4K Demo Of Film & Digital Cinema

18th August 2004

I have just received this invite from Walt Ordway, Chief Technology Office, of the DCI group. The DCI (Digital Cinema Initiatives) is the combination of the 7 Motion Picture Studios and their effort to test and suggest standards for Digital Cinema. They are making great progress and we will see the results before the end of the year.

They are inviting everyone to view a side by side test of the recent footage shot and known as the film StEM display reel. The film answer print will be displayed on one side of the screen. On the other side will be two 2k projectors setup in a side by side mode to simulate 4K.

The demo will take place this Tuesday and Wednesday. Everyone is welcome.

Demo times and location are below.

Regards,

Bill Hogan

Bill Hogan wrote :

>The film answer print will be displayed on one side of the screen. On >the other side will be two 2k projectors setup in a side by side mode to >simulate 4K.

This should be quite interesting. Please (everyone who goes) post a report on it.

$350,000 (give or take) in HD equipment vs. a $25,000 film projector. The theatre chains certainly will love those numbers!

What generation is the 35mm print -- is it a contact print off the original camera negative (as an answer print should be) or is it a first-generation print right off the film recorder neg, or is it a couple of generations removed?

That's the big issue in terms of quality of film out, IMHO.

Jeff "OCN rules!" Kreines

Jeff Kreines wrote :

>What generation is the 35mm print -- is it a contact print off the original >camera negative (as an answer print should be) or is it a first-generation >print right off the film recorder neg...?

As I said in my email this is an "ANSWER PRINT". It is off the original cut Negative. Nothing is being shown that is "Digital Intermediate" at this time.

DI would have the possibility of looking better than the "answer print" from the original negative.

(The last statement is MY observation.)

Regards, Bill Hogan

Bill :

>As I said in my email this is an "ANSWER PRINT". It is off the original >cut Negative.

Yes, I know that's what you said.

But the press release you posted may be wrong, because the official stem tests were scanned at 6K for color correction, then output at 4K.

My suspicion is that it's an "answer print" from a DI, not an answer print off of the original camera negative.

Or did they cut the OCN and strike a print from it?

Not trying to be difficult, just trying to find out what is being shown.

Best,

Jeff Kreines

Jeff Kreines wrote :

>My suspicion is that it's an "answer print" from a DI, not an answer print >off of the original camera negative.

For the THIRD time. The print being screened is an "ANSWER PRINT" made from the cut negative.
Yes there has been work on a DI version but that is NOT what is being shown.

I came originally from the "video" side of things. But I transferred the first negative on a flying spot scanner in the US. I know what OCN is and what an answer print or first generation print is.
So there.

Regards,

Bill Hogan

Bill Hogan wrote :

>For the THIRD time. The print being screened is an "ANSWER PRINT" >made from the cut negative. Yes there has been work on a DI version >but that is NOT what is being shown.

Forgive me, Bill. I thought you were just pasting in a press release, and often press releases are written by PR people who get technical details wrong. Believe me, I understand that you, personally, know the difference, and never meant to infer otherwise.

Good to hear they're presenting something printed right off the camera neg.

Any idea what the methodology is they're using to simulate 4K projection with 2 2K projectors? Pixel offset, or something trickier?

Jeff "merely curious" Kreines

Jeff Kreines wrote :

>Any idea what the methodology is they're using to simulate 4K >projection with 2 2K projectors? Pixel offset, or something trickier?

Something simpler. Stacking the right side image as over/under 2K. (2048x1080 on top of another 2048x1080).

Only 1714 lines are displayed x 2048 across on the right side of the screen.

Jim Houston
Pacific Title Imaging

Bill Hogan wrote :

>They are inviting everyone to view a side by side test of the recent >footage shot and known as the film StEM display reel.

So, anyone, what did it all look like? Haven't seen any posts here about it.

Jeff Kreines

JK wrote :

>So, anyone, what did it all look like? Haven't seen any posts here about >it.

4K digital was sharper and had better dynamic range than answer print. Color was very similar. Better (deeper) blacks in digital.

Not surprising as the 2K digital also looked better in previous screenings than the Answer print for the same reasons.

Matt Cowan

Even the digital projection engineers generally do not make claims that they are achieving deeper blacks with digital projection than a film print. If anything, that's been one of the biggest challenges.

So I'm surprised that this demo showed deeper blacks with the 4K projector. I wonder if the print had problems. Well, it's nice to hear that the blacks have improved finally with DLP. (I'm assuming this was a DLP system...)

The demo of the Kodak D-ILA 2K projector had terrible blacks and I think this is one reason why DLP has more potential for improvement than D-ILA.

David Mullen, ASC
Los Angeles

>The demo of the Kodak D-ILA 2K projector had terrible blacks and I >think this is one reason why DLP has more potential for improvement >than D-ILA.

Are you sure it had bad blacks, or could it just have been the set-up on the projector?

