Dynamic Range of "Super 35" Celluloid Film?

[Chris Davis]

Quick question: What is the dynamic range of Super 35 celluloid film?

ARRI claims a dynamic range of 13.5 stops with their Alexia.

According to wik, "Red specifies the sensor's signal to noise ratio at greater than 66 dB,[2] with 11.3 stops of total dynamic range." (here). (I do not know if this refers to the updated Mysterium X sensor).

In an article here, the author says there's not much difference in dynamic range of Super 35 and these digital cinema cameras.

I do not know what a "signal to noise ratio at greater than 66 dB" looks like, but I am wondering how many stops celluloid film has when measured in a similar way, so can be compared to these cameras.

Thanks.

 

[Justin Hartough]

That would depend greatly on the film stock you chose. So, you would need to check with the manufacturer or run your own tests. You should also remember that there is a loss in DR from camera negative to release print.

 

[Dan Hudgins]

Noise

Noise

Quick question: What is the dynamic range of Super 35 celluloid film?

There is no simple answer to that question since it depends on the sample area of the film.

The end use size and how much grain you can stand and if you use temporal noise reduction software like ARRI relativity are used.

If you use temporal noise reduction, which is used on film scans now, on Digital Cinema Camera frames as well you can have mid tone lower for the same mid tone noise in the finished result.

If you are going to DVD you can under-expose maybe a stop more than if you are going to Blu-Ray since the mixing of the pixels for the smaller format gives less noise in the end viewing.

If you are talking about a release print or the camera negative, and then what film stock since the grain in all stocks and their contrast are not the same.

Also if you pull process vs. push process you can change the usable f/ stop range of film stocks.

What level of noise one user will feel works for a project may not work for another project.

Using the right amount of fill light can offset many dynamic range issues.

Using Daylight gels on lights can improve the results with Daylight balance sensors since the red/blue blanace will be better and you get less blue noise.

The maximum dynamic range is limited by the lens coatings and intenal reflections.

To get maximum dynamic range from film you can "cem-tone" like process it using 1-PHENYL-3-PYRAZOLIDINONE (PHENIDONE A) then if its color negative, bleach and redevelop in CD-3, but you will not like the results of very long scale since the images will look very flat and lacking in contrast, even if you can see sunspots and spiders in the shadow under a rock at the same time...

Its best to expose for the highlights and add fill light for the shadows.

Kodak has a standard slope used for their CIN scanner and color negatives are developed to that standard slope angle, and then the top and a bit of the bottom are cut off to get "normal" contrast in the scan with counts 470 for 18% gray and 685 for 90% white. So Kodak seems to think that only part of their color negative range should be used for "normal" grading, and they say "soft-clip" should not be used on all shots. With 90 points per stop and a range of 95-685 the f/ stop range USED from color negative is: (685-95)/90 = 6.55 f/ stops.

So 6.55 stops from the negative ends up on the print, that's Kodaks spec from the Cineon documents, if you think I am wrong about that please quote the Cineon documents or other Kodak documents relating to film negative scans. Since they recomend using hard clipping, what else is on the negative does not show and does not matter other than some "safety" with getting the exposure wrong, but film is non-linear at the ends, so you cannot just slide up and down the S-curve like you can the straight line curve sensors have. Also film resolution is lower in the high exposure areas, and very high grain and lower contrast in the under exposed areas, so even if you use a large sample area and say "I can see some density change" at both high and low ends of the exposure range, you cannot print those ends and get as good an image as printing from the center of the S-curve.

 

[David Mullen ASC]

Kodak generally claims around 15-stops of DR for Vision-3 color negative film stock.

But all these figures, both for film and for digital, are relatively higher than the practical DR ranges that we work within generally, so it's always a bit controversial because the published figures don't really seem to correspond to real-world working situations.

