Anamorphic and the Digital world

[Landon D. Parks]

I have a question.... Could you use a Anamorphic lenses with a camera like Arri D-20 or the Genasis since they have a super 35mm sized CMOS? It seems like if this was possible, you could gain a much higher resolution in theater release than just croping the middle 2.39 section from the Super 35mm frame.

 

Thanks,

[David Mullen ASC]

The problem is that anamorphic lenses have a 2X squeeze. The chip in the Genesis is 16x9, plus the image is recorded to 16x9 HD, so the problem is the same with using anamorphic lenses on a Super-16 camera or a 3-perf Super-35 camera -- the resulting image is too widescreen (3.56 : 1 when unsqueezed) and you end up cropping just as much horizontally to get to 2.39 as you would shooting spherically and cropping vertically to get 2.39.

 

I think the Arri D20 has a 4x3 sensor, so in theory, you would get more resolution by using 2X anamorphic lenses to get a 2.39 image, just as with 35mm, assuming the recording format can use all the 4x3 pixels.

 

The Dalsa's sensor is nearly 2x1, so it makes more sense to just crop to achieve 2.39.

 

What someone really needs to start doing is building sets of high-quality anamorphic primes and zooms with a 1.33X squeeze.

 

But in terms of comparison to cropping Super-35 to 2.39, that's still more resolution than the Genesis or Arri D20; only the Dalsa comes close to that level of resolution. Remember that the Genesis records a 1920 x 1080 pixel image to HDCAM-SR even though it has a 12MP chip.

Edited November 6, 2004 by David Mullen

[Landon D. Parks]

What people really need to be inventing is recording devices capable of recording 12mp uncompressed. Now THAT would be a major improvement in the HD arena.

 

It seems like it would be possible to invent a recorder like this, even though it would eat up A LOT of hard drive space.

[David Mullen ASC]

Well, as the guy from Dalsa mentioned here, the Genesis may not be capable of producing an acceptable 12MP image since it was designed to be recorded to HDCAM-SR (take anything a rival manufacturer says with a grain of salt). Also, remember that on all these single chip cameras (Genesis, Dalsa, D20) all three colors have to be extracted from one chip, so you aren't really getting full resolution in all colors (some of them have an excess of green pixels to boost resolution in that color, sort of a physical manifestation of 4:2:2 recording solutions.)

 

They use either Bayer filters or RGB striping. It's not exactly like they are getting one-third of the resolution of the chip once separated into RBG channels, but I suspect that the 8MP image (4000 x 2000 pixels) off of the Dalsa (which is kept 8MP when recorded) is not really "4K" but practically more like "3K" in effective resolution.

 

Film, on the other hand, is made up of layers for each color info (the Foveon CMOS chip has been trying to make this approach work) and can be scanned at high resolution levels (4K, let's say) for each color separately.

 

There is an invention that can record a 12MP image at speeds up to thousands of times a second... it's called 35mm film.

Edited November 6, 2004 by David Mullen

[Landon D. Parks]

There is an invention that can record a 12MP image at speeds up to thousands of times a second... it's called 35mm film.

I understand that. I was refering to digital though.

Edited November 6, 2004 by Landon D. Parks

[David Mullen ASC]

I know you know that. I just get amused at all the problem-solving and new technology that people are trying to create simply to achieve digitally what we can already get on film... that's sort of the problem with those that want to proclaim film's death too early. You first have to achieve parity, you then have to surpass film, then it has to be cheaper and the more efficient way to do things compared to film. So getting the technology to match film's quality is only the first step.

[Phil Rhodes]

Hi,

 

One thing that has begun to incense me more and more as I work daily with 2K and occasionally 4K film scans is that the resolution the digital stuff is being asked to match is unreasonable. Grain, I can assure you, is fully visible buzzing away in a 2K scan and is absent in video material of almost equal resolution, even compressed HDCAM. If pixel artifacts were as visible on a 2K filmout as grain is on the scan, people would be screaming about it. DI technology is being made to look bad because of the faults of the capture medium, and film is being made to look bad by the interaction of its own faults with DI technology. It is my contention based on this that a true 2K 4:4:4 chain, whether you want to capture that as Bayer 4K and downscale or whatever, is probably all the world needs so long as it's shot, processed and projected as a pixel-for-pixel transfer.