Dale Launer
Writer/Filmmaker
Santa Monica

The "simulated" 4K was comparable to 35mm, even perhaps a bit sharper.

The simulated nature of it raises question marks, but not many. It was actually two 2K projectors : top and bottom. Technical info was minimal, but I still appreciated the opportunity.

Blain Brown
DP
LA

ShootMe wrote :

>The simulated nature of it raises question marks, but not many. It was >actually two 2K projectors : top and bottom. Technical info was minimal, >but I still appreciated the opportunity.

You want technical? Here's technical :

The film image on the left was a 35mm Scope answer print from original camera negative, color timed at FotoKem Burbank by the material's cinematographer Allen Daviau. It was one of a batch of 9 prints made for the tests by the ASC and DCI (Digital Cinema Initiative). The film was shown on a Kinoton 35/70 projector with a hand-picked combination lens comprised of a Schneider 85mm prime and ISCO Scope anamorphic adapter. This is the same projector/lens combination that was later used to color correct the digital material in the same room using a discreet Lustre system installed to do the timing. The right half of the image was masked with both black paper and a flag to prevent spill onto the digital side of the screen. This particular reel of film had been shown approximately 30-40 times and was beginning to show some color shifts toward yellow compared to it's original condition.

The digital data was scanned from the original negative on a Northlight scanner at 6K resolution making 16-bit tiff files. It was down rezed to 4K using minimal digital filtration and no compression. As mentioned above the material was color corrected in the same viewing environment against the print on a Lustre using 2K or lower rez proxy files. In actuality, the timing went very quickly and minimal corrections were necessary. Allen Daviau again was present for the final passes. It took far longer to transfer the 4K files from Pac Title where they were scanned to our Lab and back again for final rendering.

The digital images on the right were from two matched Christie CP2000 Digital Cinema Projectors. They were both equipped with prototype 3.0-4.3:1 very long lenses at their longest setting to get the images as small as they were (22 feet wide). Also, because of the very small size (for our theatre) we needed to downsize and defocus the xenon lamps to get down to 14 ft./Lamberts; this led to some unevenness in illumination - hopefully not too distracting. Both projectors were running in an experimental XYZ color space, not the more usual RGB or YCC. Even after careful electronic matching using a Minolta CS-1000 Spectroradiometer, further "expert viewer" corrections were dialled into both projectors down to 1/10th of one percent accuracy. Once calibrated for the Tuesday afternoon sessions, neither projector was turned off until completion of the ninth demonstration session on Wednesday morning.

Data was played back from a pair of DVS HDStation servers in a "Master-Slave" locked configuration allowing frame accuracy across the two devices. They each drove separate projectors over dual-link SMPTE 292M lines at 2048 by 857 resolution from tiff files at 10-bits per color. Unfortunately, the current DVS hardware installed at our facility does not support 12-bits per channel, however, the two servers will be upgraded to 12-bits within the near future. While we had tried some automation to sync the film projector with the servers, we ultimately resorted to doing a manual "count-down" to start the servers against the film (I got pretty good at hitting the "go" button after a few sessions).

To butterfly the images, Pac Tile took the upper and lower left quarters of the original 4K files and made these into separate sets of data for each of the DVS servers. In the projectors we flipped the data horizontally for the upper and lower right side images. These flips do not affect the data in any way.

In the centre port, the Christie projector was pointed up and to the right; the right-most projector was pointed downward. We were extremely lucky in the alignment considering the extreme angles, the slight curvature and the "belly" in the forward leaning screen that we got as good a joint between the two images as we did. At some time in the future, we might consider doing some edge blending but this becomes problematic for us on horizontal rather than vertical edges.

Neither digital projector used internal scaling or an anamorphic lens to preserve square pixels. The total image size was 2048 by approximately 1700 which represented approximately one half of a 4K (4096) image.

Is that enough technical info?

Paul K. Miller
Director, Digital Cinema Lab, ETC at USC, Hollywood
and Founder, Digital Film Image Transfer Society

>The digital data was scanned from the original negative on a Northlight >scanner at 6K resolution making 16-bit tiff files.

TIFF files?? Are you sure about that? Cineon files would be vastly preferable to 16-bit Tiff s from a CC perspective (and at 6K off 35mm 4-perf, bit-depth would be almost immaterial there'd be so much noise - IMHO the 4K 65/70mm in the StEM was way cleaner. What the StEM says to me is that 35mm just can't go where we need to go, and it's days are truly numbered). 32-bit FP TIFF might be different matter.

Tim Sassoon
Sassoon Film Design

"The road to hell is paved with clipped linear files"

>IMHO the 4K 65/70mm in the StEM was way cleaner.

And so much more problematic...

>What the StEM says to me is that 35mm just can't go where we need to >go, and it's days are truly numbered).