Generally when I shoot film, I can count on about a 10-stop range being visible... i.e. things tend to go black 5-stops under and burn-out to white about five stops over. And most digital cameras have about two stops less on the high end range. But that doesn't mean I am claiming that color negative film only has a 10-stop range and that high-end digital cameras only have an 8-stop range.

I find these DR figures to be mostly a marketing tool because manufacturers publish their own product's figures but not figures of competing cameras measured at the same time using the same method (they'll only make vague claims as to the superiority of their camera to their competitors.)

Anyone who just takes one company's figures and compares them to another company's figures is on a fool's mission. It's pointless. All that matters is relative differences revealed through testing, not the overall figure.

The thing with DR tests is that some manufacturers count detail buried in the noise (or grain) floor, as if that were really usable.

Personally, I think modern high-end digital cinema cameras are about two stops short (more or less) of color negative film's DR, entirely in the overexposure end, due to the behavior of film to become lower in contrast at the ends of the exposure range.

--

The resulting DR from various presentation systems is less relevant -- you always want more range in the origination medium, otherwise you have no ability to slide the image brightness up or down, you're stuck more or less to how you exposed it. Print stock HAS to have a high gamma or else you wouldn't get a decent black after all that projector light blasted through a print and onto a white surface.

 

[Gerald Menclik]

always a pleasure to read your explanations, David!

 

[Dan Hudgins]

DPX/CIN

DPX/CIN

One does have to ask how these ratings of 13.5 or 15 stops are packed into a DPX/CIN file?

1024/90=11.3 stops maximum.

When film is scaned base density is set just under 95 unless compensation is done to reduce the exposure in the scanner which brings down the top end, if the D-max of the negative is over the scanner range the most you can get back "linear stops" is one stop.

In order have more than 11.3 stops in a DPX/CIN file you need to have the head and toe of the curve non-linear, so your adjustment range is:

95/90 = 1.05 stops down for under exposure

1023-685 = 3.75 stops up for over exposure

If the scanner is set to have base density just under 95 then you get no more detail when you slide down, you just wash out the black areas.

When you slide up, you cut off the "stops range" under the black clip point since Kodak did not allow for soft clip in the black end of the curve.

If you put the 6.55 stop sample window at the top end, you do get more than 6.55 stops but they are non-linear.

If you use softclip, you can get 6.55+3.75 = 10.3 stops about, so that may be where David gets his 10 stops from? (softclip does not add anything at the bottom, just the top highlights, although if base density is set at just below 95 the foot is non-linear also anyway, but if the scanner is in "overexposed mode" the foot is not as non-linear so you get fewer stops in the negative scan)

So how does ARRI compute a curve that relates to packing their 13.5 stops into the standard 95-685 range of a DPX/CIN file?

Does ARRI make both the head and foot non-linear in an S-curve?

Does the mid-tone slope of the ARRI corraspond to the mid-tone slope of normal process Vision 3?

 

[Eki Halkka]

The thing with DR tests is that some manufacturers count detail buried in the noise (or grain) floor, as if that were really usable.

In my opinion, that's a perfectly viable way, maybe the best: it tells us the absolute maximum range of any given camera, and is very usable figure at that. This is a value that can be measured with a reasonable amount of precision.

But you're also right about the usable range - it's always much smaller than the absolute number. The problem is, what's "usable" is pretty much a matter of taste, how much noise or grain is acceptable to you. As there's no standard for this whatsoever, while somewhat useful, this figure is just a very, very rough ballpark figure.

Personally, I think modern high-end digital cinema cameras are about two stops short (more or less) of color negative film's DR, entirely in the overexposure end, due to the behavior of film to become lower in contrast at the ends of the exposure range.

That of course depends on how you rate the camera: if you rate Red One at ISO 320, you're probably right. If you rate it at ISO 1280, you have two more stops of highlight range at the expense of more noise, so you can then create a gamma curve that mimicks film in post.

