Any thoughts on how the 5d3 will compare on dxo mark to the Nikon D800?

[LetTheRightLensIn]

Every additional bit of DR doubles the number of luminance steps you can achieve...and they are pro-actively allocated to HIGHLIGHTS FIRST, then to darker tones.

Yes, and because of this I assumed a 12-stop DR (i.e. 5D3) to have a brightness range of [4,16383] and 14-stop (i.e. D800) a range of [0,16383], in which case I considered the highlight range to be equal. My understanding is that the extra stops that the D800 offer are primarily in the bottom end of the DR due to lower read noise etc.

Most of the blather that ensued on this forum shortly after the 5D III announcement was people complaining about the bottom 2-3 stops of DR, which generally account for maybe 20 or so distinct levels?

This was my original point, people (and apparently DXO) are making a big deal of the extra 2 stops (4 levels in total, 3 stops bottom DR would be 8 levels), when in the real world it doesn't matter at all since the brightness data is quantized beyond repair. I'm not usually very good in making my point clearly

they are not either at the top or bottom, sensor are linear capture, not like film, the range is larger vs smaller, that is the difference and if you care more about highlights then you expose less and save more, if you care more about shadows you expose longer and lose more highlights

and don't forget that DxO "print" numbers are based on a normalization to 8MP so you can things like 14.5 stops on a 14bit camera

 

[briansquibb]

I am not so sure how important that last bit (ok 2 bits) of DR is in absolute terms. I think it is lack of DR which is the issue. - as a rule I would say that DR should be about 10. Following this through it is the reduction of DR and increase of noise as the iso increases that grabs my attention. We have just had two cameras released where the high iso performance has improved both producing clean (not too noisy) pictures at 12800 (and maybe above), but look at the DR at those iso levels.

The point I am trying to make is to bump the iso levels up to gain shutter speed can have a detrimental effect of the picture IQ even though there is little noise

 

[straub]

that is the difference and if you care more about highlights then you expose less and save more, if you care more about shadows you expose longer and lose more highlights

Yes, but this hasn't got anything to do with the issue I'm talking about. Even in an optimally "exposed-to-the-right" capture, the very bottom end of DR will be quantized beyond repair. Which is why the "12-stop DR vs 14-stop DR on a 14-bit signal" path argument is pointless.

and don't forget that DxO "print" numbers are based on a normalization to 8MP so you can things like 14.5 stops on a 14bit camera

Not really. Anything above 14 stops in 14-bit signal path is zero. *If* the ADCs use dithering, *some* light below EV(-14) might register into the output, but in the end you will never know if it's a EV(-14.4) or EV(-13.7).

 

[straub]

I am not so sure how important that last bit (ok 2 bits) of DR is in absolute terms. I think it is lack of DR which is the issue. - as a rule I would say that DR should be about 10. Following this through it is the reduction of DR and increase of noise as the iso increases that grabs my attention. We have just had two cameras released where the high iso performance has improved both producing clean (not too noisy) pictures at 12800 (and maybe above), but look at the DR at those iso levels.

Precisely.

 

[JR]

I am not so sure how important that last bit (ok 2 bits) of DR is in absolute terms. I think it is lack of DR which is the issue. - as a rule I would say that DR should be about 10. Following this through it is the reduction of DR and increase of noise as the iso increases that grabs my attention. We have just had two cameras released where the high iso performance has improved both producing clean (not too noisy) pictures at 12800 (and maybe above), but look at the DR at those iso levels.

The point I am trying to make is to bump the iso levels up to gain shutter speed can have a detrimental effect of the picture IQ even though there is little noise

Bang on. It's all about compromise and real life picture. Individually all the specs can be misleading as to your point when you actually use these camera and use them with higher ISO, these same specs mean less for the final picture outcome...

 

[jrista]

Every additional bit of DR doubles the number of luminance steps you can achieve...and they are pro-actively allocated to HIGHLIGHTS FIRST, then to darker tones.

Yes, and because of this I assumed a 12-stop DR (i.e. 5D3) to have a brightness range of [4,16383] and 14-stop (i.e. D800) a range of [0,16383], in which case I considered the highlight range to be equal. My understanding is that the extra stops that the D800 offer are primarily in the bottom end of the DR due to lower read noise etc.