 

I would not argue that while we're doing this unholy gestalt film-digital-film process, 4K is probably necessary to make it look like an optical print - but I think in many cases you're just capturing grain more sharply.

 

Phil

[David Mullen ASC]

But if you don't capture the grain sharply (and I'm not talking about sharpening the grain artificially) then the image will end up softer-looking, although also less grainy.

 

You see this in 2K digital intermediates already. I think working with 35mm at lower resolutions just to reduce graininess is probably not the best solution. Truth is, people have been using 2K as a crutch to avoid worrying about graininess when shooting. With an all-4K system for D.I.'s, we can go back to trying to get it right on the negative rather than use down-rezzing as a method of de-graining (surely there must be better ways.)

 

That Millimeter magazine article on 2K versus 4K several months ago showed fairly clearly with frame enlargements that a 4K scan was better at resolving not just grain, but actual detail, whereas a 2K scan was sort of losing fine detail into blockiness.

 

Film has grain; it's part of its image characteristics, so removing it too well artificially is removing information. Phil, it's not necessarily a BAD thing if a D.I. is able to transfer the grain structure of the original negative over to a new negative, intact, without adding to it. To me, that would be an accomplishment, to end up with duplicate negatives that match the quality of the original. So if the original has grain, then the duplicate has it too. That would be an improvement over conventional duplicating through an IP/IN, especially with an optical printer, because then you aren't simply duplicating the grain of the original, you are adding to it and augmenting it. But the point shouldn't be to achieve a final image that is less grainy that the original negative (except for fixing exposure mistakes, let's say) because THAT'S when you get into trouble with generating artifacts.

 

As for the fact that people are more comfortable with film artifacts (like grain) than digital ones, what did you expect? We've been watching film grain for over 100 years.

[David Mullen ASC]

As for whether digital cameras need to achieve more than 2K resolution to reach parity with 35mm, I think they do if my experience shooting HD is any guide. Many of us have noticed in the theaters that 2K digital intermediates look great, detail-wise, in midshot and close-ups but sort of crap out a little in the wide shots on the big screen. Even straight 35mm in general is barely adequate to put a sharp image into a big screen so dropping down to 2K is definitely a problem. I think people would be complaining more if it weren't for the fact that most movies are shot tighter than they used to be, or that many people see movies in smaller screens.

 

Now if we had a 4K digital camera, we would have an image that was very close to 35mm resolution BUT be less grainy; therefore while there might not be any more detail in the image, it may survive being enlarged more (like blow-ups to IMAX.) This is why I think the goal for digital cameras has to be 4K, although I'd settle for 3K.

 

In theory, with 4K cameras being projected digitally at 4K, we could achieve better resolution than 35mm print projection and get something closer to the experience of watching 70mm prints like we used to have. So we have an opportunity to improve things if we don't settle for 2K throughout the chain.

[Landon D. Parks]

As for the fact that people are more comfortable with film artifacts (like grain) than digital ones, what did you expect? We've been watching film grain for over 100 years.

Thats true. If digital was what we had for the 100 years, and then film came a knocking then everyone would be talking about how terrible film looks. Even though a Digital Image is usually cleaner (as fare as grain anyway) than a film frame, that don't make it better.

 

We have just grown up with the fact that film is the standard, and that's how film should look.

 

Sure digital can be a lot less grainy, but that don't always make it better. I think grain adds to the image quality, as long as theres not to much of it. :)

[Landon D. Parks]

Now if we had a 4K digital camera

The Dalsa is 4k camera is'nt it? Although I dont think it records all of that detail. But then again, I don't know. I have not checked on the Dalsa site for a while to see if they have made any improvments to the Camera and its recorder.