Noise of course is yet one more very subjective thing - i personally see it as very much the same thing as film grain, even if the appearance is slightly different. If we allow similar amount of noise for i.e. Red One as with i.e. ISO 320 film, you can rate Red very high for additional highlight protection, especially with the MX sensor. Rate Red at ISO 5120 and you have four more stops of highlight protection compared to rating it at ISO 320.

 

[Dan Hudgins]

Where's the bottom?

Where's the bottom?

Rate Red at ISO 5120 and you have four more stops of highlight protection compared to rating it at ISO 320.

With fixed patern noise there is somewhat off a bottom, although you can subtract some of the fixed pattern noise in a sensor, but not in film, so that's one way sensor can gain an advantage.

"Random" and moving pattern noise can be reduced with temporal noise reduction in BOTH film and digital shooting, so you can gain maybe two stops of usable range by using temporal noise reduction.

It seems common now to use temporal noise reduction on FILM scans so that is looks closer to the range of Digital cameras, that may have built in temporal noise reduction.

There is so much digital processing going on with FILM scans to "de-gran" and de-dust them, that very few people know what "film" looks like, what they think film looks like is a digital processed version of a film scan, so it is more what does the digital workflow produce if you "START" with film, or some Digital camera, and how much CPU/GPU power you can use on the "raw" data to beautify the results.

How many "films" were made last year WITHOUT the negative going through a scanner and you watched just a film-to-film 4th generation print? Any?

If you are looking at a DVD or Blu-Ray, then you are NOT looking at "film" but some digital image based in PART on a film scan...

How does ARRI scanner vary from the Cineon specs, that they can claim that you get more stops in their results, do you see those stops in a CIN/DPX file made from their scanners, or is it just something internal to their hardware that gets clipped before the end processed frame files are made?

With film you have processing marks and gunk all over the images, so that limits some of the processing you can use to bring up shadow detail.

With sensors you have fixed pattern noise, and some moving pattern readout noise, but from what I have seen, its easer to pull up the shadow details with processing in a digital camera, than a digital scan off film.

Film has more of a "hard" bottom where the gunk and grain make it hard to get much back.

Digital has a "hard" top where you white clip and then you can only use DRX processing to pull some detail out of the red and blue since they do not clip as soon.

If you have a digital sensor on a true RAW camera you can shoot through greenish filter to get a wider DRX spread, in that way you can gain a stop at the top and bottom, then you would "match film", maybe?

Has David used the DRX highlight recovery and shoot at 800 (1600?) on M-X yet? Then run the footage through temporal noise reduction?

 

[Chris Davis]

Follow up Questions Later

Follow up Questions Later

It appears the answer(s) to my question are anything but simple, which good. Before I ask follow up questions, I need to read up on and develop a more sophisticated understanding of some of the concepts discussed here. I am enjoying the in-depth dialogue in this thread.

Will post again in a few weekends.

Many thanks,

Chris

Edit: P.S.: Looking forward to a response to Dan's last post.

 

[Eki Halkka]

Has David used the DRX highlight recovery and shoot at 800 (1600?) on M-X yet? Then run the footage through temporal noise reduction?

MX is so clean, that noise reduction isn't really needed at ISO 800. Sure, there's visible noise if you pixel peep, but not much.

 

[Chi Ying Chan]

"1024/90=11.3 stops maximum."

Thats why i said 10bit log is NOT ENOUGH!

Regards
Chan Chi Ying
DP HK

 

[Bruce Watson]

Good responses here. I would like to reiterate that film and digital are different media, each with its own strengths and weaknesses.

I can make an analogy with audio. There were these interesting fights over digital and analog audio back when CDs first made it to market (yes, I know this shows what an old guy I am, sigh...). Digital recording (16 bit) is theoretically good for 96 dB, but in practice somewhat less, call it 90 dB. Analog recording (studio tape) measured considerably less, around 80 dB, but a vinyl LP was much worse, say 60 dB.

So you'd think that the CD would be the clear winner, yes? Except that it didn't sound that way to careful listeners (aka audiophiles).

