Dynamic Range

[Floris Liesker]

I If you expose to have similar amount of grain as film, you will have filmic highlights. That's my first impression.
IBloom

That's my impression as well. I do think however that filmgrain looks much better than the noise coming off the Red sensor. It doesnt seem to be completely random, I can detect vertical stripes in it. Hopefully that is where the new daughterboard will shine...

 

[Fergus Meiklejohn]

If you expose to have similar amount of grain as film, you will have filmic highlights.
IBloom

What exactly do you guys mean by "exposing to have a similar amount of grain as film"?

 

[jbeale]

I do think however that filmgrain looks much better than the noise coming off the Red sensor. It doesnt seem to be completely random, I can detect vertical stripes in it.

If you have the time to deal with it, it should be relatively straightforward to remove any fixed-pattern background noise. You shoot for a while with the lens cap on, average those frames together, and subtract that from your regular shots.

This kind of "dark frame subtraction" is standard practice for astronomy work. It works best in linear light, before you apply any gamma curves or any other corrections. Ideally it would be built into Redcine, or the camera itself. This technique will not help if there is something nonlinear going on or the pattern is not truly "fixed" (eg. temperature dependent).

 

[I Bloom]

What exactly do you guys mean by "exposing to have a similar amount of grain as film"?

Red by comparison to other mediums is low noise. But there is noise deep in shadows on all formats.

If you underexpose on Red, you will still have a pretty image with a lot of information. Just more noise.

On film, the response curve has a pronounced toe and shoulder, over or underexposing (especially on older film stocks) also affect the contrast as values are mapped onto different parts of the curve.

Since the response curve on Red is very linear, choosing an exposure means sliding values up or down a relatively uniform graph. Contrast isn't greatly affected by where you place those values. Instead you have harsh digital clipping at the top of the graph and greater noise response at the bottom. Personally I prefer noise, but the ultraclean look is growing on me, especially if you take it back to film.

Ian

 

[Policar]

I don't know the first thing about CMOS sensors, but considering that there's no electronic gain going on and all this ISO stuff is based around post-processing...

Wouldn't it be possible to electronically gain down the sensor to ISO 80, while still rating it at ISO 320, thereby shifting the two stops from the shadows to the highlights?

 

[jbeale]

Wouldn't it be possible to electronically gain down the sensor to ISO 80, while still rating it at ISO 320, thereby shifting the two stops from the shadows to the highlights?

All CCD and CMOS sensors have some absolute maximum exposure limit, the so-called "full-well" condition. In normal operation (simplified) a photon hits the sensor and generates a charge (electron) which is stored in a well (capacitor). After enough light has come in, the well is full. Any additional light during that exposure causes no more stored charge, so it is not measured. After the exposure, the charge is measured and the well is emptied, and then you start another exposure.

Now IF the red camera approaches "full well" exposure with the current system design (and I would expect it to, because that's what you do to maximize dynamic range) then you cannot electronically "gain down" the sensor. Well you could, but it would clip at the same exposure, it would just record as grey instead of white, so it would be of little use.

Geeky aside: if there was some electronic way to change the quantum efficiency of the sensor (electrons out per photons in) then you could indeed "gain down" the sensor, just like adding an external ND filter. That would be cool. Unfortunately I don't know of such a design. Maybe it does exist, I just don't know of it.

 

[Peter McCully]

Geeky aside: if there was some electronic way to change the quantum efficiency of the sensor (electrons out per photons in) then you could indeed "gain down" the sensor, just like adding an external ND filter. That would be cool. Unfortunately I don't know of such a design. Maybe it does exist, I just don't know of it.

Mmmmm.... The quantum chip. Next big breakthrough in imaging technology. Will Red be first I wonder?

