D800E -Truly Amazing Low-Light Performance

[bjanes]

NO, I have neither data nor graphs and I have no interest whatsoever in reading any whitepapers on the subject. It is merely my personal observation that the DR advantage of my D800e disappears quite quickly as ISO is increased. Once that advantage disappears it is my personal view that there are better handling and better featured cameras out there (e.g. D4, 5dIII, D1X, OMD, etc. etc.).

A quick look at the plethora of graphs that indignantly sprung up, yours included, at my first utterance of anything approaching criticism of the D800 seem to support this view (at least the first part of it), but what do I know.

Alistair,

If you actually have these cameras, why don't you publish some photos on the forum to back up your assertions regarding ISO or else cease and desist? Ray's results show quite the opposite to what you say.

Regards,

Bill

[hjulenissen]

In fact... yes.

As long as we use cameras, the technicalities thereof are relevant.

Most drivers don't feel that a torque curve/efficiency curves are important to their driving/wallet, but in fact... they are.

Cheers,
Bernard

It may well be possible to measure everything that is relevant, but surely not everything that can be measured is relevant. To continue your car analogy: one car may have 10 dB better FM radio sensitivity than another, but this has no influence on what car will win a race...

My question was how the 18% SNR measurements extends what can be learned from DR measurements in a way that is relevant for my/your/... photography.

...The SNR at 18% saturation is virtually the same for both cameras, which is not surprising since both cameras have the same sensor size and collect the same amount of light and have similar shot noise (which is the predominant noise contribution at this saturation level).

So could we just ditch the 18% SNR measurement and state that "bigger is better" (in this respect, at least)? Does it matter? I.e. would my A3+ prints appear visually better if my 7D had better SNR@18%, "everything else being equal"?

My uneducated gut-feeling is that noise (even worse: patterned noise) in the shadows are a much more relevant problem than shot-noise at mid-gray. But perhaps I am misinterpreting the measurements and their implications. If so, I would be pleased to be educated.

-h

[bjanes]

It's now clear to me that the D800E is not an ISO-less camera, and that shots at ISO 6400, for example, compared with the same exposure at ISO 100,  have more detail in the deep shadows, and more accurate color rendition.

The 5D3 may have better specs in some respects, than the D800E, but DR and SNR are not amongst them.

Merely by looking at the DR graphs, one cannot determine if the departure from linearity at low ISO is from impaired performance at base ISO or from improved performance at higher ISO. Marianne states that the latter is the case and the D800 is better than ISOless. This DPR thread amplifies the subject.

Regards,

Bill

[Ray]

It may well be possible to measure everything that is relevant, but surely not everything that can be measured is relevant. To continue your car analogy: one car may have 10 dB better FM radio sensitivity than another, but this has no influence on what car will win a race...

My question was how the 18% SNR measurements extends what can be learned from DR measurements in a way that is relevant for my/your/... photography.So could we just ditch the 18% SNR measurement and state that "bigger is better" (in this respect, at least)? Does it matter? I.e. would my A3+ prints appear visually better if my 7D had better SNR@18%, "everything else being equal"?

My uneducated gut-feeling is that noise (even worse: patterned noise) in the shadows are a much more relevant problem than shot-noise at mid-gray. But perhaps I am misinterpreting the measurements and their implications. If so, I would be pleased to be educated.

-h

I'm confused here, h. What's your point? As I understand, DR refers to SNR at extremely low signal levels, ie, the deep shadows. SNR at 18% refers to noise levels in the mid-tones which are relevantt for smooth skin tones. MFDB excels in this area.

As I understand, the DXOMark results for DR performance could be described as SNR at 1% to 0.1%, insted of 18%. Is there more to it than that?

[ErikKaffehr]

Hi,

18% SNR is what you probably would have on facial tone or deep blue sky. These are typically surfaces you would like to have smooth tonality.

With a wide dynamic range it is well possible that we never utilize the low end. Shadow detail is normally kept pretty dark and would be even more compressed in printing.

You can extend dynamic range downwards, but it does not help on mid tones.

Keep also in mind that DxO measurements are about the amount of noise and not the quality of the noise.

