Canon 5Ds Dynamic range--any accurate information not just speculation??

[Bart_van_der_Wolf]

Not necessarily 100% representative (because from a pre-production model), but according to RawDigger I get the following suggested values based on what's assumed to be masked pixels that only exhibit read-noise:

READ-NOISE (StdDev) @ ISO 100, before Black-Point subtraction:
R  = 5.82 , G  = 5.85 , B  = 5.77 , G2= 5.83 .
Saturation (14-bit) is presumably at 14733, although one color plane has an outlier of 15441, and an un-clipped black-point of on average 2047.

I just had a look at an ISO 50 shot, and that gave:
READ-NOISE (StdDev) @ ISO 100, before Black-Point subtraction:
R =5.45 , G = 5.53 , B = 5.43 , G2 = 5.44
Saturation (14-bit) is presumably at 14733 (all 4 color-planes clip at that level), with the same Black-Point level at 2047.

That would give an average engineering dynamic range of Log(14733/5.46)/Log(2) = 11.40 stops (screen mode in DxOMark terminology), 0.09 up from ISO 100. So there seems to be a slight benefit for e.g. studio users with powerful flash, wider aperture and marginally better noise and DR performance. I do not know if the change in ISO setting induced gain leads to half 'sensitivity'. To determine the exact ISO delta, it would be required to take a series of exposure pairs at various ISO settings.

In the attached histogram of the Read-Noise sample, there seems to be a 5 ADU gap which suggests some sort of odd amplification, but it is not present at actual exposure levels. I'm not yet sure what to make of that. Maybe something to do with the masked photosites.

I also notice a slight difference in G1 versus G2 levels, like in other Canon cameras, so maybe that's intentional. It does mean that a Raw converter should treat the 4 color planes as individual planes, and not group the G planes together.

Cheers,
Bart

[Paul2660]

So far the files look very good, and for the price point for 50MP, if I had to shoot a simple bracket, so be it.  I do hope that this will add some sensibility to other 50MP cameras on the market in regards to base price.

I still wonder why both Canon and Nikon cannot add a moveable screen to these wonderful cameras.  Not all of us have the flexibility of a 20 year old anymore and the advantages of a moveable screen are huge. 

These cameras will do very well indeed.

Paul

[Bart_van_der_Wolf]

I still wonder why both Canon and Nikon cannot add a moveable screen to these wonderful cameras.  Not all of us have the flexibility of a 20 year old anymore and the advantages of a moveable screen are huge.

Hi Paul,

I agree that an articulated LCD would be very welcome. I also frequently shoot from different shooting heights, and it wears my trousers out when I need to use a hoodman loupe for focusing in Live View close to the ground. That's why I use a tethered (or wireless) Tablet with Helicon Focus for critical focusing.

Cheers,
Bart

[LKaven]

Hi Bart,

Not that it makes much of a difference, but screen mode in DxOMark terminology refers to the clipping raw value (12686 DN at ISO100) divided by the signal at which SNR = 1.  Assuming a similar fwc as the 7DII and using your values above for read noise, the signal at which SNR=1 works out to around 6.5 DN.  So 'screen' DR at ISO 100 should be about 10.93 stops.  Double check my math, given recent history

Noting that Bill Claff puts Photographic DR at 9.84 stops at ISO 100, I think DxO "screen DR" seems likely to come in higher than your estimate.  [The D810 comes in at 11.02 stops PDR at ISO 100, for comparison.]

[Jack Hogan]

READ-NOISE (StdDev) @ ISO 100, before Black-Point subtraction:
R =5.45 , G = 5.53 , B = 5.43 , G2 = 5.44
Saturation (14-bit) is presumably at 14733 (all 4 color-planes clip at that level), with the same Black-Point level at 2047.