True, the 5D III would have twelve stops because it loses something on the shadow end. That wouldn't really mean that the most common usage of DR is for shadows, though. It would simply mean that shadows are the greatest area of gain were Canon to improve their DR at all.

Most of the blather that ensued on this forum shortly after the 5D III announcement was people complaining about the bottom 2-3 stops of DR, which generally account for maybe 20 or so distinct levels?

This was my original point, people (and apparently DXO) are making a big deal of the extra 2 stops (4 levels in total, 3 stops bottom DR would be 8 levels), when in the real world it doesn't matter at all since the brightness data is quantized beyond repair. I'm not usually very good in making my point clearly

Yup, that would be entirely true as well. Its amazing what a few bits worth of least-significant data containing 0.001% of the grand total number of levels can do to people and markets.

 

[jrista]

they are not either at the top or bottom, sensor are linear capture, not like film, the range is larger vs smaller, that is the difference and if you care more about highlights then you expose less and save more, if you care more about shadows you expose longer and lose more highlights

If your talking strictly about discrete levels, yes, but technically speaking you could make the same argument about film. Were talking about dynamic range, though...successive stops of increased tonal range. Every additional bit IS a doubling of the total DR, so in that sense, no, its not linear.

and don't forget that DxO "print" numbers are based on a normalization to 8MP so you can things like 14.5 stops on a 14bit camera

I'd like to know exactly how the simple act of downscaling magically fabricates additional DR you did not start out with. In the case of the D800, at least according to DXO, you magically gain a full 2/3rds of a stop extra DR simply by scaling all of your images down to an 8x12" print. Well, either you can gain what you don't have with a little bit of resizing-foo...and that would apply to ALL cameras, including Canons; or no amount of digital magic can fabricate DR, and Mount Spokane is absolutely correct that DXO is weighting their results in favor of Nikon. Seeing as the gains Nikon has made with the D800 surpass what seems plausible and reasonable, my commitment to DXO's print DR (which I previously trusted) is wavering.

Personally, I'm of the camp that if you start out with 13.8 stops of native DR strait off the sensor, you have 13.8 stops of DR. If you muck around with a digital image in post to fabricate additional data, that is not true dynamic range, and technically speaking you should be able to do the same thing with images from ANY camera to magically increase DR. In this case, DXO claims about 1.6 times as many additional levels of luminance than you got strait out of the camera...around an additional 10000 levels!! For the math, 1 extra stop of DR is 2^15, or 32768 levels, where as 14 stops of DR is 2^14, or 16384 levels, so a gain of 2/3rds of a stop is 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity simply by scaling your image down. DXOMark is "fiddling".

 

[rpt]

Personally, I'm of the camp that if you start out with 13.8 stops of native DR strait off the sensor, you have 13.8 stops of DR. If you muck around with a digital image in post to fabricate additional data, that is not true dynamic range, and technically speaking you should be able to do the same thing with images from ANY camera to magically increase DR. In this case, DXO claims about 1.6 times as many additional levels of luminance than you got strait out of the camera...around an additional 10000 levels!! For the math, 1 extra stop of DR is 2^15, or 32768 levels, where as 14 stops of DR is 2^14, or 16384 levels, so a gain of 2/3rds of a stop is 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity simply by scaling your image down. DXOMark is "fiddling".

A few questions jrista:
a. 2^15 - I did not get where the 15 came from. Could you elaborate?

b. 2^14 - I presume it is a round off for 13.8. Is that correct? Or have I got it wrong? If wrong, could you explain?

c. 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity...
so your equation was (2^15 - 2^14) * 2/3 ??? Right?

 

[jrista]

Personally, I'm of the camp that if you start out with 13.8 stops of native DR strait off the sensor, you have 13.8 stops of DR. If you muck around with a digital image in post to fabricate additional data, that is not true dynamic range, and technically speaking you should be able to do the same thing with images from ANY camera to magically increase DR. In this case, DXO claims about 1.6 times as many additional levels of luminance than you got strait out of the camera...around an additional 10000 levels!! For the math, 1 extra stop of DR is 2^15, or 32768 levels, where as 14 stops of DR is 2^14, or 16384 levels, so a gain of 2/3rds of a stop is 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity simply by scaling your image down. DXOMark is "fiddling".