Best regards
Erik

My uneducated gut-feeling is that noise (even worse: patterned noise) in the shadows are a much more relevant problem than shot-noise at mid-gray. But perhaps I am misinterpreting the measurements and their implications. If so, I would be pleased to be educated.

-h

[hjulenissen]

I'm confused here, h. What's your point? As I understand, DR refers to SNR at extremely low signal levels, ie, the deep shadows. SNR at 18% refers to noise levels in the mid-tones which are relevantt for smooth skin tones. MFDB excels in this area.

As I understand, the DXOMark results for DR performance could be described as SNR at 1% to 0.1%, insted of 18%. Is there more to it than that?

I don't know. I wish that someone was able to convey this topic once and for all using some simple graph. Ala the placeholder included below (only designed by someone who knew the physics and engineering approximations):

[Ray]

Merely by looking at the DR graphs, one cannot determine if the departure from linearity at low ISO is from impaired performance at base ISO or from improved performance at higher ISO. Marianne states that the latter is the case and the D800 is better than ISOless. This DPR thread amplifies the subject.

Regards,

Bill

Does it matter? For the practical photographer, what matters is maximum DR and SNR at all ISOs. Currently, cameras which excel at high ISO, such as the D4 and the D3s, do not excel at low ISO to the same degree. There's a trade-off.

Cameras which excell at low ISO may not compete with cameras designed to to excell at high ISO. The D4 beats the D800 at high ISO, but the D800 beats the D4 at low ISO.

[bjanes]

I don't know. I wish that someone was able to convey this topic once and for all using some simple graph. Ala the placeholder included below (only designed by someone who knew the physics and engineering approximations):

The most complete description of noise in the DXO analysis is the full SNR plot. Emil Martinec explaines engineering DR and shows how to determine the photographic DR for any given noise floor. The full SNR plot may also be used for determination of the tonal range of a camera, which is the number of discrete gray levels that the sensor can discriminate, expressed as the number of bits necessary to encode these levels. In Michael's original ETTR post, he spoke of the number of levels in the brightest f/stop and assumed that this was related to the bit depth of the sensor. However, it does not make sense to quantify the sensor response in finer steps than the noise level.

Consider the full SNR plot for the D800e (shown below). In scientific papers it is common to plot the standard error as bars above and below the observed value to show the error of the measurement (see here for an example). We can do the same for the full SNR plot. The vertical arrowed bar is the standard deviation of the noise for a given pixel value. The true value of pixel should lie between the upper and lower bounds for this pixel level as shown by the green arrow. If one repeats the process from shadows to highlights and places the black arrows side by side, that is the number of discrete gray levels that the sensor can detect. This number may be expressed as the number of bits necessary to encode it.

Regards,

Bill

[Alistair]

Alistair,

If you actually have these cameras, why don't you publish some photos on the forum to back up your assertions regarding ISO or else cease and desist? Ray's results show quite the opposite to what you say.

Regards,

Bill

Well if you want proof of the low ISO performance of the D800e here are some from today; all OOC jpegs. As I am trying to point out, I neither need nor use high ISO and if I did I would not buy a D800/e. At this point I am going to take your desist option and bow out of the thread, this stuff is of no interest to me. Go well.

[langier]

After all the number talking and curve plotting, I'm for going out and simply shooting and enjoying the pleasant surprise of getting an image today that yesterday, the technology limitations didn't allow!

[KevinA]

Do you care to furnish some data to back up your assertions? The DXO data tell a different story. The SNR at 18% saturation is virtually the same for both cameras, which is not surprising since both cameras have the same sensor size and collect the same amount of light and have similar shot noise (which is the predominant noise contribution at this saturation level). For engineering DR using the print values to account for the differing pixel counts of the two cameras, the D4 has a slight advantage. Since the D800 has more pixels and each pixel contributes read noise it has slightly less engineering DR, since read noise is in the denominator for this determination. For photographic DR where read noise is less a factor, the values would be more comparable.

You might want to read this white paper by DXO.

I've said many a time I could not give a stuff about results that come from meter derived conclusions. I don't need data to back it up, just eyes. I would suggest you shoot them side by side and judge for yourself and not depend on third party graphs.

