D750 Photographic Dynamic Range (PDR) Adjusted at PhotonsToPhotos

[bclaff]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thank you to all collaborators around the world for providing the required raw files for analysis over the past 10 years.
If you also want to help (with any camera model) I'd love to hear from you.

I'm highlighting this update in light of the recent release of the D850 since comparisons with the D750 are affected.

How the old and the new PDR values for the D750 compare:



And here's how the comparison between the D850 and D750 now looks :



Note that the maximum D750 PDR is no longer higher than the D850 (by 0.14 stops)
The D750 still acquits itself quite well; particularly below ISO 400 before dual conversion gain kicks in for the D850.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[lickity split]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thank you to all collaborators around the world for providing the required raw files for analysis over the past 10 years.
If you also want to help (with any camera model) I'd love to hear from you.

I'm highlighting this update in light of the recent release of the D850 since comparisons with the D750 are affected.

How the old and the new PDR values for the D750 compare:



And here's how the comparison between the D850 and D750 now looks :



Note that the maximum D750 PDR is no longer higher than the D850 (by 0.14 stops)
The D750 still acquits itself quite well; particularly below ISO 400 before dual conversion gain kicks in for the D850.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

LOL err so now were starting to move the goal post's around after 3+years so one camera looks better than the other ...

I always felt DPR's studio comparison was monkeyed around with in the same way . :-(

--
https://www.flickr.com/photos/lickitysplit11111/
https://500px.com/lickitysplit11111

 

[TOF guy]

Thank you for sharing the measurements with all of us :-D

 

[bclaff]

...

LOL err so now were starting to move the goal post's around after 3+years so one camera looks better than the other ...

I always felt DPR's studio comparison was monkeyed around with in the same way . :-(

As they say: "Just because you're paranoid, it doesn't mean they aren't out to get you."

 

[scokill]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thank you to all collaborators around the world for providing the required raw files for analysis over the past 10 years.
If you also want to help (with any camera model) I'd love to hear from you.

I'm highlighting this update in light of the recent release of the D850 since comparisons with the D750 are affected.

How the old and the new PDR values for the D750 compare:



And here's how the comparison between the D850 and D750 now looks :



Note that the maximum D750 PDR is no longer higher than the D850 (by 0.14 stops)
The D750 still acquits itself quite well; particularly below ISO 400 before dual conversion gain kicks in for the D850.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Than now puts the D(X)s at the top of the heap for higher ISO PDR.

 

[TN Args]

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).

LOL err so now were starting to move the goal post's around after 3+years so one camera looks better than the other ...

I always felt DPR's studio comparison was monkeyed around with in the same way . :-(

It's all water off a duck's back if we give measurements the appropriate, modest level of importance.

But if we start quoting slightly different numbers online as an argument that one camera is going to produce better IQ than another, then (a) we have lost track of reality, (b) we are going to make poor decisions (if we live by our words), and (c) we are going to get all uppity when they change and the carpet gets pulled out from under all our strutting and arguing.

Good on Bill for maintaining his technical honesty about his numbers, but if that change makes any difference to anybody, it would be a sad commentary on their perspective.

 

[msu79gt82]

... I've noticed that many folks trolling the D850 use the old D750 numbers as ammunition.

This is tongue in cheek of course, but seriously the trolls have often brought that up.

 

[TOF guy]

if we give measurements the appropriate, modest level of importance.

(snip) it would be a sad commentary on their perspective.

Well, I am going to make a sad commentary on my perspective, then But IMO these measurements are important.

We learn that the cameras have the same SNR performance, and their "noise" either at low or high ISO should not be a factor when picking one over the other.

We also learned that progress in SNR is stalled (for the obvious reason that it's about as good as it can get with Bayer filters) and any manufacturer claim that it's improved by 1 stop is about jpeg noise reduction.

And more importantly, these are objective measurements. Beats any time the I "see" more noise in here than in there any time of the day.

But maybe you meant something else ?

