5D3 same max dynamic range as the 5D2???

[te4o]

And yes I'm outta here too. I've defended myself enough. It wastes too much time and I'm tired or arguing and if all people want to do is attack and throw accusations and with 30 smites for each applause why the heck should I ever bother posting about anything anyway, or at the least why bother going on and on in defense and offering explanations. If I do post more I will simply post and be done with it and not respond and everyone can say whatever they want after that .
[/quote]

@LTRLI (too long to spell) I applauded you for your open mind and for the boldness you have to oppose so many fans. You explained in good detail what you found and your conclusions are logical. I understand them as a warning against too high expectations for the new sensor at low ISO. However, giving an opinion about this sensor without a single production camera available, without the proper final RAW software, and without using them both yourself makes this just an approximation and a topic of interest for further research. Nothing more.
Good to learn about "masking area" though. I'd love to see your results from the lens-cap-black-frame-test.
Don't be disappointed by opposition, Nietzsche says "there are many disappointed good 'men' standing in the way of progress..."
To all the rest, I think people like LTRLI are helpful to increase our own personal DR .

 

[sarangiman]

Heck I gave you all +1's! I think all the arguing back & forth is great, as there is a lot to learn/understand! As long as we're civil about it

I have a # of questions, especially for LTRLI, neuroanatomist, & jrista:

(1) If Canon claims quantum efficiency is up due to better microlens design, and if initial tests show better ISO performance for the MkIII vs the MkII with RAW files (i.e. no influence of noise reduction, hopefully), then doesn't all that mean that SNR is necessarily higher? And doesn't higher SNR necessarily mean greater DR, for the same general sensor design (assuming nothing actually got *worse* btwn the MkIII & MkII)? Am I missing something here?

(2) Let's say those masked pixels really are sealed/blocked off from light OR are somehow turned off. However, let's say they're still being read by the electronics & so they can still be an indication of dark noise. Then, you use:

Equation 1: DR = log(base 2) [highest signal/stdev]

But for that calculated DR to be indicative of the max possible DR of the scene being recorded -- doesn't that presuppose a linear relationship between incoming light & recorded signal in the RAW file? Do we know this to be true?

(3) Related to question (2)... Has anyone actually established that Equation 1 is a standard, accepted, measure of DR? DXO says "Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture"... note it doesn't say "... a sensor records". Again, assuming a linear relationship between incoming light & signal recorded (which is not true, for example, for film) in the RAW file, I guess this'd be the same. But but from DXO's testing methodology description (http://www.dxomark.com/index.php/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range), I thought they were using actual light source/ND filter combinations to measure the SNR & DR.

(4) I must say I find it strange that in DXO's methodology in the link above, they state that they use a setup "in order to test across a dynamic range of 4 density steps (= 13.3 f-stops — a dynamic range much greater than today’s digital cameras)"... but then report that the Nikon D7000 has 13.9 stops of DR. Is this just an outdated description?

Thanks in advance!

 

[sarangiman]

I went back to the book 'Image Sensors and Signal Processing for Digital Still Cameras' and it speaks of the 'optical black' (OB) pixels around the active pixels that are necessary to determine a proper black level. I'm assuming that, by whatever mechanism, they have no light gathering capability (b/c, as neuroanatomist mentioned, light from the lens can fall on those pixels).

So, if those OB pixels are used to determine the black level, they may also reflect the noise generated by the read event & subsequent electronics. It is, however, curious that they're around 1024 for the 5D II & 2048 for the 5D III. That's highly suspicious... I'd love to know what Canon is doing there. Isn't that # roughly what the black level is set to? If so, I don't think you'd want to be setting the black level to ~2048... seems to me that'd lower DR... unless the blackest pixel in the Active Pixel area really has a lowest level of 2048 (seems highly unlikely).

I have to go back to dcraw & see what it sets as default black level for my 5D II.

Still, my other problem with this methodology is that recorded signal in the RAW file does not necessarily reflect input signal until someone shows this relationship is completely linear... at least that's what I think.