Edited November 7, 2004 by Landon D. Parks

[Filip Plesha]

Bayern sensors create images with interpolation. If a digital camera or a one-CCD video camera has for example 6Mp sensors and records 3000x2000 (6Mp) images on a tape or disk, this means that it uses interpolation. That camera never really can record 3000x2000 separate values of light. It has 3000x2000 sensors for all 3 color chanels TOGETHER, not 3000x2000 for each RGB chanel. The number of sensors are usually in 4:2:2 ratio green is usually the one that gets most pixels.

But not even the green chanel records 6mp of data. It's all interpolated.

 

So 3000:2000 image made from a 3000x2000 sensor probably has half of the

actual resolution. There are a lot of very sophisticated interpolation techniques, so that the result can even look very close to the "real thing"

 

From what I have read about genesis, it does not use a bayern pattern to make images, but uses full resolution for each color chanel. So if you wan't a HD output, it is not interpolated or anything. It's pure 1920x1080 pixels of EACH chanel. It's the real thing. So, for an HD image to be that pure you need a 6Mp sensor at least.

Genesis uses 12Mp, and if all 12Mp are actually used in the image capture, it could fill up a 2.5K image of pure data without any interpolation.

 

But it can not make a 4K image of pure RGB data. For a 4K image, 36Mp would be neded in order to have 3 chanels of color information in full resolution.

If genesis were using a bayern pattern sensor and interpolation mechanism, it would be able to make good 4K images.

 

Dalsa uses bayern pattern or any equivalent of it and makes interpolated 4K images, of course not "real" 4K images that you get in scans, but interpolated. (8Mp sensor)

 

Arri uses allso Bayern patter, but it makes a compromise. It uses bayern technique, but kind of overdoes it, so it ends up a better HD image than usual.

[David Mullen ASC]

The Dalsa does have a 4K sensor and can record all of that data onto a recorder without dropping to HD like the Genesis does. However, being a single chip camera with a Bayer filter, all three colors aren't 4K individually but have to share space on the 4K sensor, so practically speaking, I'd say it would feel more like a 3K scan off a 35mm frame, resolution-wise, even though the final signal, after being converted into separate RGB channels, may (through some uprezzing) be 4000 x 2000 pixels for each color.

 

But whether it is really a 2K, 3K, or 4K image, it is definitely higher in resolution than HD.

 

There is also the experimental Olympus Super-HD camera, which uses four HD CCD's (1920 x 1080 pixels each, or close to that), with two being for the green record, and one of the green CCD's is a half-pixel offset so that you practically double the resolution in the green channel to 3840 pixels across.

[Sam Wells]

"That Millimeter magazine article on 2K versus 4K several months ago showed fairly clearly with frame enlargements that a 4K scan was better at resolving not just grain, but actual detail, whereas a 2K scan was sort of losing fine detail into blockines"

 

There's a very interesting 2K / 4K comparison on Cintel's web site @

 

http://www.cintel.co.uk/technology/4k.htm

 

-Sam

[Landon D. Parks]

There is also the experimental Olympus Super-HD camera, which uses four HD CCD's (1920 x 1080 pixels each, or close to that), with two being for the green record, and one of the green CCD's is a half-pixel offset so that you practically double the resolution in the green channel to 3840 pixels across.

Wow, that sounds pretty cool. Do you know of any links to more information about it?

[David Mullen ASC]

http://www.olympus.co.jp/en/news/2002a/nr0....cfm?ote=0&nr=1

 

I saw a demo of footage shot with this camera at 30P (it doesn't do 24P yet) transferred to IMAX, frame-by-frame, so at 24 fps, it was slightly slo-mo. It was compared to footage shot on the F900 HDCAM transferred to IMAX as well. I saw it at the IMAX theater at the Bridge Cinema.

 

Just like you'd expect, the Olympus footage looked twice as sharp and detailed compared to the F900, sort of like 35mm compared to 16mm. Still had some video-ish look to the contrast and highlights, etc.