Turned out that with analog, there was considerable interesting signal content below the noise floor. So yes, technically, the analog medium had a s/n ratio of 60 dB, but you could hear differences in the signal in the range of 90-110 dB (depending on who you believed). So yes there was hiss, but you could still easily hear the solo oboe and the individual audience members coughs below it.

This situation resulting in years of struggle with corporations telling consumers that CDs rocked, and consumers answering back that CDs were insufficient. Now we have ready access to recording technology with 24 bits, and sample rates of 96 and even 192 kHz. Now digital audio finally starts to sound pretty darn good, even to us die hard audiophiles.

There are some complementary analogies available between digital and film video, but I've already made this post much too long. This seems to be a pretty smart crowd, and I can't cover it as well as some of the posts already made on this thread. Suffice it to say, I think digital video is getting there. And Red is one of the reasons why.

 

[Eki Halkka]

So you'd think that the CD would be the clear winner, yes? Except that it didn't sound that way to careful listeners (aka audiophiles).

Turned out that with analog, there was considerable interesting signal content below the noise floor. So yes, technically, the analog medium had a s/n ratio of 60 dB, but you could hear differences in the signal in the range of 90-110 dB (depending on who you believed). So yes there was hiss, but you could still easily hear the solo oboe and the individual audience members coughs below it.

30-50 dB is a LOT - so i'm not sure how much sounds that are that much below the noise floor matter, even IF someone could hear those (which i do doubt pretty seriously).

I'm also pretty darn sure that most audiophiles would have a very, very hard time hearing a difference between an analogue vinyl record and a digital 16/44.1 recording of the said record.

The problem was not that CD is not good enough - but rather that it's too good. CD lacks the distortion, noise and compression of an analogue record, and people tend(ed) to associate those features with "warmth" and "quality".

It is indeed analogous to the film vs. video debate though - a lot of the "film look" we associate with "quality" and "warmth" comes from the artifacts of film: jitter, grain etc.

 

[Dan Hudgins]

top or bottom

top or bottom

It is indeed analogous to the film vs. video debate though - a lot of the "film look" we associate with "quality" and "warmth" comes from the artifacts of film: jitter, grain etc.

Digital recordings of audio have the MAXIMUM distortion at lower recording levels, whereas Analog recordings have the MINIMUM distortion at lower recording levels.

Digital recordings of audio have the MINIMUM distortion near maximum recording levels, whereas Analog recordings have the MAXIMUM distortion near higher recording levels.

The ear works more like Analog recording so when you hear Digital Recordings you hear the distortion at low levels. It depends on the audio compression ratio uses as to how much the levels change while recording, with clasical recordings that vary in volume you will hear this more than in compressed recordings.

It also depends on if you have any Class B amp stages, if there are any non Class A amp stages, then you get digital like distortion at the lower volume levels, and so you would not hear much if any difference.

Since few people have "direct to disk" vinyl recordings and ALL Class A J-FET amps, few people would be able to hear any difference, although violins do sound different from CD and Tape and Vinyl.

Its not so much CAN you see or hear any difference, its more a matter of what is Practical.

If it was just what looks better EVERY movie would be shot and projected in 65/70mm, or all film-to-film IMAX. That is not going to happen.

And people will stop shooting 35mm movie film because its Practical to shoot digital since with heavy digital post production the advantage of shooting film over digital gets less and less each year. Even today it can be very hard to tell what you are looking at with regard to the camera media after heavy digital processing in post production.

I do not see any audience demanding 65/70mm film-to-film printing, nor would such films make more money than something shot Cinealta with DCI projection.

 

[Bruce Watson]

30-50 dB is a LOT - so i'm not sure how much sounds that are that much below the noise floor matter, even IF someone could hear those (which i do doubt pretty seriously).

Not an audiophile then?

I'm also pretty darn sure that most audiophiles would have a very, very hard time hearing a difference between an analogue vinyl record and a digital 16/44.1 recording of the said record.