 

[morpheusfilm]

The reason a DP is telling 6 stops is that he is measuring the range as a contrastt ratio- in other words he reads the best exposure for a scene and looks at how much contrast range holds detail. This is very subjective but common way for a DP to evaluate the real effective range. Basic daylight exposure requires at least a 5:1 contrast ratio capability in the recording media to represent it moderately well. The reason the other test is giving 11.3 stops is that it is more scientific in approach and it's basically under and over exposing an 18% grey card. You overexpose the card until it merges with white and then underexpose until it merges with black and then you have your range. By the way the 11.3 was really dissappointing to me as a DP as I'm able to get 10.7 stops with the Varicam or the new F900R with certain filter combinations. I figure those same tricks may juice the RED to 12 stops but ultimately it still pales in comparison to film which is in the 16-17+ range. This to me is the most significant shortfall of digital cinema formats, and I really believe that this is the main area that needs to be improved in the system to really claim the prize.

 

[morpheusfilm]

differing f-stop ranges

differing f-stop ranges

The reason a DP is telling 6 stops is that he is measuring the range as a contrastt ratio- in other words he reads the best exposure for a scene and looks at how much contrast range holds detail. This is very subjective but common way for a DP to evaluate the real effective range. Basic daylight exposure requires at least a 5:1 contrast ratio capability in the recording media to represent it moderately well. The reason the other test is giving 11.3 stops is that it is more scientific in approach and it's basically under and over exposing an 18% grey card. You overexpose the card until it merges with white and then underexpose until it merges with black and then you have your range. By the way the 11.3 was really dissappointing to me as a DP as I'm able to get 10.7 stops with the Varicam or the new F900R with certain filter combinations. I figure those same tricks may juice the RED to 12 stops but ultimately it still pales in comparison to film which is in the 16-17+ range. This to me is the most significant shortfall of digital cinema formats, and I really believe that this is the main area that needs to be improved in the system to really claim the prize.

 

[David Mullen ASC]

Well, remember that film has also been improving over the decades in terms of dynamic range -- something like 3-strip Technicolor was pretty contrasty and early Eastmancolor negative wasn't much better. I expect digital sensors to also keep improving in dynamic range over time.

 

[Floris Liesker]

it should be relatively straightforward to remove any fixed-pattern background noise. You shoot for a while with the lens cap on, average those frames together, and subtract that from your regular shots.

I've been trying to do exactly that, to get a 'fingerprint' of the sensor to substract that of the image. Unfortunately the pattern turned out to be very non-fixed. I could not make any logic out of it, except that it contained randomly appearing and disappearing vertical stripes. Maybe Greame found some logic in it and whispered it into his daughterboard's ear.

 

[I Bloom]

I've been trying to do exactly that, to get a 'fingerprint' of the sensor to substract that of the image. Unfortunately the pattern turned out to be very non-fixed. I could not make any logic out of it, except that it contained randomly appearing and disappearing vertical stripes. Maybe Greame found some logic in it and whispered it into his daughterboard's ear.

It seems possible that the camera already does some fingerprint subtraction.

IBloom

 

[Álex Montoya]

I don't know, Morpheus. When I look at the footage of Jannard at the races or at this pic, I find that it holds enough info. And the RAW conversion helps squeezing the latitude better than with the Varicam or the 900.

Regarding the fingerprint, I recall Jannard saying that they were gonna get rid of the fixed pattern noise, so I guess they are already using an algorithm, though it obviously doesn't work too well.

 

[david farland]

Apologies for some really dumb questions here.....

1. If film, I'm assuming this is negative stock, has this 16-17stop range...will the print also have this range and will i be able to see the 16-17 stops on the theatre screen?

2. Again, I have 16-17 stop material, what digital or filmic online processes can handle this degree of range?

3. Are there any electronic screens i.e. crt, lcd, digitial projectors etc that I can view this range on?

Dave

 

[Fredrik Harreschou]

Apologies for some really dumb questions here.....

1. If film, I'm assuming this is negative stock, has this 16-17stop range...will the print also have this range and will i be able to see the 16-17 stops on the theatre screen?