 

[Ricku]

I saw this chart in another thread, and it looks like the D800 is the new DR king for sure.


The 5D3 and D800 are pretty much comparable, but only at ISO800 and above.. Below ISO 800, the D800's DR is heading sky high, setting a new record.

 

[torger]

Numbers are hard to interpret, there are so many factors that don't show in a number. No measurement so far I've seen take noise quality (pattern etc) into account, which has a huge impact on subjective DR.

Even if 5Dmk2 and 3 measure the same, if pattern noise is less than usable DR will be better.

High ISO measurements often miss out on quantum efficiency and can be misleading due to that.

A test that I would like to see is mk2 and mk3 shooting the exact same test scene perfectly ETTR, and then push a shadow area (preferably containing a color checker) 3 stops and show the crops side by side, resolution is almost the same so no sacling is required. I suspect/hope that the mk3 will fair much better in such a test than these measurements indicate.

 

[DavidRiesenberg]

I saw this chart in another thread, and it looks like the D800 is the new DR king for sure.


The 5D3 and D800 are pretty much comparable, but only at ISO800 and above.. Below ISO 800, the D800's DR is heading sky high, setting a new record.

If this chart is true and not a result of some miscalculation, that would indicate that the D800 applies some sort of artificial curve to balance the shadows and highlights. An aggressive HTP sort of thing.

 

[sarangiman]

High ISO measurements often miss out on quantum efficiency and can be misleading due to that.

Why is that (mechanistically)?

A test that I would like to see is mk2 and mk3 shooting the exact same test scene perfectly ETTR, and then push a shadow area (preferably containing a color checker) 3 stops and show the crops side by side

Why can't you do that with the RAW files posted for both cameras on Imaging Resource right now yourself? There's a Color Checker in those shots.

If this chart is true and not a result of some miscalculation, that would indicate that the D800 applies some sort of artificial curve to balance the shadows and highlights. An aggressive HTP sort of thing.

HTP purposefully underexposes the shots (physically), & then applies a non-linear curve. If the methodology used in generating that graph for the D800 is similar to the method LTRLI used, then any sort of non-linear manipulation done after the exposure would not change the resulting calculated DR (b/c the end points of any non-linear curve would likely still be the same as for the linear curve, since there's no reason to change the absolute values of saturation or black). So I don't think that explains it.

 

[torger]

Waiting for raw support in my raw processor before I can do investigations myself.

 

[KeithR]

This is, after all, why many photogs embrace HDR.

Exactly my point - you don't need the camera to provide 6 stops of adjustment.

And I'll reiterate: how much you can get out of the shadows depends massively on the choice of converter, and some people just aren't prepared to put the effort in to find the best converter for all their needs.

If you need deep shadow recovery without banding, use a converter that can do it. They exist.

It's a damn' sight cheaper than jumping ship, and a damn' sight less stressful than the photographer wasting his life away waiting for Canon to produce the perfect camera.

Maybe I'm unusual in this, but I actually enjoy the challenge of finding solutions to problems: it's a far more productive way to use my time than endlessly complaining about this thing or that on internet forums.

 

[KeithR]

The 5D3 and D800 are pretty much comparable, but only at ISO800 and above.. Below ISO 800, the D800's DR is heading sky high, setting a new record.

Bill Claff, whose site this is from, is a long-time fervent Nikon guy; and these graphs are based on estimated values for both the D800 and the 5D Mk III.

Big pinch of salt in order here...

 

[bvukich]

The 5D3 and D800 are pretty much comparable, but only at ISO800 and above.. Below ISO 800, the D800's DR is heading sky high, setting a new record.

Bill Claff, whose site this is from, is a long-time fervent Nikon guy; and these graphs are based on estimated values for both the D800 and the 5D Mk III.

Big pinch of salt in order here...

Why is everyone getting all worked up about some guy's GUESS about sensor performance?