 

It's still a prototype though.

[John Sprung]

The Dalsa does have a 4K sensor and can record all of that data onto a recorder without dropping to HD like the Genesis does.

<{POST_SNAPBACK}>

If I'm correctly informed, Genesis does let you send the full datastream down a fiber optic cable. The problem is that, like Dalsa, they don't have anything particularly practical to put on the other end of that cable. Shlepping a rack of disk arrays the size of a fridge around on location isn't going to fly. But at least for TV shows, recording to HDCam SR 4:4:4 would be more than adequate.

 

 

 

-- J.S.

[Landon D. Parks]

Shlepping a rack of disk arrays the size of a fridge around on location isn't going to fly

What is it thats keeping recorder's from recording full Uncompressed Signal from, say the, Genesis?

 

does the technology currently exist to record all that information, and its just a matter of how many Disk's is would take to get that recorded... Or is it a matter that you can't record that amount of Data do to limitations in Disk Speeds, Computer Hardware, ect?

 

Because if it IS possible, then I think it would be worthy to build a recorder the size of a fridge just to test how the FULL uncompressed data looks compared to 35mm as fare as resolution and contrast, ect. So that we would know how the actual full signal looks compared to film.

 

As of right now, even though we have these cameras with 12mp CMOS's and stuff, Nothing "exist's" to record it. Thats seems like a waste to me.

[Phil Rhodes]

Hi,

 

> What is it thats keeping recorder's from recording full Uncompressed Signal

> from, say the, Genesis?

 

Practicality. I spent some of today working on a Baselight 8 colour corrector, a device designed for realtime 4K grading. The disk arrays occupy 40U of 19" rack space and with the processing nodes are about three feet deep. When industry produces multiterabyte drives and a way to put 1.2Gb of data onto them every second in a single unit, I imagine it'll shrink a bit!

 

> does the technology currently exist to record all that information, and its just a

> matter of how many Disk's is would take to get that recorded... Or is it a matter

> that you can't record that amount of Data do to limitations in Disk Speeds,

> Computer Hardware, ect?

 

You can do more or less anything you like, parallelism being the order of the day - it's just a case of what you want, against what you are willing to pay for it.

 

> Because if it IS possible, then I think it would be worthy to build a recorder the

> size of a fridge just to test how the FULL uncompressed data looks compared to

> 35mm

> As of right now, even though we have these cameras with 12mp CMOS's and

> stuff, Nothing "exist's" to record it. Thats seems like a waste to me.

 

See Kinetta. I appreciate that because the Genesis is a Panavision camera then it will become received knowledge that it is the best available, but Kinetta has or will have an onboard recorder capable of what you mention - based on the logic that Genesis has a 12MP sensor from which to draw Bayer RGB and is thus equivalent to around 2 or 3K raw. Dalsa is therefore much more interesting to me as an indicator of what the gold standard should be. They can, by their own claim, shoot 4K, or thereabouts - to the requisite fridge-sized disk array. Filmlight have the solution to grade it, and an Arrilaser will film it out, or better, there are now 4K projectors. Yes, it would be interesting to see.

 

Phil

[David Mullen ASC]

The problem with field data recorders is that you have to figure that on average you may shoot 1.5 to 2 hours of footage a day, so you need enough recorders to do that for at least two or three days in a row. Why? Because those recorders need to be sent to a post facility and be emptied to an even BIGGER data storage unit, or to computer tapes, and then set back to sent, empty again.

 

I was planning once on shooting 4:4:4 on the Viper and I think I needed something like eight or more S-Two DFR's on the set to be covered, and that's HD resolution, not four times that.

 

The beauty of the Genesis over other designs is that the recorder is built-in, like a camcorder, so no cabling to a deck or data recorder.

Edited November 11, 2004 by David Mullen

[John Sprung]

> What is it thats keeping recorder's from recording full Uncompressed Signal from, say the, Genesis?