What StEM said to me was that a well shot 35mm negative has so much more in it than we are able to see in even the best quality contact answer print - forget release printing! The best paradigm seems to be: shoot 35mm 4 perf., flat, Super-35, or Scope; scan everything at 4K (or better) on a real scanner (not a telecine on steroids), do a DI in a room with a 2K projector (or better), and then do a digital cinema release at 2K (or better). Everything exists except enough 2K projectors for this to work.

The negative will last for centuries; data can always be rescanned from it when higher rez machines are built; and the studios know all about the re-purposing of old films.

65mm is just not going to happen again for anything other than "Special Venue" installations like IMAX. The studios won't go for it, labs are not equipped for it, and the infrastructure just does not, and will not, exist for it to be practical again.

Paul K. Miller
Director, Digital Cinema Lab, ETC at USC, Hollywood
and Founder, Digital Film Image Transfer Society

Paul Miller wrote:

>The negative will last for centuries

Really? In what universe? I'm sorry, but despite Kodak's contentions, I see no evidence of the "permanence" of film records, unless you have no problem with fading and have the ability to maintain absolutely proper vault conditions until the end of time. Film deteriorates just like everything else. The only answer to "permanent" storage of rather fragile stuff like film is to recycle it on to new formats on a regular schedule. While it is true that at this point in time only film has the virtue of being completely format agnostic (i.e., to recover the information, you simply shine a light through it), as well as having an incredibly high "data packing density," let's not get carried away.

These qualities make it the best archival element for now - but not permanent by any stretch of the imagination.

Mike Most
VFX Supervisor
IATSE Local 600
Los Angeles

>65mm is just not going to happen again for anything other than >"Special Venue" installations like IMAX. The studios won't go for it, labs >are not equipped for it, and the infrastructure just does not, and will not, >exist for it to be practical again.

Perhaps you've missed both CFI and FotoKem's significant recent investments in 65/70mm. Give Robert Dennis a call at CFI, or Andrew Oran, just hired away from Gulliver/Paris, at FK. FotoKem has even bought Imagica US (don't know if it's announced yet), so now they have large format opticals, scanning and recording. Despite that, I agree with you, as someone who works with 65/70mm all the time. It's way too fussy for most shows. Which is what's going to drive 4K digital acquisition, and why 1920x1080 isn't the answer, long term. I firmly believe that before this decade is out, features need to take a big step up in image quality.

Tim Sassoon
Sassoon Film Design

"more opinions than is wise" and "fussy, yes, but damn, it looks good"

>TIFF files?? ...Cineon files would be vastly preferable to 16-bit Tiff's >from a CC perspective (and at 6K off 35mm 4-perf, bit-depth would be >almost immaterial there'd be so much noise...

I would've thought uncompressed 6K Tiff's would be decent quality (as well as huge files) - but why would there be so much noise necessarily - even at 16 bit depth ?

Just curious as I aim to learn more about this.

Any place one can see comparisons ?

Mark Doering-Powell

Mark Doering-Powell wrote:

>I would've thought uncompressed 6K Tiff's would be decent quality (as >well as huge files) - but why would there be so much noise necessarily >- even at 16 bit depth?

You replied to the previous message stating: ("and at 6K off 35mm 4-perf, bit-depth would be almost immaterial there'd be so much noise...")

Not everything that you read on the CML forum or elsewhere on the net makes it true. You can say anything. I guess we see that everyday on the CML.

These files were just fine.

Regards, Bill Hogan

>TIFF files?? Are you sure about that? Cineon files would be vastly >preferable to 16-bit Tiff s from a CC perspective (and at 6K off >35mm 4-perf, bit-depth would be almost immaterial there'd be so much >noise

The StEM was scanned to 16-bit DPX at 6K. Because of poor support for DPX in many packages and servers, we converted the 4K masters to 16-bit TIFF but kept the same printing density values so that the Lustre wouldn't have any trouble dealing with them.

Bit-depth wasn't immaterial. I was able to see a distinct (but small) improvement using 16-bit scanning mode vs 10-bit mode. Even at the settings of 6K and 16-bit, the date sampling of the film emulsion is not fully clean which suggest that even higher super-sampling would improve the film signal to noise ratio. (relative to 5218 and 5245 which were the films used in the test)

If you are going for a "perfect" film scan including capturing all of the grain exactly as on the film, you would want to scan at a rate 3X higher than the information density on the film. It is 3X because you not only want higher than the Nyquist sampling rate, but want to capture the phase of the info relative to your sampling array as well. For 35mm film, this suggests the impossibly high number of 15000 samples in 24mm.

So let's just leave 'Perfection' on the shelf with the other textbooks...

Jim Houston
Pacific Title Imaging

Michael Most wrote :

>Paul Miller wrote: The negative will last for centuries Really? In what >universe? I'm sorry, but despite Kodak's contentions, I see no evidence >of the "permanence" of film records, unless you have [...] the ability to >maintain absolutely proper vault conditions until the end of time.