 

[TN Args]

if we give measurements the appropriate, modest level of importance.

(snip) it would be a sad commentary on their perspective.

Well, I am going to make a sad commentary on my perspective, then But IMO these measurements are important.

I never said they are unimportant i.e. to be ignored. I said they need to be weighted sensibly among other factors.

We learn that the cameras have the same SNR performance, and their "noise" either at low or high ISO should not be a factor when picking one over the other.

They previously/already were so close that the same conclusion should have been drawn.

We also learned that progress in SNR is stalled (for the obvious reason that it's about as good as it can get with Bayer filters) and any manufacturer claim that it's improved by 1 stop is about jpeg noise reduction.

You will fail to find a reference where Nikon said the SNR is one stop higher -- so you have built a straw man there. They said "image quality", and to conflate that two would be simple misrepresentation and argumentation. They also said JPEG and raw, clearly. But "image quality" means the overall effect of all changes, including for example increased resolution. So why can't they say that a D850 raw image at double the D810 ISO will "look just as good overall" as a D810 raw image at half the D850 ISO, if it actually does?

And more importantly, these are objective measurements. Beats any time the I "see" more noise in here than in there any time of the day.

It certainly does not, and any self-respecting scientist would support my view. Have you not noticed how noise in images from different cameras looks different, and one can react more negatively to one look than the other, even if the number is the same?

Technology experts, involved in technology that is ultimately judged by human senses for its success, understand that measurements are proxies, that give an indication of a possible human preference at the end, but that you really need to take it and move on to the actual human subjective evaluation to know for sure. Every single time.

 

[S t e f a n]

Oh nos! Well, it was good while it lasted. Just the other day I took some very nice (I thought, not knowing any better back then) pictures in a dark tunnel with my D750. (Somebody needs to download them to my laptop, and then put them up somewhere, so maybe I could share one.) But now...

Seriously, as one of the notorious D850 apologists here already pointed out, all these cameras, the D600/610, the 750, the 800/810s and the D850, are within a quarter stop or so from each other, which might just be the measurement error.

Update: one of the pics from yesterday afternoon.


End view of an accelerator magnet (Tevatron dipole)

I like this picture because it's something of technical significance that not many people get to see. You could argue that these Tevatron magnets are the grandfathers of all the MRI and other commercial superconducting magnets around the world. This is what started the superconducting wire and magnet industry.

I think the picture is fitting in this context since the grey metal makes it easy to see the color noise at the pixel level, yet at reasonable magnification the picture looks just fine. No adjustments/noise reduction, JPEG straight out of the camera. D750 with (freshly overhauled at APS) Nikkor 20-35/2.8 at 35 mm f/2.8 1/50s ISO 6400.

And another one, also mostly grey tones, at even slightly higher ISO:


A view of the DZero detector (south side muon pixel counter array)

I think the (low) level of noise at these ISO sensitivities is amazing.

 

[TOF guy]

We learn that the cameras have the same SNR performance, and their "noise" either at low or high ISO should not be a factor when picking one over the other.

They previously/already were so close that the same conclusion should have been drawn.

Do not quite agree. There was a distinct advantage to the D750 before.

We also learned that progress in SNR is stalled (for the obvious reason that it's about as good as it can get with Bayer filters) and any manufacturer claim that it's improved by 1 stop is about jpeg noise reduction.

You will fail to find a reference where Nikon said the SNR is one stop higher -- so you have built a straw man there. They said "image quality", and to conflate that two would be simple misrepresentation and argumentation.

Better IQ at higher iso means better snr. Anything else is "building a straw man".

They also said JPEG and raw, clearly.

Yes, but it cannot be raw.

But "image quality" means the overall effect of all changes, including for example increased resolution. So why can't they say that a D850 raw image at double the ISO will "look just as good overall" as a D810 raw image at half the ISO, if it actually does?

Because it does not

And more importantly, these are objective measurements. Beats any time the I "see" more noise in here than in there any time of the day.