[hjulenissen]

I've said many a time I could not give a stuff about results that come from meter derived conclusions. I don't need data to back it up, just eyes. I would suggest you shoot them side by side and judge for yourself and not depend on third party graphs.

"Experience keeps a dear school, but fools will learn in no other"
-Franklin, Benjamin

There is no contradiction between doing good measurements and doing good side-by-sides.

-h

[jeremypayne]

I've said many a time I could not give a stuff about results that come from meter derived conclusions.

You have.

You read these threads in which say you have no interest, and then you declare your 'independence'.

Right on cue. 

[RFPhotography]

"Experience keeps a dear school, but fools will learn in no other"
-Franklin, Benjamin

There is no contradiction between doing good measurements and doing good side-by-sides.

-h

Not a contradiction but an over-reliance on measurements isn't good either. 

I haven't used the term in many months, having been chastised for it in the past, but this type of discussion is precisely where the term 'measurebator' comes from.  Test analysis from outlets like DxO Mark and others are fine and using that information as a part of the research into a new camera is fine but in the end few, if any, of us spend our time making images in a highly controlled lab setting.  Even a studio with 'controlled' lighting isn't as static as a lab. 

In addition, too many people tout the hype of the test numbers without inquiry.  DxO Mark's headline test numbers for the D800/e are terrific.  But those numbers are massaged.  How many people actually realise that?  I'd venture it's a minority.

Most of us are actual photographers.  I'd like to think that it's the relative minority that pore over images at sub-pixel level magnifications on a computer screen for hours on end looking for the most minute imperfection.  Or spend countless thousands of dollars on cameras that are capable of making spectacular images with immense amounts of data captured in a raw file to then simply post 720x540 pixel jpegs on the web.  Photography is a visual endeavour.  So while we may look at test results from 'reputable' outlets, in the end it is what we see with our eyes that tells the tale.  Clients don't ask what the SNR of a picture is.  Clients don't ask what the dynamic range of a camera is.  Clients don't ask if a sensor is ISO-less.  They couldn't care-less. 

[Slobodan Blagojevic]

... Clients don't ask what the SNR of a picture is.  Clients don't ask what the dynamic range of a camera is.  Clients don't ask if a sensor is ISO-less.  They couldn't care-less. 

Clients also do not know or care what f/stop or ISO is, but shouldn't you?

They don't because they assume you do. And you do because you learned it at some point. And learning it required measuring... yours, or someone's else.

And that's what we are doing here, in this thread.

[RFPhotography]

It will likely come as no surprise, Slobodan that I disagree.  There's a difference between the technical skill or knowledge required to get the desired result in the form of a photograph and measurebating or pixel-peeping.

I don't think it takes a full understanding of, for example, Emil's treatise on noise to really understand what noise in a digital image is and that less is generally better.  I don't think it takes an engineering degree to realise that, for the most part, today's cameras are better than ones that came before in terms of noise performance/characteristics and dynamic range, among other things.  It doesn't take measuring in a lab to determine that.  The original post in this discussion thread bears that out quite well. 

I don't give a flying fadoo whether a particular camera measures 10db or 100db.  I don't care what the actual measurement is because it's a meaningless number.  But I want to see what the difference is between the two cameras.  The number tells me that the second camera is inherently better - has a better SNR - but I also don't need to measure it to see it.  So no, I don't believe that learning necessarily involves measuring, at least not to the exclusion of anything else.  I don't believe the two are interdependent.

Contrary to what your comment may seem to indicate, I'm not opposed to lab tests; and said as much in my previous remarks.  I read them and take the information into account when I'm evaluating a new camera or lens.  But my point is there's more to it than just the lab numbers and that an over-reliance on the lab numbers alone isn't a good approach. 

[hjulenissen]

It will likely come as no surprise, Slobodan that I disagree.  There's a difference between the technical skill or knowledge required to get the desired result in the form of a photograph and measurebating or pixel-peeping.

I know from experience that my family thinks that using a DSLR or pointing your flash towards the ceiling is the equivalent of "pixel-peeping": they use their compact cameras in full auto, have the internal flash making for the "deer in headlights", but they focus on the content. So should we all do as them?