A properly washed and stored black and white negative can last for centuries. Some studies suggest 1000 years if held at 20 deg C.

A good northern cave can handle that...

We know for sure that color negatives won't last that long.

Jim Houston
Pacific Title Imaging

Dale Launer wrote :

>The demo of the Kodak D-ILA 2K projector had terrible blacks. Are you >sure it had bad blacks, or could it just have been the set-up on the >projector

D-ILA is still struggling a bit with contrast issues, but the new small JVC 1920x1080 has very good blacks that looked equivalent to the DLP to me. (They claim 2000:1 contrast ratio)

Jim Houston
Pacific Title Imaging

David Mullen wrote :

>Even the digital projection engineers generally do not make claims that >they are achieving deeper blacks with digital projection than a film print. >[...] So I'm surprised that this demo showed deeper blacks with the 4K >projector. I wonder if the print had problems.

Hmmm. Well film blacks are still deeper but..

Some shots in the StEM material had better blacks on the digital side than on the film print. The flare curve for reproduction (including the projector lens) was probably greater on the film side than on the digital side.

There were several scenes where I had to lift the blacks in order to match the film print better. In other scenes I left the digital blacks alone and they looked better in digital.

Relative to the 2K projection, some blacks looked better because there was more definition in the image and thus local contrast edges were improved. This leads to the illusion that the black is better when it is more likely that it was just more well-defined from one section of the image to the next.

Jim Houston
Pacific Title Imaging

Jim Houston wrote :

>A properly washed and stored black and white negative can last for >centuries. Some studies suggest 1000 years if held at 20 deg C. A >good northern cave can handle that

Sorry, I'm still not buying it. Since motion picture film has only been around for about 100 years, there's no way that such "studies" can be empirically proven. I, for one, remain quite sceptical of such claims, based on known history.

Mike Most
VFX Supervisor
IATSE Local 600
Los Angeles

Michael Most wrote:

>Sorry, I'm still not buying it. Since motion picture film has only been >around for about 100 years, there's no way that such "studies" can be >empirically proven. I, for one, remain quite sceptical of such claims, >based on known history.

Ok, so we have a hundred years of experience with film. Comparing that to how many years we have with Digital media, which seems to be a constantly moving target, how would be go about evaluating the longevity claims made from the digital side?

Mark Smith

>Comparing that to how many years we have with Digital media, which >seems to be a constantly moving target, how would be go about >evaluating the longevity claims made from the digital side?

True, but I believe that is the issue. It is not reasonable to practically claim the shelf life of any media, even film, in the terms some claim for film. I've been near vaults containing nitrate negatives that are much less than 100 years that only a bomb expert would go near and that even the owners of the vaults are afraid of.

GEORGE C. PALMER
HDPIX, INC.
www.hdpix.com

George C. Palmer wrote:

>It is not reasonable to practically claim the shelf life of any media, even >film, in the terms some claim for film. I've been near vaults containing >nitrate negatives that are much less than 100 years that only a bomb >expert would go near and that even the owners of the vaults are afraid >of.

Thank you, George. Yes, that was exactly the point I was trying to make. I was not turning this into yet another senseless film v. digital discussion.

Mike Most
VFX Supervisor
IATSE Local 600
Los Angeles

Mark Smith wrote :

>Ok, so we have a hundred years of experience with film. Comparing that >to how many years we have with Digital media, which seems to be a >constantly moving target, how would be go about evaluating the >longevity claims made from the digital side?

The problem on the digital side is not just the media, but the longevity of the data recording device. How many 8" floppy disk readers are still operational?

Noel Sterrett
Baytech Cinema
www.baytechcinema.com

>If you are going for a "perfect" film scan including capturing all of the >grain exactly as on the film, you would want to scan at a rate 3X higher >than the information density on the film.

How does this relate to scanning a high contrast image from film? Would I be right in presuming if you have "captured all of the grain exactly" you will preserve all the contrast information so to speak ?

>So let's just leave 'Perfection' on the shelf with the other textbooks

What _is_ then the minimum requirement in terms of capturing contrast -- say I wanted to preserve as much information possible from an Ektachrome 100D original ?

(Am I even phrasing the question in the right way ?)

Sam Wells

Thanks for a great screening!

The only area where I felt the 4K Digital projection was not matching or exceeding the shown print was in color depth. I am curious to know if anyone reading this list, who is familiar with this matter, knows if this is due to the colour space conversion from RGB to the experimental XYZ colour space, or maybe something in the DLP projector, or is 10bit with a 2.6 gamma not sufficient as a presentation format? (I noticed that the projection system is being upgraded to 12 bit so I am assuming I didn't get my eyes crossed.)

Aasulv Wolf Austad
DP, Los Angeles

#Bit-depth wasn't immaterial. I was able to see a distinct (but #small) improvement using 16-bit scanning mode vs 10-bit mode.