 

Nobody has yet designed and built a tape machine that extreme. A few years back there was an uncompressed HD tape format called D-6, which was sort of like D-1, only more so. It used the same size cassettes as D-1, but it ran the tape and drum faster. It had 32 heads on the drum. I saw a hotel room demo of it, and people would back away from the machine during spin-up. It sounded like a Unisaw ripping thin plywood.

 

You could probably divvy the datastream up and spread it around on, say, half a dozen D-5 machines -- maybe even with some redundancy, like RAID disk arrays. But again, that leaves us tethered to a rack of equipment.

 

 

> does the technology currently exist to record all that information, and its just a matter of how many Disk's is would take to get that recorded...

 

Yes, exactly. That's what Dalsa and Lockheed-Martin have been doing. Both use big racks full of disk drives.

 

 

> Because if it IS possible, then I think it would be worthy to build a recorder the size of a fridge just to test how the FULL uncompressed data looks compared to 35mm

 

Yup. That's happening now.

 

 

> As of right now, even though we have these cameras with 12mp CMOS's and stuff, Nothing "exist's" to record it. Thats seems like a waste to me.

 

Nothing that would be practical for widespread use in entertainment production. "Tethered to the fridge" mode is adequate for experimental purposes.

 

Is it a waste? Well, in research and development of new technologies, we can't expect all the pieces to come together at exactly the same time. Cameras are ahead of recording technology at the moment. If it takes a really long time, say over ten years, for the recorders to catch up, then I'd say it was a waste to have gotten so far ahead with the cameras. But if they get there in 3-5 years, I'd say that's just the way things go on the cutting edge of technology.

 

 

 

-- J.S.

[Mike Brennan]

>>But it can not make a 4K image of pure RGB data. For a 4K image, 36Mp would be neded in order to have 3 chanels of color information in full resolution.

If genesis were using a bayern pattern sensor and interpolation mechanism, it would be able to make good 4K images.<<<

 

Hmm... a prism with 3 of the Genesis 12.1 million pixel ccds would do the job.

Of course you would need a set of new lenses.

 

 

 

Mike Brennan

[Landon D. Parks]

Hmm... a prism with 3 of the Genesis 12.1 million pixel ccds would do the job.

Of course you would need a set of new lenses.

I understand that you can't use a film lense with a prism (Unless you use a 35mm Lense adapter).... But I have always wondered why? What makes a video lense and Film lense so different that one works with a prism and one don't? :unsure:

[David Mullen ASC]

You can use cine lenses on prism-block 3-CCD cameras as long as they cover the target area and you have a mount adaptor; the only issue is that lenses specifically designed for prism-block cameras will probably be a little sharper because they were designed for the fact that the light is split three ways and the three primary colors have to focus on three different surfaces. The fact that the Zeiss Digi-Primes were designed with this in mind is why Zeiss says that they can't be easily altered to work on Super-16 cameras.

 

By the way, isn't it time that Zeiss or Cooke came out with a set of lenses specifically designed for Super-16? In other words, physically smaller than the typical PL-mount lenses for 35mm, with shorter available focal-lengths? It seems you really have to bodge together an odd lens package to shoot in Super-16 if you want wide-angle primes.

[John Sprung]

I understand that you can't use a film lense with a prism (Unless you use a 35mm Lense adapter).... But I have always wondered why? What makes a video lense and Film lense so different that one works with a prism and one don't?  :unsure:

<{POST_SNAPBACK}>

The big difference is purely mechanical. On a three chip camera, you have a cube of glass somewhat larger than the width of the image between the lens and the focal plane. Some film lenses, particularly wide angles and zooms, often have rear elements far enough back that they have to be where the prism block is. Three strip technicolor had exactly that same problem, so old Technicolor lenses would work fine -- well maybe not fine -- but at least they'd be able to be mounted in a way that would produce an image. Mirror shutter reflex film cameras have this problem to a lesser extent, the shutter has to pass through that space at a 45 degree angle. Look at some Panavision and Arri lenses, and you'll see how they're designed to not collide with the shutter.

 

 

 

-- J.S.