 

[3kramd5]

Question for those who may know.

Do these test (the OP's method, those in links or DXOlabs' stuff) include sensor level processing?

The Sony A580 and Nikon D7000 use the same sensor, yet the Nikon is evidently capable of an extra .6 stops of DR. I assume that's due to Nikon's Active D Lighting processing, but maybe it's due to generally better front-side optics.

Any ideas?

http://snapsort.com/compare/Nikon_D7000-vs-Sony_Alpha_DSLR-A580

 

[nightbreath]

Question for those who may know.

Do these test (the OP's method, those in links or DXOlabs' stuff) include sensor level processing?

The Sony A580 and Nikon D7000 use the same sensor, yet the Nikon is evidently capable of an extra .6 stops of DR. I assume that's due to Nikon's Active D Lighting processing, but maybe it's due to generally better front-side optics.

Any ideas?

http://snapsort.com/compare/Nikon_D7000-vs-Sony_Alpha_DSLR-A580

It might be an improved algorithm in ADC or the optics. I never saw how DxO measures DR of a specific camera and I doubt they use the same lens for different brands.

 

[Ricku]

If you need deep shadow recovery without banding, use a converter that can do it. They exist.

It's a damn' sight cheaper than jumping ship, and a damn' sight less stressful than the photographer wasting his life away waiting for Canon to produce the perfect camera.

Interesting.

Please name one. I'm using LR3 right now. Soon LR4, but if there is something better out there for cleaner shadow recovery, I'd like to know.

 

[Stvx]

Wow.
Some guys here sound like Public relation/marketing/lawyers from Canon.

I am actually amazed that this thread is still going on

I started a thread one week before the 5DMkIII was launched, about the sensor technology (was it a fine tuning of the existing MkII model or something completely new):

My thread was simply deleted by the "administrator" after 5 minutes...

You can't obviously be too critical with Canon products on a site supporting the brand. If you want a more critical approach i guess you need to go on some other sites. ;D

 

[jrista]

Some small notes about Sony sensors that may affect the discussion. First, and importantly: They have a variety of hardware-level built-in noise reduction technology. Two key types, if I remember correctly, as well as a couple other features that improve their sensors. One to pretty much eliminate fixed pattern noise by accounting for differences in transistors at the hardware level. Another that uses an extra accumulator that measures dark current noise before computing the charge accumulated in the photodiode itself (CDS, or correlated double sampling). In addition to these explicit noise reduction technologies, Sony Exmor sensors also use column-parallel ADC. High frequency electronics have the potential to introduce more noise. Sony's CPADC, being hyperparallel, means each ADC has to do less work, and therefor does not need to operate as fast or on as many pixels as ADC's in Canon cameras. This also leads to lower noise. They use copper wiring which allows the CMOS to be thinner, along with a double layer of microlensing (which appear to be gapless) to direct more light into each well. Sony holds patents for all of these technologies, as well as for an approach to backlit sensor fabrication (which does not seem to be used in Exmor sensors, not sure why.)

For one, Canon does not seem to have patents on similar technology in most cases. They do offer some hardware-level noise reduction in line with Sony's CDS, or correlated double sampling, to eliminate dark current noise before reading the photodiode charge. Canon seems to use a different approach to fixed pattern noise however, whereby they set a non-zero pixel level floor, such as 1024 (or possibly in the case of the 5D III, 2048). This information seems to be derived from masked off pixel borders around the edge of the sensor, and can be used to adjust the black point of a CR2 RAW image at time of processing. The non-zero floor allows both positive and negative adjustments to black level, depending on how much things like fixed pattern noise exhibit. However, at least based on information I've been able to find so far, this non-zero floor does not seem to "push" the maximum saturation value up at all. As far as I can tell, this leads to a fixed limitation on dynamic range since your eating into the maximum saturation level with a higher-than-zero black level...and may very well be the reason Canon sensors have been stuck below 12 stops of DR.