Even #at the settings of 6K and 16-bit, the date sampling of the film #emulsion is not That is not a 16 bit issue, is it? The real resolution of these scanners is not even close to 16 bit. A real 16 bit scanner has more bits than you will ever need. The last couple of bits are noise.

#If you are going for a "perfect" film scan including capturing all #of the grain exactly as on the film, you would want to scan at a rate #3X higher than the information density on the film. It is 3X #because you not only want higher than the Nyquist sampling rate, #but want to capture the phase of the info relative I'm not aware of anything in sampling theory /Nyquist that supports this. A bit more than 2 times is enough for the phase as well.

Where is this coming from?

#D-ILA is still struggling a bit with contrast issues, but the new #small JVC 1920x1080 has very good blacks that looked equivalent #to the DLP to me. (They claim 2000:1 contrast ratio)

I find the blacks of this device still mediocre and unsatisfactory. Dark scenes look washed out and lack punch and 3 dimensionality.

I had the chance to switch between it and a high end CRT projector and the difference in black level and image beauty of darker material was sobering. The best DLP's have a better black level and about twice the CR. But they also can't compete with CRT in the black level and shadow detail department.

Michel Hafner
www.imdb.com

Michael Most wrote :

>Sorry, I'm still not buying it. Since motion picture film has only been >around for about 100 years, there's no way that such "studies" can be >empirically proven. I, for one, remain quite sceptical of such claims, >based on known history.

Then why is it that I can buy a copy, on DVD, of "Birth of a Nation" with Lillian Gish shot in 1915? That's a film that's basically 90 years old. I'm sure film preservation techniques weren't very good in 1915, but they still were able to pull good video off of it!

Paul K. Miller
Director, Digital Cinema Lab, ETC at USC, Hollywood

Dale Launer wrote :

>The demo of the Kodak D-ILA 2K projector had terrible blacks Are you >sure it had bad blacks, or could it just have been the set-up on the >projector.

Actually, the problem with that demo was bad information, not bad hardware. The material that Kodak prepared in-house for themselves from "Moulin Rouge" was spectacular. The scenes from some earlier materials mastered for "old" DLP's, looked as bad as the "old" DLP's. Had they used different original materials, I'm sure the results would have been much better.

Paul K. Miller
Director, Digital Cinema Lab, ETC at USC, Hollywood

Michel Hafner commented on the following statement :

>"the new small JVC 1920x1080 has very good blacks that looked >equivalent to the DLP to me. (They claim 2000:1 contrast ratio)

He said :

>"I find the blacks of this device still mediocre and unsatisfactory. Dark >scenes look washed out and lack punch and 3 dimensionality."

Michel :

You comment about this projector. It has only been shown in US at NAB. Has it been shown in Europe or were you at NAB? The model number of the projector being discussed is the "DLA-HD2K".

More details here :

http://pro.jvc.com/pro/pr/2004/nab/cybernewsroom/releases/HD2K.htm

Many of us saw this Projector at NAB and it was very good. Both Color and Blacks were quite outstanding.

Just where did you have this projector next to a High End CRT projector? You just don't believe that any other technology can be better than a CRT projector. In my option you are just plain wrong. BUT it does not mean than all DLP or D-ILA projectors are better than some CRT projectors.

Regards,

Bill Hogan

>The StEM was scanned to 16-bit DPX at 6K...we converted the 4K >masters to 16-bit TIFF but kept the same printing density values...

Hmmm. Maybe I'm not as smart as I look, but that still reads a bit ambiguous to me, since DPX allows a lot of things that Cineon does not.

>I was able to see a distinct (but small) improvement using 16-bit >scanning mode vs 10-bit mode.

You've got amazingly good eyes. The precision jump from 10 to 11 bit is .04882812 percent, 11 to 12 bits half that (as a refinement of the previous), and so on. And at best any random pixel of, say, a scanned LAD card is going to be plus or minus as much as 5% off the median or perceived value (in log). In other words, due to film grain a lot of pixels are running at only 5 or 6 bits precision. Even in a fairly large patch one would be hard-pressed to find a single pixel actually at the patch's average or median value. Which is why Kodak felt quite safe specifying 10-bit storage of scanned film images (and after subtracting values outside of normal printing range, you're left with only 9 and change bits of image).

>...3X because you not only want higher than the Nyquist sampling rate, >but want to capture the phase of the info relative to your sampling array.

If you say so. With all due respect - your positions are thought-provoking, and the StEM turned out quite nice, but I think I'll stay on the Kodak side of the spatial argument, too. As previously stated, the 65/70mm section is the convincer in this regard (thanks to Peter et al).

Tim Sassoon
Sassoon Film Design
Santa Demonica, CA

(How did I end up defending Rochester? Sheesh. Where's Bev Pasterczyk?)

reference :

http://www.kodak.com/country/US/en/motion/support/dlad/index.shtml

Michel Hafner wrote :

> A real 16 bit scanner has more bits than you will ever need.