Decades ago I had several pairs -- vinyl and CD, from the same master. We actually organized some double-blind tests to find out the answer to this very question. Turns out, if you have half decent equipment, it's pretty easy for anyone to hear the difference. I'm talking kids, disinterested wives, and the old guy from across the street -- not just audiophiles.

The problem was not that CD is not good enough - but rather that it's too good. CD lacks the distortion, noise and compression of an analogue record, and people tend(ed) to associate those features with "warmth" and "quality".

Um... no. CD wasn't "too good." If it were too good, we'd never have seen SACD, and companies like Sound Devices wouldn't be making equipment to record 24/96 and better.

What CDs are is good enough. CD won because it solved a host of problems for the vast majority of customers -- from wow/flutter to ticks and pops. But it didn't record sound better. It won because it was vastly more convenient. And if people like anything, it's convenience.

It is indeed analogous to the film vs. video debate though - a lot of the "film look" we associate with "quality" and "warmth" comes from the artifacts of film: jitter, grain etc.

Yes. There's surely a lot to this.

But digital is, if nothing else, much more convenient. For example, it's nice to know you nailed your exposure on the spot. And that tricky follow focus. And that intricate dolly move. It's nice to be able to have your rushes so quickly. And it's really nice not to have to deal with tens of thousands of feet of film on a shoot -- not having to load/unload magazines, run the film to the lab, process it, etc.

IMHO, with Red and Arri giving film a run for it's image quality money, the sheer convenience of digital is going to push film over the edge of the cliff. C'est la vie.

 

[Victor Papalucas]

It would be cool that RED make 5K film scanner, 5K projector and 5K DCP.
Arri would be totaly dead... All these stuffs for 300.000USD


I've seen 35mm DPX 16bit 2K material in baselight and it was great !

 

[Eki Halkka]

Not an audiophile then?

Working as a part time audio engineer for quite some time has weeded that side out of me ;-)

Decades ago I had several pairs -- vinyl and CD, from the same master. We actually organized some double-blind tests to find out the answer to this very question. Turns out, if you have half decent equipment, it's pretty easy for anyone to hear the difference. I'm talking kids, disinterested wives, and the old guy from across the street -- not just audiophiles.

You understood me wrong: what i suggest is that you take the vinyl recording, digitize it (with snap, crackle, pop and all) to a good quality 44.1 / 16 kHz digital recorder, or just run it through an AD/DA which is essentially the same thing. Not that you compare a vinyl and a CD from the same master (unless you also have the said original master at hand for further comparison).

Um... no. CD wasn't "too good." If it were too good, we'd never have seen SACD, and companies like Sound Devices wouldn't be making equipment to record 24/96 and better.

"Too good" was in reference to vinyl recordings - lack of subtle distortion etc. artifacts was perceived as digital being too "Harsh". Also, with CD it was possible to use louder high frequency (and very low frequency) content on the mix, which made some masters sound very different too (it was physically impossible to use those with vinyl).

24/96 is in the realm of "nice to have", especially for recording source material - it's essentially the same as shooting 4K for 2K release. But it's actually pretty rare to make much use of the extra quality: the analogue gear you record with (mics, preamps), and the environment (background noise) make it pretty rare to run into situations with higher than 96 dB required dynamic range. Some classical music recordings might be in this category i guess.

What CDs are is good enough. CD won because it solved a host of problems for the vast majority of customers -- from wow/flutter to ticks and pops. But it didn't record sound better. It won because it was vastly more convenient. And if people like anything, it's convenience.

Here we tend to disagree - in my personal opinion, a CD is a much more faithful reproduction of the sound on master tape than a vinyl is (i've mixed / mastered for both, though just our own small-time garage bands' indie releases - all my paid audio work is video-related). Vinyl may sound and especially "feel" more pleasing, but that's another issue.

