Epic sensor's dynamic range

[david farland]

Has there been any word on the increase in dynamic range of Epic v1 over the current Red?

 

[Michael Lindsay]

Not that I've noticed.

I haven't seen a stills camera with noticeably better dynamic range than Red so I do wonder what is actually possible. I'm sure it will be better but what is realistic to expect I have no idea. Anyone else? It might be worth Red jumping in to help 'manage' our expectations. What is possible?

As it is there is plenty of latitude for my studio shoots.

3+ extra stops with 14bit sampling would cover all bases and make all criticisms purely academic (IMHO). I suspect that is too much to expect. If we did get that I would sell my Ultra Primes and get MP.

At the moment certain wide city night exteriors are compromised/tricky and run and gun high contrast day stuff needs perfect judgement. Anything more will help. Even if it is only a stop.

hopefully

Michael

 

[Patrick Tresch]

Have you seen a picture shot with Nikon D3 at 1600 asa?

It's "like" shooting with RED at 500 asa.

Future will be brighter!!!

Patrick

 

[Michael Lindsay]

I have seen some pretty amazing D3 lowlight images. However I thought that the the larger photo sites made this possible.

regards

Michael

 

[R. Schorman]

It will be interesting to see how they manage to squeeze more dynamic range out of higher pixel density.

 

[Charles Angus]

It will be interesting to see how they manage to squeeze more dynamic range out of higher pixel density.

To a certain extent those are the same thing. Despite conventional wisdom, more dense photosites decreases noise at the baseline resolution. For this comparison, 4k would be the baseline (existing RED camera) and 5k would be the new resolution. Downrez the 5k to 4k intelligently and you will see lower noise.

Because you have less noise, you have more dynamic range because more toe is usable.

 

[Kyle Presley]

Makes sense, but what about sensitivity? Smaller photosites means fewer photons gathered by the individual site. Won't the blacks clip sooner, giving less usable range?

 

[Daniel Browning]

Makes sense, but what about sensitivity? Smaller photosites means fewer photons gathered by the individual site.

Right, but there are more sites. If the sensitivity per pixel decreases by the same amount as the increase in the number of pixels, then the net result is the same.

A sensitivity of 20% incident photons measured per pixel in a 4K camera is the same as 15% per pixel in a 5K camera.

Look at the cheap digicams. For many years, their ultra-tiny 1/1.7" sensors (40 square mm) have had double or quadruple the sensitivity (per area) of any DSLR. They've long had gapless microlenses and fill factors that are just now starting to trickle up to larger sensors (the D3 is the first, notably).

Obviously, having four times the sensitivity per area is undone by having an area that is 25 times smaller, but it proves that sensitivity does not necessarily get worse as pixel size shrinks.

Won't the blacks clip sooner, giving less usable range?

The RAW data, if it's engineered correctly (not like Nikon) does not clip blacks: it preserves the noise in entirety. Charles Angus Taylor addressed the question of noise very well.

 

[Daniel Browning]

Has there been any word on the increase in dynamic range of Epic v1 over the current Red?

Not since Jim said that Scarlet will have "more DR than a RED ONE (before upgrade)".

He didn't qualify if that was per pixel or at equal output sizes. Let's assume "more" means a 67% improvement (one third-stop).

If it's on the basis of equal output sizes, then it means Epic will have at least 1 and and 1/3 stop additional DR.

If that's on a pixel basis, then it means the Epic will have at least 2/3 stop more DR.

 

[Hans von Sonntag]

Since Red obviously matured to a cine tech company one can count one with remarkable success in a astounding short time, I'm sure they know by now how the game is played.

Epic is aimed at the pro market, mainly rental houses and private owner who either have a RedOne already or still shoot 35mm and wait for the next genration that will have all the tedious bugs fixed that often arise with the first generation.

Sure there are many quirks Epic can fix, especially if we talk about ergonomics, but mainly people will focus on DR (as David, I and many others do) and the sensors performance over all.