Hi Michel

I'm afraid you are incorrect in your assumption.

A 16 bit scanner would be ideal for working with the OpenEXR format - a 16 bit format that was used here for the VFX work on both 'Troy and 'Harry Potter and the Prisoner of Azkaban'.

> The last couple of bits are noise.

They once were - but technology has moved on. The new generation of scanners now work with a 16 bit pipeline.

Regards

Martin Parsons
The Moving Picture Company
London W1F 0NL
http://www.moving-picture.com

>I was able to see a distinct (but small) improvement using 16-bit >scanning mode vs 10-bit mode.

>You've got amazingly good eyes.

Gee thanks!

Actually though, I looked at a difference image showing exactly what detail was pulled out in a 16-bit scan that wasn't there in the 10-bit. I then used that to look at sharpness and noise in exactly the areas that should have shown the most difference. So I had a cheat sheet.

Tsassoon writes: "The precision jump from 10 to 11 bit is .04882812 percent, 11 to 12 bits half that (as a refinement of the previous), and so on. And at best any random pixel of, say, a scanned LAD card is going to be plus or minus as much as 5% off the median or perceived value (in log). In other words, due to film grain a lot of pixels are running at only 5 or 6 bits precision. Even in a fairly large patch one would be hard-pressed to find a single pixel actually at the patch's average or median value. Which is why Kodak felt quite safe specifying 10-bit storage of scanned film images (and after subtracting values outside of normal printing range, you're left with only 9 and change bits of image)."

I think you are confusing dynamic range with stepwise resolution. Estimates of precision leaving 9bits and change is really only looking at dynamic range (in the presence of a certain signal to noise ratio). You also want to evaluate what can be seen by a typical human. So how big or small a step in luminance is visible. A group that I'm in (the SMPTE digital cinema color ad-hoc committee) recently tested that for digital projection and came up with a minimum threshold of 11-bits (and maybe 12 for a few people in the front row). The results are being published later this year in the SMPTE Journal.

Kodak's 10-bit standard is just short of what several studies have shown is needed for cinema projection (digital or film). In practice, it has been very good, but there are still a (very) small number of cases in film output where you can discover that it wasn't quite enough. The Vision2 stocks challenge this assumption even more because of their reduced grain.

Jim Houston
Pacific Title Imaging

Aasulv Wolf Austad wrote :

The only area where I felt the 4K Digital projection was not :

>matching or exceeding the shown print was in color depth. [....is this] >due to the colourspace conversion from RGB to the experimental XYZ >colourspace, or maybe something in the DLP projector, or is 10bit with >a 2.6 gamma not sufficient as a presentation format?

All good questions. We are still looking at it because several of us still share a concern about this. We are waiting for a 12-bit interface on the DVS server to eliminate at least one variable. But your other suggestions are possible as well and are going to be looked at. It could just be an implementation issue. [and you can add to the list, was there too much quantification in the 3D lookup table?]

Jim Houston
Pacific Title Imaging

Jim Houston wrote :

>Kodak's 10-bit standard is just short of what several studies have >shown is needed for cinema projection (digital or film). In practice, it >has been very good, but there are still a (very) small number of cases in >film output where you can discover that it wasn't quite enough. The >Vision2 stocks challenge this assumption even more because of their >reduced grain.

Adding to this,

10 bits was/is a good round number to fit into 32bits, 2.048 printing density is less than the maximum capabilities of the Kodak print stocks, but in practice most of the images we scan fit into this range, we get the occasional sequence that pushes this in dynamic range terms. It is a standard from over 10 years ago that fit with the available technology.

So the question comes - is 12 bits an increase in precision, range or both. I'd like to see both, but maybe this means more than 12 bits. Certainly remember that your sensor bit depth needs to be greater still if your going to 'log' the image (or apply a power function) and the image processing an even bigger number to prevent too many errors creeping in.

| Kevin Wheatley | These are the opinions of |
| Senior Do-er of Technical Things | nobody and are not shared |
| Cinesite (Europe) Ltd | by my employers |

Mark Doering-Powell wrote :

> Everyone, start printing those jpegs onto archival paper...

Don't laugh! (But worry about those inks... most of them fade.)

Many pre-1914 films survive only because of a flaw in the Copyright laws. Before then, you couldn't copyright a film, but you could copyright a photographs. Edison (and others) quickly realized that there was no reason that said photograph couldn't be 35mm wide and 1000 feet long...and so they printed their films onto 35mm paper, and submitted them to the Library of Congress.

In about 1940, Hal Wallace at the LOC found 3000 of these films under a staircase, and since then there have been multiple attempts to copy them from paper back to film. The efforts were erratic (some films, with usable perforations, were easy to copy; others, with no perforations, had to be hand-registered frame-by-frame).