A few questions jrista:
a. 2^15 - I did not get where the 15 came from. Could you elaborate?

b. 2^14 - I presume it is a round off for 13.8. Is that correct? Or have I got it wrong? If wrong, could you explain?

c. 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity...
so your equation was (2^15 - 2^14) * 2/3 ??? Right?

Sorry, I was a bit sloppy. First, yes, C is correct: (2^15 - 2^14) * 2/3. Its rough, but basically, if you add ONE additional bit of precision to the sensor (one additional stop of DR), that takes you from 14 bits to 15 bits. The discrete numeric range of 14-bit data is 2^14 (16384), and the range of 15-bit data is 2^15 (32768). Since the D800, according to DXO-mark, has supposedly gained 2/3rds of a stop (14.4 - 13.8, or 0.6), I multiplied the difference between 15 and 14 stops by 2/3rds. Thats a bit aggressive, I guess. Since the raw sensor DR is actually 13.8, you actually probably gain a bit less than 10000 additional levels.

So if we make our interpretation a bit more conservative, DXO is claiming the D800 somehow magically gains around some 7500-8000 discrete tonal levels (or luminance levels) by the simple act of DOWNSAMPLING (as they claim, anyway)? Fishy. And even if that WAS somehow possible, whatever you "gain" is artificial...the camera itself is still the limiting factor when you press the shutter button and capture a scene, in which case the D800, according to DXO's "screen DR" results, captures somewhere around 13.8 stops of DR.

I hope that is more clear.

 

[rpt]

Personally, I'm of the camp that if you start out with 13.8 stops of native DR strait off the sensor, you have 13.8 stops of DR. If you muck around with a digital image in post to fabricate additional data, that is not true dynamic range, and technically speaking you should be able to do the same thing with images from ANY camera to magically increase DR. In this case, DXO claims about 1.6 times as many additional levels of luminance than you got strait out of the camera...around an additional 10000 levels!! For the math, 1 extra stop of DR is 2^15, or 32768 levels, where as 14 stops of DR is 2^14, or 16384 levels, so a gain of 2/3rds of a stop is 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity simply by scaling your image down. DXOMark is "fiddling".

A few questions jrista:
a. 2^15 - I did not get where the 15 came from. Could you elaborate?

b. 2^14 - I presume it is a round off for 13.8. Is that correct? Or have I got it wrong? If wrong, could you explain?

c. 2^15 - 2^14 * 2/3. Sorry, but mathematically, you don't gain that much additional fidelity...
so your equation was (2^15 - 2^14) * 2/3 ??? Right?

Sorry, I was a bit sloppy. First, yes, C is correct: (2^15 - 2^14) * 2/3. Its rough, but basically, if you add ONE additional bit of precision to the sensor (one additional stop of DR), that takes you from 14 bits to 15 bits. The discrete numeric range of 14-bit data is 2^14 (16384), and the range of 15-bit data is 2^15 (32768). Since the D800, according to DXO-mark, has supposedly gained 2/3rds of a stop (14.4 - 13.8, or 0.6), I multiplied the difference between 15 and 14 stops by 2/3rds. Thats a bit aggressive, I guess. Since the raw sensor DR is actually 13.8, you actually probably gain a bit less than 10000 additional levels.

So if we make our interpretation a bit more conservative, DXO is claiming the D800 somehow magically gains around some 7500-8000 discrete tonal levels (or luminance levels) by the simple act of DOWNSAMPLING (as they claim, anyway)? Fishy. And even if that WAS somehow possible, whatever you "gain" is artificial...the camera itself is still the limiting factor when you press the shutter button and capture a scene, in which case the D800, according to DXO's "screen DR" results, captures somewhere around 13.8 stops of DR.

I hope that is more clear.

It is now. Thanks.

 

[V8Beast]

So if we make our interpretation a bit more conservative, DXO is claiming the D800 somehow magically gains around some 7500-8000 discrete tonal levels (or luminance levels) by the simple act of DOWNSAMPLING (as they claim, anyway)? Fishy. And even if that WAS somehow possible, whatever you "gain" is artificial...the camera itself is still the limiting factor when you press the shutter button and capture a scene, in which case the D800, according to DXO's "screen DR" results, captures somewhere around 13.8 stops of DR.