That would give an average engineering dynamic range of Log(14733/5.46)/Log(2) = 11.40 stops (screen mode in DxOMark terminology)

Hi Bart,

Not that it makes much of a difference, but screen mode in DxOMark terminology refers to the clipping raw value (12686 DN at ISO100) divided by the signal at which SNR = 1.  Assuming a similar fwc as the 7DII and using your values above for read noise, the signal at which SNR=1 works out to around 5.68 DN.  So 'screen' DR at ISO 100 should be about 11.12 stops.  Double check my math, given recent history

Jack

[Jack Hogan]

In the attached histogram of the Read-Noise sample, there seems to be a 5 ADU gap which suggests some sort of odd amplification, but it is not present at actual exposure levels. I'm not yet sure what to make of that. Maybe something to do with the masked photosites.

I see it in the histogram of the actual image as well.  In the Nikon world this is quite common and known as prescaling or pre-conditioning.

[AlterEgo]

Not all of us have the flexibility of a 20 year old anymore

apparenly elderly japanese males are still quite slim and agile   

[Bart_van_der_Wolf]

Hi Bart,

Not that it makes much of a difference, but screen mode in DxOMark terminology refers to the clipping raw value (12686 DN at ISO100) divided by the signal at which SNR = 1.  Assuming a similar fwc as the 7DII and using your values above for read noise, the signal at which SNR=1 works out to around 5.68 DN.  So 'screen' DR at ISO 100 should be about 11.12 stops.  Double check my math, given recent history

Hi Jack,

When I measure an EOS 7D Mark II file (ISO 100 is the lowest I have found) the same way I did the 5DSR one, I get:
R = 5.52 , G = 5.48 , B = 5.46 , G2 = 5.40, with a clipping level of  13583 (BP = 2047).
That would produce Log(13583/5.47)/Log(2) = 11.28 stops, DxOMark reports 11.11 stops.
If I subtract the black point and keep the read noise I'd get Log(11536/5.47)/Log(2) = 11.04 stop, so that would be closer.

When I measure an EOS 5D Mark III file (ISO 50) the same way I did the 5DSR one, I get:
R = 6.32 , G = 6.27 , B = 6.13 , G2 = 6.21, with a clipping level of  15283.
That would produce Log(15283/6.23)/Log(2) = 11.26 stops, DxOMark reports 10.97,
If I subtract the black point and keep the read noise I'd get Log(13236/6.23)/Log(2) = 11.05 stop, so that would be closer.

In the same vein, with BP subtracted, the 5DS R would come out as 11.18 when using my quick and dirty method, a bit above both predecessors.

So you're right that I'm overstating the value that DxOMark will probably report as Screen DR, thanks for pointing that out. Quick and dirty estimation is just that, quick and dirty.

It still means that the 5DS R has probably (we'll see for sure when DxO gets a production model to test)  a similar, or even a bit better DR than both, with a smaller sensel pitch. I've seen some samples that Iliah Borg processed of ISO 6400 shots with his Rawconverter, and they even looked usable compared to others. Not that I would likely choose ISO 6400 on the 5DS/5DS R, It probably tops out at 1250 or 1600 and then further pushing in postprocessing is better, but still.

Cheers,
Bart

[LKaven]

Jack, did I go into a timewarp?  By all appearances, I replied to your message before you wrote it.  How did that happen?

[Bart_van_der_Wolf]

Noting that Bill Claff puts Photographic DR at 9.84 stops at ISO 100, I think DxO "screen DR" seems likely to come in higher than your estimate.  [The D810 comes in at 11.02 stops PDR at ISO 100, for comparison.]

Bill Claff uses a higher ('photographic') noise level as basis, which obviously is a bit arbitrary, so his DR numbers will always be lower. Some people accept a higher noise level as minimum, others lower levels. I prefer to use Engineering DR (lowest possible) if solid data (requires subtracting equally exposed image pairs) is available. Lowest possible noise levels make sense if e.g. software is used that can average multiple exposures, or apply effective noise reduction.

Cheers,
Bart

[LKaven]

Bill Claff uses a higher ('photographic') noise level as basis, which obviously is a bit arbitrary, so his DR numbers will always be lower. Some people accept a higher noise level as minimum, others lower levels. I prefer to use Engineering DR (lowest possible) if solid data (requires subtracting equally exposed image pairs) is available. Lowest possible noise levels make sense if e.g. software is used that can average multiple exposures, or apply effective noise reduction.

All fine.  But I think this would still suggest that the DxO "print" DR for the 5Ds would be higher than Jack calculated. 