It certainly does not, and any self-respecting scientist would support my view. Have you not noticed how noise in images from different cameras looks different, and one can react more negatively to one look than the other, even if the number is the same?

I've certainly noticed that people can convince themselves of the conclusions they like to reach . Does not seem very scientific to me.

Noise may appear different only because of different processing. Otherwise the noise contribution is basically the same, the Bayer filters the same +/- minor tweaks and it's a given that if Bill was to measure noise per channel he would find the same since it's the same electronics under the pixels of different colors.

Morale: I prefer objective measurements than conclusions that people reach unless they take specific precautions (like compare the images w/o knowing where they come from when they rank them).

 

[bclaff]

TOF guy wrote:

... Otherwise the noise contribution is basically the same, the Bayer filters the same +/- minor tweaks and it's a given that if Bill was to measure noise per channel he would find the same since it's the same electronics under the pixels of different colors.

...

Actually, I do measure each channel separately.
Although there are exceptions (PDAF pixels come to mind) the 4 channels generally have the same behavior (I normally average them).

Regards,

 

[syberman7]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thanks for clearing that up. It never made any sense to me how much better the D750 appeared to be than the D810 in your charts.

 

[TN Args]

We learn that the cameras have the same SNR performance, and their "noise" either at low or high ISO should not be a factor when picking one over the other.

They previously/already were so close that the same conclusion should have been drawn.

Do not quite agree. There was a distinct advantage to the D750 before.

Tiny. I would not like to think it would appear distinct in images, just because it appears distinct in a graph.

We also learned that progress in SNR is stalled (for the obvious reason that it's about as good as it can get with Bayer filters) and any manufacturer claim that it's improved by 1 stop is about jpeg noise reduction.

You will fail to find a reference where Nikon said the SNR is one stop higher -- so you have built a straw man there. They said "image quality", and to conflate that two would be simple misrepresentation and argumentation.

Better IQ at higher iso means better snr. Anything else is "building a straw man".

Sorry, but let's be crystal clear, since you are spraying mist everywhere.

Nikon said "image quality". That's what they said. QED.

For you to say, "oh, they actually mean SNR", and then shoot down SNR as impossible for raw files, that is the very definition of constructing a straw man argument. Look it up if you need to but don't debate it with me, please.

They also said JPEG and raw, clearly.

Yes, but it cannot be raw.

See above. Your straw man.

But "image quality" means the overall effect of all changes, including for example increased resolution. So why can't they say that a D850 raw image at double the ISO will "look just as good overall" as a D810 raw image at half the ISO, if it actually does?

Because it does not

See above. Your straw man.

And more importantly, these are objective measurements. Beats any time the I "see" more noise in here than in there any time of the day.

It certainly does not, and any self-respecting scientist would support my view. Have you not noticed how noise in images from different cameras looks different, and one can react more negatively to one look than the other, even if the number is the same?

I've certainly noticed that people can convince themselves of the conclusions they like to reach . Does not seem very scientific to me.

I see you deleted the extra paragraph of my post, which is unfortunate because it explains how science uses quantitave proxies to assist, but never replace, human perception, when developing a technology that that is ultimately judged by human senses -- photography being one example, and sound recording another.

Morale: I prefer objective measurements than conclusions that people reach unless they take specific precautions (like compare the images w/o knowing where they come from when they rank them).

Well, Nikon spoke about "image quality" and one cannot help noticing that one word is 'image', which you cannot assess without looking, and the other word is 'quality' which is by definition a relative word, i.e. qualitative not quantitative.

And we all know, I hope, that there are numerous factors that contribute to the totality that is "image quality" (and each of them in turn has several quantitative measurement proxies, yet further removed from 'image quality' but chosen because there is at least a tenuous link back to 'image quality' that makes them useful if not definitive). So, let's literally interpret the words Nikon said by literally interpreting image quality as perceived excellence. What they said literally means, "If I look at a high-ISO D810 image, the overall quality as a qualitative assessment is about the same as if I look at a D850 image taken at double the ISO". And the only way to refute that is with actual images, since there is a lot more than noise alone that differs between the images.