If I understand correctly, DxO claims an increase in DR by the simple act of downsampling an image? This would seem to favor big MP cameras, and I question the legitimacy of the subsequent results.

From a standardized testing procedure, I fully understand that they downsample to level the playing field between sensors of different resolutions. On the other hand, I can't possibly see how this scenario applies to actual shooting technique in real life. Obviously, if you buy a 36 megapixel camera, part of the appeal is the massive resolution. I can't see why you'd pay money for all those pixels, just to downsample an image and throw them away. However, this is precisely the scenario that DxO is simulating, is it not?

 

[V8Beast]

It simply means that for the supposed extra two stops of DR of the D800, the only brightness values that can actually be resolved are EV(-12 1/2), EV(-13) and EV(-14). Anything between EV(-12) and EV(-14) is quantized into one of these three EVs, so there is hardly any detail left, and certainly no gradations. Check https://en.wikipedia.org/wiki/File:Neighborhood_watch_bw.png, that is pretty much what is captured between EV(-13) and EV(-14).

From a use case perspective of using the extra DR to recover shadow detail, the 2-stop advantage in this context is nonexistant. And as such, using it as a metric for a sensor "score" is simply wrong IMO.

From the bantering that ensued after you post this, if I understand correctly, the D800's 2-stop DR advantage over the 5DIII might show up in the highlights, but it is most easily measured in shadow detail that's difficult to see outside of a lab test? If so, this would explain why I'm having such a hard time distinguishing much difference in DR at all between the two bodies based on the sample images that are now trickling out.

Some claim the Nikon's advantage is closer to three stops. That equates to 8x the volume of light. If you quote figures like that to most photographers, many of which aren't so technically inclined as far as things like DxO are concerned, they're going to expect a significant and obvious difference in perceived DR when looking at sample images. Just think about what happens to an image when to slap on a 3-stop ND grad filter in front of your lens. You go from not being able to see any clouds in the sky at all, to having the clouds pop out at you with very rich detail.

Thus far, I am not seeing this type of difference in the samples from the D800 and 5DIII. Consequently, I'm not going to go poopoo on Canon sensors just because they perform better in real-world scenarios than some lab tests suggest

 

[jrista]

So if we make our interpretation a bit more conservative, DXO is claiming the D800 somehow magically gains around some 7500-8000 discrete tonal levels (or luminance levels) by the simple act of DOWNSAMPLING (as they claim, anyway)? Fishy. And even if that WAS somehow possible, whatever you "gain" is artificial...the camera itself is still the limiting factor when you press the shutter button and capture a scene, in which case the D800, according to DXO's "screen DR" results, captures somewhere around 13.8 stops of DR.

If I understand correctly, DxO claims an increase in DR by the simple act of downsampling an image? This would seem to favor big MP cameras, and I question the legitimacy of the subsequent results.

As far as I understand, thats what their "print DR" tests are. I have been digging around DXO's site trying to find more explicit details about what "print DR" actually means, and whether they do any additional processing or not. Their site seems really slow and sketchy these days, so I keep giving up the search. Either way, it just seems extremely fishy to me that you can GAIN DR by downsampling. I'm fully aware that downsampling a high MP image to a normalized image size is a legitimate way to compare IQ between different cameras, however that is normally because of the normalization of noise and resolution...I've never seen it used anywhere else to claim a legitimate increase in DR beyond what the sensor itself is physically capable of. If DXO had claimed 14.0 stops rather than 14.4 stops of DR for the D800, I would have assumed that downsampling simply helped you realize the full potential of the sensor. Now, I am extremely skeptical of their "Print DR" ratings...

From a standardized testing procedure, I fully understand that they downsample to level the playing field between sensors of different resolutions. On the other hand, I can't possibly see how this scenario applies to actual shooting technique in real life. Obviously, if you buy a 36 megapixel camera, part of the appeal is the massive resolution. I can't see why you'd pay money for all those pixels, just to downsample an image and throw them away. However, this is precisely the scenario that DxO is simulating, is it not?

Yup, couldn't agree more.

 

[grahamsz]

If I understand correctly, DxO claims an increase in DR by the simple act of downsampling an image? This would seem to favor big MP cameras, and I question the legitimacy of the subsequent results.