Soon a new system will be in place to capture these prints digitally, process them for stabilization, dirt, dust, scratches, and stains, and output them to 35mm B&W stock, Estar based, which should be good for a bit of time, at least 100 years at the right humidity/temperature combinations.

Of course, one can't be sure -- everyone thought acetate based film was safe, and then vinegar syndrome reared its ugly head. But in this case, the material will also be saved in digital form -- and the original paper prints will of course be kept.

Jeff Kreines
Kinetta

>The Vision2 stocks challenge this assumption even more because of >their reduced grain.

You mean it was OK to "dumb it down" before because the stocks were grainy anyway ?

>we get the occasional sequence that pushes this in dynamic range >terms.

This talk is what concerns me. Feel free to consider my view "naive" - it certainly is from a technical perspective, as I would be clearly 'out of my bit depth' challenging much of anything in this thread on an engineering level. Me, I just make film images and look at them.

There is a hundred year history of "sequences" that I'm sure "pushes this in dynamic range terms" True enough, anyone creating any kind of dynamic range, then, now, tomorrow must realistically do so taking into reasonable consideration destination materials, medium.

But I want to get a sense here as to what extent these proposals for Digital Cinema are a case of really capturing the fullest range of brightness and color vs. forcing round pegs into square holes so to speak (polygonal grain into square/rectangular pixels)

I read on this thread (can't pull up the exact quote) that - paraphrased - given the information extractable from a 35mm negative (we've heard at least opinion stating the scanning/4K projection chain was giving visual results superior to the graded contact print from same negative) that "we need better" than 35mm projection.

So, how conditional is that last statement ??

Sam Wells

A brief definition of the term StEM?

Thanks

Blain Brown
Alhambra

A brief definition of the term StEM?

StEM is the "STandard Evaluation Material" created by the ASC and DCI. It is about 25 minutes of material from multiple film formats. It is also known as the DCI mini-movie.

Jim Houston
Pacific Title Imaging

>StEM is the "STandard Evaluation Material" created by the ASC and >DCI.

I guess it's not what I was looking for. Friend of mine works in a post house (mostly audio for TV shows) and he is always referring to the "dialog stem" the "music stem" that sort of thing. I thought they might be the same.

Thanks

B. Brown
Alhambra

This was a very fair comparison. There was much in the 4k presentation against the stem material on film that was revealing. The 12bit interface should help. However the one thing that was disturbing (other DP's agree) was the quality of the skin tones.

There seemed to be little or no depth or luminosity or pearlescence (or call it whatever you want) to the close up faces. Later that evening we had a projection sub-committee meeting there. We reviewed the Stem material over and over on 2k. The discussion was really great. After many viewings and stopping at various spots we were able to start to see things that you just don't notice on casual viewing. I think it helped everybody there identify things to look for when testing other systems.

The last thing we did that evening was to fire up one of the projectors used for the 4k experiment and compare that to the 2k projection. We only had a 1/4 or a screen and the material wasn't exactly synced. But the really interesting thing was that even at the back of the theatre you could see the (large) quality difference between 2k & 4k. Not scientific - but a BIG difference.

Steven Poster ASC

Steven Poster wrote :

>"There seemed to be little or no depth or luminosity or pearlescence (or >call it whatever you want) to the close up faces."

I noticed this in the 10 bit RGB vs. XYZ test comparison. The XYZ picture was waxier looking. Were you looking at RGB or XYZ source? Our calculations indicate that an extra bit or two in XYZ should help make it comparable to 10 bit RGB.

Matt Cowan.

Matt Cowan wrote:

>I noticed this in the 10 bit RGB vs. XYZ test comparison.

Could someone tell us more about the XYZ format? Thanks.

Jeff "ABC -- 123 --XYZ" Kreines

Jeff Kreines wrote:

> Could someone tell us more about the XYZ format? Thanks.

Do a Google search on "cie xyz colour space." You'll find almost 120 sources of information. Also, Scott Billups is very interested in the subject, perhaps he could post a summary.

Mike Most
VFX Supervisor
IATSE Local 600
Los Angeles

"There seemed to be little or no depth or luminosity or pearlescence (or call it whatever you want) to the close up faces."

XYZ has a much "longer" luminance vector than RGB. Perhaps its most impressive attribute.

Scott Billups - LA

#Michel :

You comment about this projector. It has only been shown in US at NAB. Has it been shown in Europe or were you at NAB?

I saw it at Cebit and again in a private demo organised by JVC in a 99% light controlled room next to a Sony G90. We played HD from several sources and DVD. The black level was as I said. Whether they got the best out of the device I can't judge, but what they had there were greys for black and the difference to the CRT was big. It was not the production model but the Cebit model.