All that aside, Canon has made explicit claims about improving the dynamic range of their new sensors. From the Canon EU site (http://cpn.canon-europe.com/content/education/technical/inside_canon_eos_5d_mark_iii.do; see under Image Quality, Processing & Playback):

Like the EOS-1D X’s sensor the full-frame CMOS sensor of the EOS 5D Mark III offers increased light sensitivity, image noise reduction at higher ISO speeds, and a wider dynamic range.

The above is from a caption. They also provide the below information:

CMOS sensors make use of microlenses to direct the light into each pixel well. Canon’s first full-frame sensor with a gap-less microlens design is found in the EOS-1D X and the sensor in the EOS 5D Mark III features the same technology for improved light gathering ability. Gap-less microlenses mean that no matter what angle the light arrives at the sensor from it is directed into a pixel well where it can be used most effectively. In essence it means that no light that gets to the sensor is wasted by not making it into a pixel.

Unless Canon truly does not have an issue getting caught in an outright lie (i.e. if their new sensors still don't achieve more DR than the 11.95 stops measured by DXO or 10.3 stops measured by DR), such a claim has to mean something has been improved with 1D X and 5D III dynamic range. (Or at least, so one would hope...)

 

[Loswr]

I started a thread one week before the 5DMkIII was launched, about the sensor technology (was it a fine tuning of the existing MkII model or something completely new):

My thread was simply deleted by the "administrator" after 5 minutes...

You can't obviously be too critical with Canon products on a site supporting the brand. If you want a more critical approach i guess you need to go on some other sites.

No, you can be quite critical here, that's not the issue. What leads to threads being deleted is when they turn into a flame war instead of a civil discussion.

Also, there was a period a little before the 5DIII was announced when a bunch of people started threads of a very similar nature about the new model, and the mods went through and merged them into just a couple of threads to decrease clutter. If you look at those threads (some run to 20 pages) you can see that the post titles jump around a lot because many threads were merged into one main one. So, it's quite possible that your thread was not deleted, but simply 'disappeared' as a separate entity because it was merged with others.

 

[t.linn]

This is, after all, why many photogs embrace HDR.

Exactly my point - you don't need the camera to provide 6 stops of adjustment.

I think it is good to keep this point in mind. The fact that there is a work around—and with the HDR feature on the new 5D3 perhaps a relatively easy one—means that with some degree of extra effort whatever DR one needs is available to 5D3 shooters. (To be clear, when I refer to the "HDR feature" I refer to the fact that the 5D3 may be able to effectively capture a group of appropriately bracketed RAW images with minimal effort.)

And I'll reiterate: how much you can get out of the shadows depends massively on the choice of converter, and some people just aren't prepared to put the effort in to find the best converter for all their needs.

If you need deep shadow recovery without banding, use a converter that can do it. They exist.

The DR limits being calculated are irrespective of any RAW converter. So maybe there differences between RAW converters but lets be clear that the difference in performance between a Canon and Sony sensor is not about the RAW converter being used. It's a real, quantifiable difference.

Now this is not to say that one RAW converter won't handles shadows and banding better than another. However the choice of RAW converter is not simply about output; it's also about workflow. LR users have potentially invested thousands of hours creating a catalog to manage hundreds of thousands of images. Switching converters can be like switching camera systems. It's not something that can necessarily be done on a whim.

It's a damn' sight cheaper than jumping ship, and a damn' sight less stressful than the photographer wasting his life away waiting for Canon to produce the perfect camera.

Maybe I'm unusual in this, but I actually enjoy the challenge of finding solutions to problems: it's a far more productive way to use my time than endlessly complaining about this thing or that on internet forums.

Like many others involved in this discussion, I've got a 5D3 on pre-order; I'm not wasting my life waiting for a perfect camera. ;D And I don't consider most of the posts in this thread as complaining. It is threads like this on forums like this that raise awareness of issues with other users and even sometimes with Canon. Case in point: If you shoot video at all, you know that hundreds of threads like this related to the crippled video functionality of the 5D2 at its introduction (no manual control; a single odd frame rate) helped push Canon to make the changes necessary to turn the 5D2 into a professionally useful tool for video.