2. Again, I have 16-17 stop material, what digital or filmic online processes can handle this degree of range?

3. Are there any electronic screens i.e. crt, lcd, digitial projectors etc that I can view this range on?

Dave

2. Floating point compositing can handle any dynamic range AFAIK.

3. The Sunnybrook HDR display supposedly has a DR of 40,000:1. I'm not sure how to translate that to stops. There are probably others too. Not cheap.

 

[Graeme Nattress]

The only film neg curves I've seen are completely flat at the ends of a 4.0OD dynamic range of exposure. That's 13.3 stops. About 2.5OD (8.3 stops) of that range is linear (ie shadow detail and highlight detail are not in serious compression) I have no idea which usable motion picture film stock would achieve 17 stops (5.1OD range)!

 

[David Mullen ASC]

1. If film, I'm assuming this is negative stock, has this 16-17stop range...will the print also have this range and will i be able to see the 16-17 stops on the theatre screen?

As Graeme said, color negative is more like 13 and a half (maybe 14-stops, which is what Kodak usually says).

No, print stock has a smaller range. This is because: (1) copying film to film causes an increase in contrast (contrast is additive), so the negative has to be lower in contrast than the final print, and (2) when projecting an image onto a white surface, the black density (D-Max) in the print has to be very high in order to look black on the screen, so print stock is designed with a high contrast.

Now the advantages of color negative's wider dynamic range are: (1) more information in color-correction to play around with to optimize for the smaller dynamic range of film prints or video presentation, (2) more natural roll-off into white due to the flatter response to light at the high end of exposure, (3) greater ability to handle mistakes in overexposure.

 

[Harry Lime]

Well, remember that film has also been improving over the decades in terms of dynamic range -- something like 3-strip Technicolor was pretty contrasty and early Eastmancolor negative wasn't much better. I expect digital sensors to also keep improving in dynamic range over time.

And they were slow. I think the 3-strip process they use on "The Red Shoes" / "Black Narcissus" was 25asa and had a rather narrow DR. Perhaps similar to Kodachrome 25?

Both films looked stunning though.... (Jack Cardiff I humbly bow before you...)

 

[David Mullen ASC]

I've spent some time trying to figure out the ASA value of 3-strip Technicolor based on footcandles and f-stops being used on movie sets back then, plus the speeds of b&w film used inside the camera. Remember that the prism block split the light in half, and then the light had to pass through a deep narrow-cut filter. So I would suspect that the system lost at least three-stops before it even hit the film.

Kodak doubled the speed of its b&w stocks in 1938, which is when 3-strip Technicolor also doubled its speed, just in time for "Gone with the Wind". Then there were changes to the prism block and filters that improved speed, and in the mid 1940's, changed the color balance from daylight (scenes had to be lit by carbon arcs) to tungsten.

My estimate is that early 3-strip Technicolor was an effective 5 ASA. By 1938, for "Gone with the Wind", it had become around 10 ASA.

By the late 1940's, it was around 16 ASA. When Kodak introduced color negative in 1950, it was 16 ASA, which they said was "equivalent to currently available color processes." So that's another clue that 3-strip was around that speed by that time.

Another clue was that DP's regularly talked about needing 1000 fc on 3-strip sets, and most photography was done wide-open at f/2.8 on 3-strip -- so that's around 10 ASA.

Of course, you had some leeway to use less light and underexpose the negatives. When I saw a restoration of the 3-strip "Joan of Arc" I noticed that all of Ingird Bergman's close-ups were somewhat milky. I asked Bob Gitt, the restorer, about that and he said all of her close-ups were underexposed. Now maybe that was a stylistic choice or maybe she simply didn't want to act under the necessary footcandles to get a proper exposure.

 

[Harry Lime]

On the Criterion DVD of 'The Red Shoes' there is an interview with Jack Cardiff and if I remember correctly he talks a little about the lighting challenges due to 3-strip being so slow. I believe he also mentions some ASA values.

Can you imagine anyone working with such slow speed films these days?
(or a camera the size of a small fridge)

I'm not a DP, but a still photographer and am still trying to figure out how Brassi shot Paris at night with film around 20 asa and a f3.5 lens.