Roger Evans has been doing great work for many years. And he's continued to improve on what he's previously done. He's a pragmatist. He's helped me to see things in a more practical light.

 

But that still doesn't mean the concept of "undetectable differences" makes any sense at all.

 

C

[Nicholas Kovats]

Once again, Carl, you hit it out of the park!

 

"One way to create a ribbon of low bit sensors is to use nano-technology. For example one could use light sensitive silver bromide particles, perhaps suspended in cellulose acetate."

[Anthony Schilling]

 

Indeed. The Retro8 is extremely cool. It is really well made and really affordable. And produces excellent results (from what I've seen so far). Roger Evans has been doing great work for many years. And he's continued to improve on what he's previously done. He's a pragmatist. He's helped me to see things in a more practical light.

 

But that still doesn't mean the concept of "undetectable differences" makes any sense at all.

 

C

It means if your watching a clip scanned at 720P and upscaled to 1080P, next to the same clip scanned at 1080P, you wouldn't be able to tell the difference between the two.

[Carl Looper]

It means if your watching a clip scanned at 720P and upscaled to 1080P, next to the same clip scanned at 1080P, you wouldn't be able to tell the difference between the two.

 

Yes I know what it means, but that still doesn't mean it makes any sense.

 

If there is a difference it will be detectable. Simply because, there is no such thing as an undetectable difference. The qualification of "undetectable" is a way of saying the difference doesn't matter. It's a way of saying that there is actually a difference, but if you put a blindfold on, you won't be able to see it.

 

However, detection usually involves removing one's blindfold rather than putting one on.

 

Now certainly one can fail to find a difference, but on what basis then, can one say there is a difference in the first place? Either one finds a difference (ie. it was detectable) or one doesn't find it, ie. one doesn't know if there is a difference or not. To say an "undetectable difference" is a way of hedging one's bets. It's a way of saying either you did not find a difference or if you did, you decided it was inconsequential. The latter is more likely the case.

 

Either way it becomes incorrect to say an "undetectable difference". Either a difference was detected or the question of whether there is difference is unknown.

 

In any case, there is a difference and it is detectable. The bigger the difference in sensors the bigger will be the difference. The difference between 720HD and 1080Hd is certainly a small difference, but it's not an undetectable difference.

 

C

Edited August 24, 2013 by Carl Looper

[Carl Looper]

To be a little clearer, it's like saying because there is little difference (or "no detectable difference") between an image that is 640 pixels wide and one that was 641 pixels wide, that therefore anything bigger than 640 pixels is superfluous.

 

That is the logic behind the concept of "undetectable difference". It's a way of being happy with 640 pixels.

 

But one can be happy with 640 pixels anyway. One doesn't need the concept of "undetectable difference".

 

Carl

[David Cunningham]

Anyone who has really worked with Super 8 (especially fine grained ones like E100D or Vision3 50D) will tell you that there is a HUGE difference between scanning at 720p and 1080p. I'm quite certain I could "tell the difference". I'm sure once the video is compressed for the Internet, the difference is much less "detectable". But, experimenting quite a bit with E100D and Vision3 50D has proven to me that Super 8 even benefits from a 2K scan. This is not because there is actually more picture detail on the film than 720p. This is because it takes at least 1080p if not 2K to fully resolve the grain. The esthetic is VERY obvious when you compared 10-bit uncompressed versions of SD, HD and 2K.

 

So, I'm sure the difference would be quite detectable. Test charts might not show it and and grain-reduced images might not show it. But, if you are talking about E100D or even Velvia 50D and compare it to the projected image... it's not even close.

[Anthony Schilling]

The native scan is 720P but you have the option to output the file at 1080P. An up-rezed 1080P was compared to a native 1080P scan and was indistinguishable, so the added costs of building the unit with a native 1080P sensor seemed unnecessary. These are not my words or decision, just the info from the maker. Personally, I have yet to process a capture at 1080P, i'm happy with the resolve at 720P and smaller file size. But technically, i agree that in order to get the full resolve of a fine grain S8 stock, technically speaking 2K is better, and 4K for larger formats. The real question is how far do you need to go to justify added costs that outway any noticable benifit? Unless your shooting for the megaplex, it's not much of an issue,

[Carl Looper]

The native scan is 720P but you have the option to output the file at 1080P. An up-rezed 1080P was compared to a native 1080P scan and was indistinguishable, so the added costs of building the unit with a native 1080P sensor seemed unnecessary. These are not my words or decision, just the info from the maker. Personally, I have yet to process a capture at 1080P, i'm happy with the resolve at 720P and smaller file size. But technically, i agree that in order to get the full resolve of a fine grain S8 stock, technically speaking 2K is better, and 4K for larger formats. The real question is how far do you need to go to justify added costs that outway any noticable benifit? Unless your shooting for the megaplex, it's not much of an issue,

 

Yes. The real question is the cost/quality trade off. But that doesn't mean differences in quality are not real questions as well.