 

[LetTheRightLensIn]

Oh I really shouldn't bother posting anymore but this is reasonable talk so:

Heck I gave you all +1's! I think all the arguing back & forth is great, as there is a lot to learn/understand! As long as we're civil about it

I have a # of questions, especially for LTRLI, neuroanatomist, & jrista:

(1) If Canon claims quantum efficiency is up due to better microlens design, and if initial tests show better ISO performance for the MkIII vs the MkII with RAW files (i.e. no influence of noise reduction, hopefully), then doesn't all that mean that SNR is necessarily higher? And doesn't higher SNR necessarily mean greater DR, for the same general sensor design (assuming nothing actually got *worse* btwn the MkIII & MkII)? Am I missing something here?

It means that SNR would be better up in the realm where photon noise dominated read noise. But it does relatively little, in comparison, for the performance in the deep shadows compared to the read noise. So it can be that they have the same SNR in the deepest shadows near black but that one has better SNR say at semi-dark gray and brighter.

(2) Let's say those masked pixels really are sealed/blocked off from light OR are somehow turned off. However, let's say they're still being read by the electronics & so they can still be an indication of dark noise. Then, you use:

Equation 1: DR = log(base 2) [highest signal/stdev]

But for that calculated DR to be indicative of the max possible DR of the scene being recorded -- doesn't that presuppose a linear relationship between incoming light & recorded signal in the RAW file? Do we know this to be true?

It does and as far as I have scene there is and it is.

(3) Related to question (2)... Has anyone actually established that Equation 1 is a standard, accepted, measure of DR? DXO says "Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture"... note it doesn't say "... a sensor records".

I think they say records and captures because the read electronics can mess up what the sensor itself captured. I believe it has been shown that the 1D4 sensor, for instance, actually captures close to 14 stops but then the readout loses a few stops and that is what gets recorded in the RAW file and it's only 11.5 stops instead of almost 14.

(4) I must say I find it strange that in DXO's methodology in the link above, they state that they use a setup "in order to test across a dynamic range of 4 density steps (= 13.3 f-stops — a dynamic range much greater than today’s digital cameras)"... but then report that the Nikon D7000 has 13.9 stops of DR. Is this just an outdated description?

The 13.9 is only after they have normalized it to 8MPs. Although it is true that even the basic camera DR does come in 13.35 for the D7000 so I think that description of their setup may be out of date. Prior to the Exmor sensors it may have been valid (although maybe some of those old sigmas?).

 

[LetTheRightLensIn]

I went back to the book 'Image Sensors and Signal Processing for Digital Still Cameras' and it speaks of the 'optical black' (OB) pixels around the active pixels that are necessary to determine a proper black level. I'm assuming that, by whatever mechanism, they have no light gathering capability (b/c, as neuroanatomist mentioned, light from the lens can fall on those pixels).

So, if those OB pixels are used to determine the black level, they may also reflect the noise generated by the read event & subsequent electronics. It is, however, curious that they're around 1024 for the 5D II & 2048 for the 5D III. That's highly suspicious... I'd love to know what Canon is doing there. Isn't that # roughly what the black level is set to? If so, I don't think you'd want to be setting the black level to ~2048... seems to me that'd lower DR... unless the blackest pixel in the Active Pixel area really has a lowest level of 2048 (seems highly unlikely).

I have to go back to dcraw & see what it sets as default black level for my 5D II.

It's sort of arbitrary what they decided to use as the blackpoint in the RAW file and it probably makes sense to chose some power of 2 rather than say 1117 or 2003 or something.

Still, my other problem with this methodology is that recorded signal in the RAW file does not necessarily reflect input signal until someone shows this relationship is completely linear... at least that's what I think.

RAW files are said to be stored in linear fashion.