I'm dead certain the DR will be significaltely broader. Epic must have a 14 bit AD-transformer, a theoretical range of 13+ f-stops and a usable DR of 11 stops which will bring it in the realm of 35mm. Otherwise Red will lose their leadership.

That may delay things a bit but keeps them ahead of the rest of the crowd which are woken up by now. Epic MUST become the "film slayer" otherwise all will be disappointed.


Hans

 

[Alberto Caprioglio]

Otherwise Red will lose their leadership.

what an assumption.

Epic MUST become the "film slayer" otherwise all will be disappointed.

Here Red one is already beginning to replace a lot of film, and people are not quite disapponted about it's image quality, having a final higher resolution once projected on a big screen, and having a dynamic range next to film - certainly closer to film than all the previous video machines.
Not to mention s16mm. I bet lot's of people will stop using s16mm with it's PAL-like resolution and film cost - and cameras that sometimes are more expensive then a Red - and will pass to Red Scarlet (same sensor size, probably), or directly upgrade to Red One.
Anyway, I don't care about what one does and what does not.
I only need batteries always ready and bigger compact flash cards, and I dream of a future RAID-1 red drive, since I don't want to loose my footage at the first drive failure. (strange, uh?)
The remain is just a sitting room for frogs, talking and smoking forever.

 

[Stephen Pruitt]

I don't care what they do to the sensor with respect to size. I just want less noise and a greater dynamic range.

Stephen

 

[Alberto Caprioglio]

I don't care what they do to the sensor with respect to size. I just want less noise and a greater dynamic range.

I agree. We all want more and more dynamic range, no matter what they give us.
But I confess I am already quite happy with Red one dynamic range at this moment. More dynamic range would be great, but we are not going badly at all.
We are quite relaxed. If something cannot absolutely be done at this state of art, well, never mind, we are getting very very close to that, and most importantly, this doesn't prevent us from taking great images.

 

[Alberto Caprioglio]

hey, wait.
The possibility of using ALSO bigger format lens would be very useful, actually. I would love it.
On the tripod, masses of photographers may use Epic also instead of some type of studio, medium format and bellow still camera. Epic look quite compact, for such a beast that it is.
One only mega instrument for studio photography and film making.
If it has a big sensor, it may be done.
It could truly became a world standard for imaging, both still and motion.

 

[Pawel Achtel]

EPIC will have 14 stop DR using 16 bit Red Raw

EPIC will have 14 stop DR using 16 bit Red Raw

Of course, this is pure speculation on my side. However, it is based on what is achievable currently with today's technology, not what my wants are and not completely out of touch with reality :clown2:

CMOS sensors have been able to achieve 14 stop dynamic range for some time (ref 1.). However no commercially avialble camera can offer this capability yet, even if the sensor itself can. This is because of DSP precision and in particular ADC quantization level. Introduction of either dual ADC or, better still, providing quantization step that is below noise level of the CMOS, can effectively provide 14 stop DR out of the CMOS sensors available today.

In order to extract 14 stop dynamic range out of CMOS sensor, one would either need to use double channel quantization or minimum 15-16 bit ADC. I opt for a 16 bit quantization because it allows for 1 stop over the saturation level and 1 stop under the noise floor. It also offers greater precision than 15-bit that may be still too large step in comparision to the noise level of the sensor. Also 16 bit would allow easy (or easier) upgrade to a better sensor in the future. So, RED Raw would need to be 16 bit too.

Therefore just by harnessing the available technology it is quite possible and realistic to design a camera offering full 14 stops of dynamic range.

In order to go beyond the 14 stops, one would have to make some breakthroughs in CMOS sensor technology itself and, in my speculative view, this is unlikely and probably not essential at this point as 14 stops would be the best in the industry and certainly exceeding the best of the best of film stock negatives.