 

Roger Evans says there is no difference, (or no detectable difference) but despite what Roger says there is a difference and it is detectable.

 

The difference may not be worth the additional cost associated with an HD sensor but that's an entirely different argument.

 

Roger is just being practical. As you say the real question is whether the additional cost of HD is worth the increase in quality you can obtain. Roger thinks not and I would probably agree with him. Indeed one could argue what Roger is saying is just a manner of speaking: it is easier and more comprehendable to say "no detectable difference" than to elaborate price/cost/quality tradeoffs.

 

But that is where the discussion should really take place. There is a law of diminishing returns operating here. The cost of increasing sensor size (and associated data/control systems) increases at a rate that is much faster than the increase in quality. It becomes increasingly cost in-effective the further you push the quality envelope.

 

However you can, in principle, build your own DIY scanner, and do so cost effectively. And it's in this context I'm arguing against the misdirection that "undetectable differences" can suggest. Whether you'll be able to do better than Roger's excellent machine is a different story.

 

C

Edited August 25, 2013 by Carl Looper

[Carl Looper]

Anyone who has really worked with Super 8 (especially fine grained ones like E100D or Vision3 50D) will tell you that there is a HUGE difference between scanning at 720p and 1080p. I'm quite certain I could "tell the difference". I'm sure once the video is compressed for the Internet, the difference is much less "detectable". But, experimenting quite a bit with E100D and Vision3 50D has proven to me that Super 8 even benefits from a 2K scan. This is not because there is actually more picture detail on the film than 720p. This is because it takes at least 1080p if not 2K to fully resolve the grain. The esthetic is VERY obvious when you compared 10-bit uncompressed versions of SD, HD and 2K.

 

So, I'm sure the difference would be quite detectable. Test charts might not show it and and grain-reduced images might not show it. But, if you are talking about E100D or even Velvia 50D and compare it to the projected image... it's not even close.

 

Indeed. But it's not just resolving the grain. The grain is just a side effect of a much more important property - the way in which film encodes an actual signal. There is more actual signal in a higher definition scan than there is in a lower definition scan. The signal is not necessarily higher resolution, in the traditional digital sense of that term. It might be better to call it "dynamic resolution" or "dynamic definition" because it evades close analysis. A higher definition scan transfers more dynamic definition than does a lower definition scan. And this dynamic definition is visible whether the work is blown up large on a big screen or indeed reduced to a 720 pixel wide image. Strangely enough. This is because it is not localised in the same way that conventional resolution might be visible (using MTF). But it is visible. It is very visible.

 

C

[Carl Looper]

We can calculate how many K would be required to resolve the smallest silver bromide particles in black and white Super8 film and use this as a rough limit. After processing the silver bromide actually dis-integrates into even smaller particles (of silver) but they aggregate around the site of the silver bromide particle so we can use the size of the bromide as a limit.

 

Bromide particles range in size (as a ballpark figure) from 2um down to 0.2um

 

So to resolve the area of the smallest particle (0.2um) would require, for Super8, a scan that was

 

5.69 mm / 0.2 um = 28,450 pixels

4.22 mm / 0.2 um = 21,100 pixels

 

A 28K scan!

 

Arguments against this have always been in terms of either resolution, ie. that beyond a certain fuzzy threshold there is no more resolution available - that scanning at such a mega-definition would be superfluous - or that scanning at such a mega-definition would only reveal noise.

 

However this assume that it's only resolution one gets from a high resolution scan. It also assumes that the apparent noise visible at close proximity has no signal value. Both assumptions are incorrect. The apparent noise only lacks value when considered in terms of a small sample area, but statistically, ie. at much larger scales it has a huge amount of signal value. But a lot of that value is lost when performing a large area sample on it. It might cancel out the apparent noise but it also cancels out the very fine variation in that noise. It is the variation which holds a signal.

 

Statistics is a very important tool when assessing something like film - much more so than when assessing a digital image. This is because the relationship between a signal and the random size and shape of every single particle, is encoded in the film. This encoding is statistical rather than mathematical. And our eye/brains are completely tuned to seeing statistical patterns. This statistical relationship is completely visible.