The production model was promised to have a better black level. Reviews of the Sony Qualia also say that it does not have the black level of the best DLP s which in turn are not as good as high end CRT. The technology is simply not there yet with CR at chip level of <= 5000:1. It was probably as good as mediocre to bad film projection concerning black level. I have definitely seen better blacks in good film projection and from DLP (and far better from CRT).

#They once were - but technology has moved on. The new generation of =scanners now work with a 16 bit pipeline.

I don't doubt it. but do you honestly think that a red, green or blue image scanned from film needs to be as good or better defined as music on a CD, going from ppp to fff? Especially considering what you can project in a real world situation? We are talking acquisition and storage format, not precision when you apply one or more digital filters in a row.

Michel Hafner

>I think you are confusing dynamic range with stepwise resolution. >Estimates of precision leaving 9bits and change is really only looking at >dynamic range (in the presence of a certain signal to noise ratio).

No, I'm fine on that one. I may have confused others, though. I was saying that *even though* the viewable range of the image was being stored at a paltry 9 bits, the relative lack of bit depth was being masked by noise (grain), which dithered the image so that one couldn't perceive stepping. I'll be the first to say that as we move to less-noisy systems, more precision will be required.

>Kodak's 10-bit standard is just short of what several studies have >shown is needed for cinema projection (digital or film). In practice, it >has been very good, but there are still a (very) small number of cases in >film output where you can discover that it wasn't quite enough.

I think it's absurd to discard a standard which has proven itself so reliable, just because one can identify a few exceptions. I could identify a few ten years ago - an encoded maximum density of 2.0 won't even get you started with 5369, for an easy example, but so what? If you have your own film recorder you can get away with all sorts of nonsense - I remember the nasty truncated log system we used at Dream Quest back then. Doesn't at the end of the day make it a good idea, though. Yes, ultimately we'll have to tag files with nm/Kelvin starts and stops and footcandles or whatever, so we can get X-ray plasmas and Bose-Einstein condensates into the same frame, but that's then, this is now.

>The Vision2 stocks challenge this assumption even more because of >their reduced grain.

In practice, I haven't *yet* seen grain levels that low. When this conversation started, I wanted to know how you were using the spectrum on your 16-bit Tiff's for the StEM, and you weren't saying. Were they linear print density, Cineon log with a TIFF header, full-range linear in an EXR-kinda way? Where were the black point and (90%) white points? DMin calibration? Maybe we can get to the bottom of the waxy-looking skin thing - sounds like luminance and chrominance values getting out of step to me.

Tim Sassoon
Sassoon Film Design
Santa "Deakins and Szigmond almost knocked me over tonight" Monica

>Then why is it that I can buy a copy, on DVD, of "Birth of a Nation" with >Lillian Gish shot in 1915? That's a film that's basically 90 years old. I'm >sure film preservation techniques weren't very good in 1915, but they >still were able to pull good video off of it!

Anyone chime in here, I would be curious to know if the video master of "birth" was scanned from original neg or any duplicate even close to it. Or is it, more likely, an extensive "digital" restoration job (as it has been done with many other classics) from some newer print or far removed from the original duplicate neg?

When discussions drifts to 10-12-16 bit, servers, bit depth, colourspace and xyz I find it drifts farther and farther away from actual cinematography and more and more in the realm of post/engineering.

Just my two cents

Daniel Villeneuve, c.s.c.
Directeur-Photo/Director of Photography
Montréal, Canada
demo à / at : http://dvdp.ca

I wanted to know how you were using the spectrum on your 16-bit Tiff's for the StEM, and you weren't saying. Were they linear print density, Cineon log with a TIFF header, full-range linear in an EXR-kinda way? Where were the black point and (90%) white points? DMin calibration?

They were Cineon printing density (therefore log), d-min scanning set at 6080, full white (Dmax) at 65535 [which we only hit in one scanned scene]. Actual low values via histogram usually ranged from 2500 to 57600. These were all the values in the 16-bit DPX files.

The TIFF XYZ files were created (after de-gamming) by the matrix

X 0.464 0.2692 0.1610 R
Y = 0.2185 0.7010 0.0805 G
Z 0.0000 0.0457 0.9087 B

than a gamma of 1/2.6 was applied to put the file back into DLP projector land.

So one place to look for the culprit in the waxy look of the flesh tones is the multiplier for red (0.464). We are losing a bit there which may be a problem. We lost 2 bits in playback as the server couldn't handle 12-bit per color. Then there are the internals of the projector converting the XYZ back to the DLP RGB values. So there are still a few things to figure out.

Jim Houston
Pacific Title Imaging

>They were Cineon printing density (therefore log), d-min scanning set at >6080, full white (Dmax) at 65535 [which we only hit in one scanned >scene]...So one place to look for the culprit in the waxy look of the flesh >tones is the multiplier for red (0.464).

I've been travelling, and just realized I hadn't thanked you for taking the time to explain your process; it was much appreciated.

Tim Sassoon
Sassoon Film Design