 

[Chris Davis]

Follow Up Questions: DSLR's, Celluloid Film, Signal to Noise Curve, DPX/CIN files

Follow Up Questions: DSLR's, Celluloid Film, Signal to Noise Curve, DPX/CIN files

With fixed pattern noise there is somewhat off a bottom, although you can subtract some of the fixed pattern noise in a sensor, but not in film, so that's one way sensor can gain an advantage.

"Random" and moving pattern noise can be reduced with temporal noise reduction in BOTH film and digital shooting, so you can gain maybe two stops of usable range by using temporal noise reduction....

I’ve read a bit and metered with my still camera a lot. I think I have enough info to post a follow up question or two…

Film has more of a "hard" bottom where the gunk and grain make it hard to get much back.

Digital has a "hard" top where you white clip and then you can only use DRX processing to pull some detail out of the red and blue since they do not clip as soon...

I have read elsewhere that shooting RAW on DSLR’s is somewhat analogues to digital audio recording – that is, “expose to the right but don’t clip.” Or, make the audio signal as “hot” as you can without peaking above 0 db.

The logic is something like this: sensor noise continues to decrease up until you clip in at least one channel, because you are using more of the available bits in a 12 or 14 bit RAW file. Middle gray is subjective, and there is actually more noise at middle gray than a few stops over. If you want the greatest signal to noise ratio, you may even do better to overexpose in certain (maybe rare) low contrast situations, just enough so that the highlights are perhaps a stop or so away from clipping; then “pull process” the RAW file back in Lightroom so that the photo is the “correct” brightness. Purposeful overexposing in this way would seem to be used more with studio lighting or an overcast day than on a sunny day just before dusk.

But this idea is a total curveball for me. Is there any truth to this? Maybe it applies to analogue gain sensors (e.g. Canon and Nikon DSLR’s), but not digital gain sensors (e.g. Redcode Raw)? Maybe the idea is just bunk? If it is true, does the signal to noise ratio decrease leniarly, say from 3 stops over to 4 stops under middle gray?

...Generally when I shoot film, I can count on about a 10-stop range being visible... i.e. things tend to go black 5-stops under and burn-out to white about five stops over. And most digital cameras have about two stops less on the high end range. But that doesn't mean I am claiming that color negative film only has a 10-stop range and that high-end digital cameras only have an 8-stop range....

...If you use softclip, you can get 6.55+3.75 = 10.3 stops about, so that may be where David gets his 10 stops from? (softclip does not add anything at the bottom, just the top highlights, although if base density is set at just below 95 the foot is non-linear also anyway, but if the scanner is in "overexposed mode" the foot is not as non-linear so you get fewer stops in the negative scan)...

Is “softclip” sort of like well-done audio compression, where 3.75 stops of are squeezed to into a maybe stop or less (sort of like 12 db of peak audio dynamic range compressed into 3 db)?

In dp review, crop sensor DSLR’s commonly have around 8 stops of “usable” range here, and full size sensor DSLR’s have around 10 stops (stretchable to 12 stops with color shifts) here.

Recently experimenting with metering on my still camera, I see a 3-4 stop difference between the left and right side of a person’s face when she is standing perpendicular to the sun 1-2 hours after dawn (please correct me if I’m wrong here, or if this difference is mainly to glair, as I did not use a polarization filter).

I am a curious hobbyist, and I have never worked with DPX/CIN files, I’m sure David is correct re: their dynamic range, but it seems I commonly see both the sky and people’s faces in movies, sometimes with the orange (not white) sun flickering through the trees. It seems I see much more dynamic range than this in movies, but maybe it’s just careful metering and clever lighting? How do I reconcile this difference?

Where I was originally coming from with my questions, is that I don’t remember ever seeing pure white on color still photographs, but I see it all over still pictures taken with DSLR’s. But this may not relate at all to DPX/CIN files I see in movies.

Thanks for your time, Dan and David.

Sincerely,

Chris