 

C

[Roger Evans]

I think this is not as complicated as some might make it seem. In my business, even though we have sold hundreds of Sniper-HD units over the years, the telecine shops buying them tell me that they mostly still sell regular DVDs to their customers and not BluRays. So, to be clear, there is always a difference between scanning at one resolution and the next just as there is always a difference between a picture shot with your iPhone 5 and a Nikon DSLR. But, relative to intended viewing size, not the mention the sophistication of the viewer, the end result may be a distinction without a difference. The Retro-8 does a good job at 720 and satisfies the market share that would never have a display big enough - nor the sophistication - to appreciate the difference between a 720 or 1080 scan. Why drive up the cost of the unit to accomodate the exception and make it unaffordable for everyone just to have technical bragging rights at 1080? And, realistically, anyone that has the budget to buy a Spirit isn't going to say, "Gee, I think I'll save myself a quarter million dollars and buy the Retro-8, instead." The Retro-8 has an intended market, which is for private use or small telecine shops that just need a clean, easy, dependable way to transfer to DVD or BluRay. In that sense, I think it's a good value. But, depending on your needs, your mileage may vary. :)

[Robert Houllahan]

I think this is an interesting time in the world...;-)

 

Spirit Telecine suites are very far from a quarter million dollars these days, and that is as is should be. I think a SDC-2000 (Spirit-1) is worth about $20k right now, probably $25k with a DaVinci 2K plus the electricity to run it, which is allot let me tell you...

 

I think the little Retro Scan is an interesting device, and it's great to see that there is no projector or claw involved, maybe Cinelab should buy one and make it available as a system to be used by people who want to come in and DIY as someone suggested. 720P is not my favorite resolution but running it through a scaler like a Teranex can make it 1080P without loss or appreciable difference from the original 720P. That said I think it is clear that a Super-8mm frame needs an oversampled 2K scan to "get everything" including resolving the grain structure. I am pretty sure if you side by side a 3K+ Super-8 scan with the 720P or the uprezzed 720P the 3K scan will look considerably better. This may be a point which is meaningless to someone's 98 year old grandma who is watching a DVD or Blu-Ray on a $199 LCD set of her wedding in 1942.

 

Digital scanning will become less expensive and proliferate (it is) as time goes by for the same reason we have DSLR video and RED cameras, Moore's law goes everywhere.

 

"Cinelab sales pitch on"

 

We have evaluated a number of sensors for our upcoming Xena servo scanner and just purchased a Imperx Bobcat 3340 Camera-Link sensor camera for it, this will make a 3.4K color mask scan at 17FPS at 12bits. This machine is doing machine vision registration of multiple perfs for stabilization. (Color Bayer Cameras typically can resolve 2/3 of total resolution, so a 3K Bayer color camera like an Alexa makes nice 2k pics, compared to a "True" film scan which is full raster monochrome for each channel like our Pin Registered Xena at 4K for each RGB) also Resolution is only a small part of the equation there is Dynamic range, Noise, Tap Balancing, Blooming performance and the quality of illumination. If anyone is interested we have a couple of 2K (2048x2048 color) Kodak CCD sensor based cameras for sale in the classifieds for anyone interested for a DIY system.

 

"Cinelab sales pitch off" ;-)

 

I think driving the cost of scanning down and more scanning houses with varying degrees of quality and price will help to keep using film relevant, and I think the lab and Kodak also are changing and finding ways to survive in this new world. We have already tried to keep our scanning minimums to under $150 for either data or HD and I think that will be part of how things are going forward.

 

-Rob-

[Roger Evans]

 

"Spirit Telecine suites are very far from a quarter million dollars these days, and that is as is should be. I think a SDC-2000 (Spirit-1) is worth about $20k right now, probably $25k with a DaVinci 2K................"

 

I agree with much of what you are saying but I think you will also agree there is more to setting up a high end telecine suite than just buying a used Spirit for $20-30,000. Otherwise it's like saying that you now have a darkroom because you bought a used Leitz enlarger for $25 at a pawn shop. There are so many other things to consider in the way of peripherals such as various recording devices and control interfaces, film cleaning and storage, space, trained labor and maintenance contracts for all associated equipment, not to mention the sometimes hard to find 8mm gates, that I would be surprised if somone could set up a viable commercial venture for under a $200,000 investment. And even at half that price, no one with that kind of financial resources would be swayed by the low price tag of the Retro-8. It was never intended as a replacement for a high end telecine suite so any favorable comparisons between the two are the result of diligence and creativity on the part of the user as much as anything the Retro-8 brings to the table.

 

On a related note, can a clever person build something that will do 2K as a DIY project cheaply? Certainly. Can they offer it as a commercial product with all the maintained monthly overhead such as advertising, production and warranty for the same per unit price as a DIY project? Not a chance. So discussions that scoff at the $2495 price of a Retro-8 because it doesn't do 2K reflects little serious research on what commercial film scanners really cost to produce, operate and maintain.