[Bart_van_der_Wolf]

All fine.  But I think this would still suggest that the DxO "print" DR for the 5Ds would be higher than Jack calculated. 

That's possible, but I'll wait for DxOMark to come up with their real numbers. So far things look slightly better than anticipated. Very usable (due to low pattern noise) but not earth shocking. It's as announced.

Cheers,
Bart

[Jack Hogan]

Jack, did I go into a timewarp?  By all appearances, I replied to your message before you wrote it.  How did that happen?

Luke, it's all about black holes and quantum mechanics

As far as the estimates are concerned I agree with Bart, for consistency and comparability it's best to wait for DxO to do their thing, and you are probably right that the 5DS' pixel is a little better than the 7DII's.  As far as PDR is concerned, keep in mind that Bill's threshold for the denominator shifts the weight to shot noise vs read noise, compared to DxO.

Jack

[E.J. Peiker]

DPReview has added a page of tests and discussion of the 5Ds with regard to dynamic range:
http://www.dpreview.com/previews/canon-eos-5ds-sr/7

In summary, a slight improvement but not up to the level of cameras equipped with Sony sensors.  They attribute the difference to off chip A/D converters on the Canon cameras vs integrated on chip A/D coverters on Sony sensored cameras.

[dwswager]

In theory yes but often the need to composite or HDR an image is because it falls outside the DR capabilities of any photographic sensor currently available. Sure we all want more of everything but wanting a Sony sensor solely for its DR just (for me) became obsolete if all else remains unchanged (36Mpix and 13 stops DR) on the new A7rII as the 5DsR files I just downloaded look amazing.

My point wasn't that a D810 for example would cover the DR of every situation.  The point was that having 14 stops is better than 11 stops even when one is forced to make a single capture of a scene with a wider DR than 14 stops.  Those extra 3 stops allow you to capture more of the scene DR even if your intended output won't hold that range.  It allows more options in post processing.  Note that we are talking about a very small number of photo opportunities where DR exceeds even 10 stops!

[dwswager]

Indeed but the main criticism I've read why people think the 5Ds is a fail is all down to DR even before they have seen (until now) RAW files that demonstrate what the camera can do. Sure, some people don't like the effort of using polarisers or grad filters to tame contrast and only recognise a good camera as one which can be pushed and pulled in post but I see no problem now with the 11.8 stop DR of the 5Ds. Maybe the Sony will be 56Mpix, even more DR and better colour and have a greater pixel level detail rendering this topic irrelevant but until that day I'm still very interested in the Canon.

I don't see it as a fail, but it is a real world limitation and makes the price not quite as attractive.  A big portion of the audience for a 50MP camera are landscape guys which is the same audience for extended DR cameras!  I'm biased to my D810 because I already own it, but if I was deciding between a 36MP, 14 Stop, $3000 camera, a 50MP, 11 Stop, $5000 camera just doesn't seem like such a great trade for my needs.  If I shot in a studio or something, then trading some DR and cash for MPs might be a no brainer.

[jjj]

Note that we are talking about a very small number of photo opportunities where DR exceeds even 10 stops!

Not as small as you think. Backlit shots are pretty common as are low light shooting where the light sources in shot will clip, just like sunsets and dappled light through trees is particularly hard work. DR being outside my cameras ability seems quite common to me.

[Bart_van_der_Wolf]

Not as small as you think. Backlit shots are pretty common as are low light shooting where the light sources in shot will clip, just like sunsets and dappled light through trees is particularly hard work. DR being outside my cameras ability seems quite common to me.

The real question then becomes, "Is that a problem?"

With film I used fill-flash if feasible, with digital I use a bit of shadow recovery, but if the exposure makes the shadows dark, then they probably should be dark. If the exposure blows out the background, maybe it should blow out the background (otherwise one loses the sense of the actual lighting conditions).

It is my impression that many of the purported noise problems (in general) come from over-correction of shadows, or lack of (possibility to adjust) technique. Also, when the sensor noise is well-behaved (no banding or pattern noise), noise reduction software is very effective, in those cases that shadows do benefit from more lifting. In many cases, bracketing is a very good option.