It is easy to claim that obsessing with DOF and scheimpflug and golden angle and such can stand in the way of capturing the great scenes that come and go. Perhaps it is so. I think it is fair to choose ones own approach to a hobby or profession. You may call it "measurebating" as much as you want, it simply does not make your arguments any more intelligent or relevant for your readers.

I don't think it takes a full understanding of, for example, Emil's treatise on noise to really understand what noise in a digital image is and that less is generally better.  I don't think it takes an engineering degree to realise that, for the most part, today's cameras are better than ones that came before in terms of noise performance/characteristics and dynamic range, among other things.  It doesn't take measuring in a lab to determine that.  The original post in this discussion thread bears that out quite well.  

I believe that there are numerous examples where "hands on" people have given advice such as "use ISO 160 instead of ISO 100 or ISO 200, as it is less noisy" or "this camera sensor have more headroom, but that camera sensor have more footroom", where it turns out that their practical "show me the images" testing was flawed, and a more physically motivated testing methology have shown that their advice was poor.

This thread indicates that among "dedicated photographers" on this blog, there is disagreement over what current camera models gives the best result at low light levels vs at high light levels. If people buy a D4 but could have had better results with a D800, is that not reason enough to welcome good, relevant measurements (and fair side-by-sides)? I am not defending poor measurements or misinterpreted measurements, nor am I defending unfair/irrelevant subjecive side-by-sides.

-h

[Slobodan Blagojevic]

... I don't think it takes an engineering degree to realise that, for the most part, today's cameras are better than ones that came before in terms of noise performance/characteristics and dynamic range, among other things.  It doesn't take measuring in a lab to determine that....

On the contrary, it is precisely thanks to the measurbators, pixelpeepers and lab measuring that "today's cameras are better than ones that came before."

So, say "thank god for measurbators," and if it is beyond your comprehension or interest, skip the thread and move on.

Complaining about measurbators in their own thread is as meaningful as attending a church service to argue about god existence.

[jeremypayne]

Contrary to what your comment may seem to indicate, I'm not opposed to lab tests; and said as much in my previous remarks.  I read them and take the information into account when I'm evaluating a new camera or lens.  But my point is there's more to it than just the lab numbers and that an over-reliance on the lab numbers alone isn't a good approach. 

I see ... you are just looking out for those of us who aren't quite as sophisticated as you are when it comes to avoiding such "over-reliance".

What a load of horse-shit!  That's a pretty condescending and self-aggrandizing position to take.

[RFPhotography]

Not really sure why you've taken my comment so personally, Jeremy.  There was nothing arrogant, self-aggrandising or condescending  intended in the statement.  What I was merely saying was that too great a leaning on one or the other approach to evaluating a camera, or lens, or flash, or whatever isn't the best approach.  To read that as my 'looking out for' anyone is beyond absurd.  There's not a shred of logic in such an interpretation.  Agree or disagree all you want but if you're going to take such personal affront to my stating my opinion, well then fuck you.

Slobodan, I'm sorry but that's nonsense.  You're claiming that sites like DPReview or DxO Mark are the only reason there has been innovation in the camera market?  That's like saying Ralph Nader is the only reason there's been innovation in cars.  It's preposterous.  You're saying that companies won't innovate on their own?  Come on.  It's not beyond my comprehension.  Nor is it entirely without interest.  For crying out loud read what's actually written. 

h, no that's not what I'm saying.  And again read what I said about the technical knowledge to take a good picture being different from the knowledge of the inner workings of a camera.  I don't need to know how precisely a rotary engine works to be able to drive a Mazda.  WRT the 'use ISO 160' or whatever other similar message, again, I'm not opposed to data.  If someone can post accurate data that refutes a subjective statement then that's fine.  I think we both know that there's ample 'bad data' out there too.  This article and the referenced video on Vimeo stirred up quite a bit of controversy a while back.  Good data or bad data?  But it doesn't need to be a doctoral dissertation on the subject to do that.  As far as 'better' results with a D4 or D800, better is a wildly subjective term.  What is 'better' in one instance for one person may not be for another.  Yes, I do think that at times people can get caught up in the details and lose good pictures as a result.  It's also true; however, that technically perfect pictures aren't always the best either.