And in fact that statement, above, is getting a lot of verification from D850 owners who also have a D810. Even shooting raw.

 

[Joerg V]

Only those who never conducted scientific tests don't face the possibility of making a measurement error.

Thanks for your great graphs.

Thanks for your honesty about the mistake that happened. The way you handled that only increases my trust for you.

 

[L Copps]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thank you to all collaborators around the world for providing the required raw files for analysis over the past 10 years.
If you also want to help (with any camera model) I'd love to hear from you.

I'm highlighting this update in light of the recent release of the D850 since comparisons with the D750 are affected.

How the old and the new PDR values for the D750 compare:



And here's how the comparison between the D850 and D750 now looks :



Note that the maximum D750 PDR is no longer higher than the D850 (by 0.14 stops)
The D750 still acquits itself quite well; particularly below ISO 400 before dual conversion gain kicks in for the D850.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

What I don't understand about the curve in the second table is how the D750 can have lower dynamic range at 560 ISO than it does at 800 ISO.

--
“In my mind's eye, I visualize how a particular sight and feeling will appear on a print. It is an intuitive sense, an ability that comes from a lot of practice.” – Ansel Adams

 

[S t e f a n]

[...]

What I don't understand about the curve in the second table is how the D750 can have lower dynamic range at 560 ISO than it does at 800 ISO.

That's the D850, and ISO 800 supposedly is where the higher gain amplifier kicks in. It (re)gains more range at the high end than adding noise at the low end; but Bill should explain.

 

[j_photo]

Better IQ at higher iso means better snr. Anything else is "building a straw man".

Unless I'm misunderstanding what you are saying, I disagree. I believe there is more to image quality than just signal to noise ratio.

I appreciate Bill's charts and think they have a lot going for them. I'm not dismissive of them at all. But they don't necessarily tell the whole story. I just compared two cameras, Df and D800, that I own and regularly shoot at higher iso's. I'm certain that the Df at 12,800 produces better images than my D800 at 6400. But Bill's charts show them not nearly so far apart. I don't doubt Bill's data. I just think there must be more to IQ.

 

[Glack]

At PhotonsToPhotos sensor measurements are always "live".
When new data changes the results then new results are published.
Normally any change is slight and no one notices.

In March of 2017 I republished over 150 Photographic Dynamic Range (PDR) results.
Many values were slightly lower, but these slight changes went unnoticed.

I have determined that the previously published Nikon D750 PDR values were (noticeably) too high; mea culpa (my bad).
I have discovered an error in the initial data collection for PDR for that camera.
Normally PDR data collection errors result in dramatically low PDR values, the error is obvious, and the data rejected. That didn't happen in this case.
New data has been collected and processed.

Thank you to all collaborators around the world for providing the required raw files for analysis over the past 10 years.
If you also want to help (with any camera model) I'd love to hear from you.

I'm highlighting this update in light of the recent release of the D850 since comparisons with the D750 are affected.

How the old and the new PDR values for the D750 compare:



And here's how the comparison between the D850 and D750 now looks :



Note that the maximum D750 PDR is no longer higher than the D850 (by 0.14 stops)
The D750 still acquits itself quite well; particularly below ISO 400 before dual conversion gain kicks in for the D850.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Many thanks for this clarification!

I'm wondering why in the DXOMARK DR plot for D850 there is no this "zigzag", seen in your plot at ~ISO400?

 

[Mark Blauhoefer]

Interesting you should bring up sound recording. In the early days of CD experimentation, deciding on the optimum recording frequency beyond Nyquist to avoid audible aliasing (and arriving at 44.1 kHz), there was a noticeable disorienting over-clarity upon listening, with harmonics sounding too harsh. They introduced a small amount of white noise (less than 1%) to rectify the perceived flaw, and agreed it gave a more enjoyable experience. They've been adding that noise to CDs ever since