 

Roger

[Carl Looper]

On a related note, can a clever person build something that will do 2K as a DIY project cheaply? Certainly. Can they offer it as a commercial product with all the maintained monthly overhead such as advertising, production and warranty for the same per unit price as a DIY project? Not a chance. So discussions that scoff at the $2495 price of a Retro-8 because it doesn't do 2K reflects little serious research on what commercial film scanners really cost to produce, operate and maintain.

 

 

I've put together a DIY 5K scanner for under $1500 for doing my own work. It's not a way of making money as such. It's a way of getting my film work scanned at a definition that I'm after. I'd prefer even more definition but that's what I've achieved so far. However I've made it modular so can upgrade aspects as I go along.

 

A DIY product isn't typically a way of making products that one sells as such. Its usually a solution to the problem of not having much of a budget to buy something already made. One makes it instead. And in my case it was actually worse than that. The thing I needed to have made didn't actually exist. So I had to make it exist.

 

DIY projects tend to be one off projects. A bit like making a film. The film is a one off. You might go onto make other films but not the same film.

 

To frame DIY projects as if they were meant to have the same priorities and concerns as the factory line is a bit of a misunderstanding of what motivates most DIY projects. Often it's just making something for the hell of it. Something to do.

 

C

[Will Montgomery]

That said I think it is clear that a Super-8mm frame needs an oversampled 2K scan to "get everything" including resolving the grain structure. I am pretty sure if you side by side a 3K+ Super-8 scan with the 720P or the uprezzed 720P the 3K scan will look considerably better. This may be a point which is meaningless to someone's 98 year old grandma who is watching a DVD or Blu-Ray on a $199 LCD set of her wedding in 1942.

Another real factor is how crappy most Super 8 is actually shot; extremely poor focus being the main issue. I'm sure the film itself would benefit from oversampling but only if it's actually in focus when shot. I see plenty of Super 8 negative shot recently that looks like 16mm from a few years ago when scanned on a Spirit or Shadow. For me it comes down to about 80% the colorist; as long as it's an HD scan and not heavily compressed, I'd pay more money for a good colorist vs. the best machine in the world.

[Carl Looper]

Another real factor is how crappy most Super 8 is actually shot; extremely poor focus being the main issue. I'm sure the film itself would benefit from oversampling but only if it's actually in focus when shot. I see plenty of Super 8 negative shot recently that looks like 16mm from a few years ago when scanned on a Spirit or Shadow. For me it comes down to about 80% the colorist; as long as it's an HD scan and not heavily compressed, I'd pay more money for a good colorist vs. the best machine in the world.

 

Even "crappy" Super8 will look better when scanned at higher definitions. Even out of focus Super8. Can look gorgeous.

 

A higher definition scan can't bring back into focus something shot out of focus (at least not in terms of scan alone). And it doesn't need to be considered as if it would do. A higher def scan simply ensures more of the original image is transferred rather than less. If the original is out of focus, so what? Should a few out of focus shots require all the shots be therefore transferred at less quality?

 

The argument always seems to be that the worse an original film is in terms of certain criteria the less information you should therefore put into a transfer.

 

It's like saying the longer you leave some food out on the table (instead of in the fridge) the longer you should therefore leave it out on the table.

 

Or the more times you put a photocopy back through a photocopy machine the more times you should therefore put it back through a photocopy machine.

 

Super8 needs better transfers (in terms of definition and dynamic range) precisely because it starts with a signal that is weaker in the first place (weaker than larger formats). To put it another way it needs all the help it can get. Economically, of course, this doesn't make any sense. Many shooting Super8 do so due to budget constraints, so they are hardly likely to spend a mini-fortune on scans.

 

But economic sense is purely contextual. It depends entirely on the circumstances.

 

The theory side can be discussed independantly of economic circumstances. Theories on how to get to Mars and back depend on economics but economics doesn't determine whether the theory will be correct. The economics is a constraint. For example, economics might dictate how you build a spacecraft but it can't dictate things like the law of gravity. Or rather, if it did, the resulting design of your spacecraft might produce some unfortunate consequences.

 

C

[Carl Looper]

But none of what I'm saying is meant in any way whatsoever as a criticism of Roger Evans spaceship. On the contrary, it is a well built spaceship that takes into full and complete account, both the laws of physics AND economics.

 

C

Edited September 5, 2013 by Carl Looper

[Roger Evans]

But none of what I'm saying is meant in any way whatsoever as a criticism of Roger Evans spaceship. On the contrary, it is a well built spaceship that takes into full and complete account, both the laws of physics AND economics.

 

C

Oh I'm not that thin skinned. I understood your analogy. Not insulted at all. :)