I wouldn't mind a sensor that is capable of huge dynamic range, but underexposure still means that fewer photons made it to contribute (color is already perceptually reduced in shadows) to the image. If at all possible, better technique directly provides more solid image quality.

Maybe Canon prefers a more professional type of use for the studio version of the 5 series, where lighting technique is a skill that the users do master? That's probably also why they capped the highest ISO setting to 6400, with an optional digital extension to 12800.

Granted, they presumably also do not have the chip design for on sensor amplification, but that may be the result of a deliberate choice made earlier. They concentrated more on motion/video design, and other technology. Their research budget is still immense, so it's not due to their lack of knowledge, or lack of filed patents (they rank in the very top in the USA for more than a decade already). Maybe they also anticipated a decline in the potential market size, and reassigned budget to where it made more sense (like lenses, in anticipation of their upcoming high MP offerings)?

I think they are glacially slow in the perception of brand loyal users, but they are probably also very clever in the long run.

Cheers,
Bart

[DeanChriss]

All else being equal everyone would choose a camera with more DR. It would be a black and white decision if not for the fact that all else is seldom equal. Questions like the following show the many shades of gray involved in such a choice. If your success rate in photographing wildlife went up considerably due to the speed and flexibility of a 200-400mm zoom with an instantly switchable 1.4x teleconverter, do you give it up for more DR? Will you miss more shots due to lack of DR or spending time swapping a teleconverter in and out? Different systems don't have all of the same lenses, and when they do the quality is often appreciably different. New sensors come along frequently relative to lens updates. Would you rather wait for a new sensor/camera model or for an update of a given lens that's lacking compared to what you already have? If you have $20-$30K invested in lenses and other brand specific equipment, how much DR difference is needed to justify switching versus waiting? There are lots of similar questions and there is no correct answer to any of them. They don't all apply to everyone and answers to the ones that do depend on everything from the type and style of photography you do, how often you run into high DR situations that you can't handle, and your budget. No single company has the best image sensor, widest selection of best quality lenses, best feature set, best service and support, and best everything else. The best system really depends on your individual needs and DR is one important consideration among many. Fortunately it has become difficult to put together a system that is incapable of capturing outstanding photographs, but it is still important to know the strengths and minimize the weaknesses of any system and use them accordingly.

[dwswager]

The real question then becomes, "Is that a problem?"

With film I used fill-flash if feasible, with digital I use a bit of shadow recovery, but if the exposure makes the shadows dark, then they probably should be dark. If the exposure blows out the background, maybe it should blow out the background (otherwise one loses the sense of the actual lighting conditions).

It is my impression that many of the purported noise problems (in general) come from over-correction of shadows, or lack of (possibility to adjust) technique. Also, when the sensor noise is well-behaved (no banding or pattern noise), noise reduction software is very effective, in those cases that shadows do benefit from more lifting. In many cases, bracketing is a very good option.

I wouldn't mind a sensor that is capable of huge dynamic range, but underexposure still means that fewer photons made it to contribute (color is already perceptually reduced in shadows) to the image. If at all possible, better technique directly provides more solid image quality.

There is a lot of good in your post.  When I first got the D810, I forgot that while fill light is a good way to reduce DR to fit the sensor (since I rarely needed it), it is also good to flatter the subject by filling in shadows, defining shapes and showing texture.  Some times a little fill flash or reflected light makes the difference.  I got enamored with the DR of the D810, but it only took a few weeks of spending tons of time in post fixing things to remind me.

And Dean, that is why I added "all else equal".  In your case, you are trading something in the body for effectiveness and flexibility of a particular lens.  Each of us has to decide whether a trade is worth it or not.  Since all the lenses I need or would buy (except maybe the T/S lenses) are available in for Nikon, I don't need to make that trade.

All I know is after switching to Nikon at the end of the film era, and riding intro to digital with them, when they lagged, in the sensor area to Canon, Nikon shooters didn't defend them.  They pissed, and moaned and complained and switched to other brands and basically forced Nikon to fix the problem.  I just think it is more likely Canon would fix their sensor issues if their user's weren't quite so 'forgiving'!