References (very good reading):

1. Noise, Dynamic Range and Bit Depth in Digital SLRs
2. Canon White Paper

and relevant discussion on RedUser in a thread started by jbeale:

3. http://www.reduser.net/forum/showthread.php?t=20838

 

[I Bloom]

Of course, this is pure speculation on my side. However, it is based on what is achievable currently with today's technology, not what my wants are and not completely out of touch with reality :clown2:

CMOS sensors have been able to achieve 14 stop dynamic range for some time (ref 1.). However no commercially avialble camera can offer this capability yet, even if the sensor itself can. This is because of DSP precision and in particular ADC quantization level. Introduction of either dual ADC or, better still, providing quantization step that is below noise level of the CMOS, can effectively provide 14 stop DR out of the CMOS sensors available today.

In order to extract 14 stop dynamic range out of CMOS sensor, one would either need to use double channel quantization or minimum 15-16 bit ADC. I opt for a 16 bit quantization because it allows for 1 stop over the saturation level and 1 stop under the noise floor. It also offers greater precision than 15-bit that may be still too large step in comparision to the noise level of the sensor. Also 16 bit would allow easy (or easier) upgrade to a better sensor in the future. So, RED Raw would need to be 16 bit too.

Therefore just by harnessing the available technology it is quite possible and realistic to design a camera offering full 14 stops of dynamic range.

In order to go beyond the 14 stops, one would have to make some breakthroughs in CMOS sensor technology itself and, in my speculative view, this is unlikely and probably not essential at this point as 14 stops would be the best in the industry and certainly exceeding the best of the best of film stock negatives.

References (very good reading):

1. Noise, Dynamic Range and Bit Depth in Digital SLRs
2. Canon White Paper

and relevant discussion on RedUser in a thread started by jbeale:

3. http://www.reduser.net/forum/showthread.php?t=20838

I think you may be right and I think this will be the main advantage over Red One when selling Epic to producers: "Want to shoot in bright sunlight? Not much opportunity for fill? No Problem!"

IBloom

 

[Dj Joofa]

14 stops would be the best in the industry and certainly exceeding the best of the best of film stock negatives.

A simple calculation with 1% JND assumption would reveal that you would need at least 17 linear stops to equal 10 bit film DI scan (0.6 gamma assumed).

 

[Pawel Achtel]

A simple calculation with 1% JND assumption would reveal that you would need at least 17 linear stops to equal 10 bit film DI scan (0.6 gamma assumed).

I assumed no gamma (gamma=1.0).

I assumed the DR of good quality negative film is between 11-13 EV. What is the scan quantization and whether it is linear or log, is irrelevant. You can not get 17 linear stops out of the film, especially with a 10 bit scan.

 

[Dj Joofa]

I assumed no gamma (gamma=1.0).

Not sure, you can assume a gamma of 1. The gamma of film stock varies, and is also partially dependent upon the type of processing involved (e.g., push/pull). However, for the sake of this calculation we shall assume:

(1) A very standard "average" gamma of 0.6, and,
(2) Just take into account the "linear" portion of film negative and ignore toe/shoulder.

I assumed the DR of good quality negative film is between 11-13 EV. What is the scan quantization and whether it is linear or log, is irrelevant. You can not get 17 linear stops out of the film, especially with a 10 bit scan.

Firstly, we shall evaluate is 1% JND (which is a reasonable assumption many times, except some) a good assumption for film negative scan quantization step? With 1% JND the required minimum quantization step in density scan is 0.0026. Assuming 0.6 gamma on film negative and standard base density (0.19 density = 95 code value), one would get a range of 1.86 density. 1.86/928 = 0.002, which is less than 0.0026, so we are fine.

Now lets calculate the number of linear stops required for a reasonable range that may be captured on a negative. Assume that the brightest object in our scene is 1000 times brighter than the lowest, that would map to a negative scan code value of 995, which is under 1023 (10 bit range), so we are ok. However, with 1% JND and a brightness range of 1000, it turns out that you need 17 stops to cover that range at least as well as the film (assuming that the least bright object gets mapped to code value 1).