I believe they are making some interpretation of "usable dynamic range". On some level, any 14 bit sensor produces 14 stops of dynamic range - if you look at the difference between the highest and lowest recorded bit.

I'm not intimately familiar with their process but I suspect that they must come up with some determination about when the noise in the shadows in unacceptable and they use that to assess the lower bound.

Downsampling will reduce the perceived noise in the shadows so that may well boost the usable dynamic range and may well drive it close to 14. The only way I can see that you could get over 14 stops would be if there was some kind of non-linearity in the adc response.

 

[lightblood]

Perhaps the real measure is to consider the camera/lens combination. However, DxO's lens ratings are quite the eyebrow raisers? The new Canon EF70-200/2.8L II IS rates below both preceding models. The 24-70/2.8 L on the 5DII (22) rates below the older 28-70 on the original 5D (24). I have owned all of those devices and I KNOW for a fact that DxO is not quite right there. Does it make anyone else wonder whether considering DxO's test results require high sodium intake...or am I reading things incorrectly?
For small format work, I've been a Canon user since the EOS 1. Prior, I was exclusively Nikon (from F2AS to F4). I have a D800 on order. So not a "fanboy" of any type. But I really treasure truly useful information and I wonder if that's to be had from DxO.
Peace and great pictures to all.

 

[lightblood]

Just came across this rough comparison test between the 5DIII & D800 on CNET Asia:
http://asia.cnet.com/shootout-canon-eos-5d-mark-iii-vs-the-nikon-d800-62214094.htm

 

[LetTheRightLensIn]

that is the difference and if you care more about highlights then you expose less and save more, if you care more about shadows you expose longer and lose more highlights

Yes, but this hasn't got anything to do with the issue I'm talking about. Even in an optimally "exposed-to-the-right" capture, the very bottom end of DR will be quantized beyond repair. Which is why the "12-stop DR vs 14-stop DR on a 14-bit signal" path argument is pointless.

and don't forget that DxO "print" numbers are based on a normalization to 8MP so you can things like 14.5 stops on a 14bit camera

Not really. Anything above 14 stops in 14-bit signal path is zero. *If* the ADCs use dithering, *some* light below EV(-14) might register into the output, but in the end you will never know if it's a EV(-14.4) or EV(-13.7).

yes, really, normalization means you can effectively get higher numbers at a certain scale

and use the cams yourself and compare and tell me you don't see a large, noticeable, usable difference

 

[KKCFamilyman]

All thats great but does anyone think the d800 sensor will be that much better than the 5d3?

 

[LetTheRightLensIn]

So if we make our interpretation a bit more conservative, DXO is claiming the D800 somehow magically gains around some 7500-8000 discrete tonal levels (or luminance levels) by the simple act of DOWNSAMPLING (as they claim, anyway)? Fishy. And even if that WAS somehow possible, whatever you "gain" is artificial...the camera itself is still the limiting factor when you press the shutter button and capture a scene, in which case the D800, according to DXO's "screen DR" results, captures somewhere around 13.8 stops of DR.

If I understand correctly, DxO claims an increase in DR by the simple act of downsampling an image? This would seem to favor big MP cameras, and I question the legitimacy of the subsequent results.

From a standardized testing procedure, I fully understand that they downsample to level the playing field between sensors of different resolutions. On the other hand, I can't possibly see how this scenario applies to actual shooting technique in real life. Obviously, if you buy a 36 megapixel camera, part of the appeal is the massive resolution. I can't see why you'd pay money for all those pixels, just to downsample an image and throw them away. However, this is precisely the scenario that DxO is simulating, is it not?

It doesn't favor any MP level, that is the point of normalization, to make things comparable on a fair basis.

It doesn't really matter they normalize to so long as you compare the cameras relative to one another. The absolute numbers they report won't match what you get out of it unless you view a print of a specific size from a specific distance matching their normalization criteria.

 

[briansquibb]

All thats great but does anyone think the d800 sensor will be that much better than the 5d3?

I think that all the better digital cameras are first class nowadays. I believe that the majority of people would be struggle to hit the limits for example on the 1DX, 5DII, D800 etc on a regular basis. There are always of course exceptions to this rule usually where people are very specialised.

It is just a question of choosing the package that suits you most. With the body it is probably going to be the lens which impacts the choice, especially for specialists.