December 02, 2016, 03:26:24 PM

Author Topic: DxOMark Sensor Performance: Nikon vs. Canon  (Read 126926 times)

elflord

  • 5D Mark III
  • ******
  • Posts: 693
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #300 on: October 17, 2012, 09:30:38 PM »
Downsampling doesn't "create detail", but it moves the black point.

* Photographic dynamic range is the range of useful detail.

* As you conceded above, down sampling does not create detail. In fact, it throws detail away.

* Down sampling therefore cannot increase photographic dynamic range.

* DxO's normalized results are therefore false.

The last point does not follow. In fact the negation of it follows.

Look, suppose my saturation point per pixel is 12EV above my blackpoint.

Downsampling, as we agreed, lowers the blackpoint.

After downsampling, because my blackpoint moved down 1EV, my saturation point per pixel is 13EV above my blackpoint.

If I want my dynamic range scores to be the same for both images, I need to normalize, otherwise I will give a higher score to the downsampled image (13EV vs 12EV)

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #300 on: October 17, 2012, 09:30:38 PM »

elflord

  • 5D Mark III
  • ******
  • Posts: 693
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #301 on: October 17, 2012, 09:31:48 PM »
Please feel free to post any unanswered "critiques"

All of them from what I can see.

Be a little more specific perhaps ?

elflord

  • 5D Mark III
  • ******
  • Posts: 693
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #302 on: October 17, 2012, 09:35:48 PM »
I must say that all of this “my camera is better than your camera” talk gives me a headache, and I’m not sure I want to get in the middle of this, um, whatever contest.  I have read in detail the DxO results, along with many other test results and feel this debate is “full of sound and fury signifying nothing!”  Bottom line to me is whether or not the camera in question is the best match for my needs.

I'm not into the whole idea of being a "fan" of any particular camera, but I am a scientist, and I do take some offence at scientists (DxO) being attacked by camera "fans".

As far as my own gear is concerned, I use a 5DII and am quite happy with it. Dynamic range is not a key issue for what I do.

elflord

  • 5D Mark III
  • ******
  • Posts: 693
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #303 on: October 17, 2012, 09:44:18 PM »
You're making the same mistake as the others. Noise inflates the black point. Down sampling reduces noise, lowering the black point. But the black point is not actually the floor of dynamic range. The floor is the point where usable detail ceases to exist. Down sampling actually throws away detail. It cannot extend DR.

I understand that a 0db floor may be too liberal a baseline for some, but you haven't made the case that this makes much difference to results. For benchmarking you can't use something as subjective as "usable detail".

Regarding downsampling not extending true dynamic range, we do agree on this point.

However, the logical conclusion is that you should normalize. If you normalize, dynamic range defined in terms of the black point will be invariant under downsampling.

It is when you do NOT normalize that downsampling will change your dynamic range.

ishdakuteb

  • Canon 7D MK II
  • *****
  • Posts: 476
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #304 on: October 17, 2012, 10:19:14 PM »
sorry for you, the pictures show the real pixel values  and I present them in the same way, this is the difference in lower levels

you do not have to feel sorry for me since there is not a big gap between your understanding about dslr comparing to my understanding even though you have started photography since i was in elementary.  what would you like to challenge here?  most of technical things in dslr relate to physics and software, which part would you like to challenge? or want to challenge about getting correct exposure with out of camera jpeg... :)

actually, i have to feel sorry for you mr. mikael ridsesal, having no guts about what you have said and not being able to show your real imges (have requested your raw files few time, remember?).  if you want to learn more about pixels, buy your self mathlab, study and play with it.

MarkII

  • Canon AE-1
  • ***
  • Posts: 71
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #305 on: October 18, 2012, 08:24:30 AM »
Pay careful attention: down sampling cannot increase*** photographic dynamic range ***
In photography dynamic range is the range of usable detail, not the range between white and black points.
Actually, no. I think you misunderstand the concept of DR as expressed by the DXO measurements, and I suggest you read the definition of dynamic range on Wikipedia (they even have an entry for photography on the page):

  http://en.wikipedia.org/wiki/Dynamic_range

DXO state quite carefully what their how they define dynamic range for the purpose of their measurements, and their figures - including the 14.3 bit value for the normalised images are not implausible. Note that the 14.3 figure refers to the DR of a single pixel *after* downsampling. Just like the newsprint and the simplified math examples I outlined this can have more DR (bits per pixel in any given pixel) than the original.

(And if you think my analogy with print is invalid, you need to *explain* why.  I may be wrong, but simply repeating your assertion that this is so does not show it to be the case).

It is correct, however, to say that downsampling does not increase 'detail' in an image - you can not gain any information. What it does is to allow you to trade-off spatial resolution against luminous resolution (noise, and quantisation - which is itself really just another form of noise).

(This is like pulling teeth!)
Quite.

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2626
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #306 on: October 18, 2012, 10:12:31 AM »
DXO state quite carefully what their how they define dynamic range for the purpose of their measurements, and their figures - including the 14.3 bit value for the normalised images are not implausible. Note that the 14.3 figure refers to the DR of a single pixel *after* downsampling. Just like the newsprint and the simplified math examples I outlined this can have more DR (bits per pixel in any given pixel) than the original.
So instead of calling it DR can we call it DxODR and bury the hatchet (or hatchets)?

I think DR is the unit of measure and its use should be consistent. That same link talks of it in bits (sorry not trying to be sarcastic or anything - please bear with me... Thanks) and I think of bits as indivisible. So fractional bits of DR make no sense to me specially when you can get a higher value than the ADC can generate. And I understand it is maths around the original image so let us call it what it is. It is not DR. It is something similar in spirit that somebody invented.

My 2c...

Rustom

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #306 on: October 18, 2012, 10:12:31 AM »

MarkII

  • Canon AE-1
  • ***
  • Posts: 71
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #307 on: October 18, 2012, 12:20:29 PM »
So instead of calling it DR can we call it DxODR and bury the hatchet (or hatchets)?
I hope that there is no hatchet to bury...

I think DR is the unit of measure and its use should be consistent. That same link talks of it in bits (sorry not trying to be sarcastic or anything - please bear with me... Thanks) and I think of bits as indivisible. So fractional bits of DR make no sense to me specially when you can get a higher value than the ADC can generate. And I understand it is maths around the original image so let us call it what it is. It is not DR. It is something similar in spirit that somebody invented.

If it helps, call it something different...

Fractional bits represent the notional information content of the data rather than the storage size (which will be an integer number of bits).

For example, if I have a two bit number I can store { 0, 1, 2, 3 }. However, if my data for some reason are only able to take the values { 0, 1, 2 }, then I have more values that I can store in one bit, but fewer than I really need two bits for - in other words, a fractional number of bits.

rpt

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2626
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #308 on: October 18, 2012, 12:26:26 PM »
Fractional bits represent the notional information content of the data rather than the storage size (which will be an integer number of bits).

For example, if I have a two bit number I can store { 0, 1, 2, 3 }. However, if my data for some reason are only able to take the values { 0, 1, 2 }, then I have more values that I can store in one bit, but fewer than I really need two bits for - in other words, a fractional number of bits.
Can you point me to some reading material?

jrista

  • Canon EF 600mm f/4L IS II
  • **********
  • Posts: 5334
  • EOL
    • Nature Photography
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #309 on: October 18, 2012, 02:21:04 PM »
(And if you think my analogy with print is invalid, you need to *explain* why.  I may be wrong, but simply repeating your assertion that this is so does not show it to be the case).

I think this is a critical point, and possibly the root of the contention of the "DXO DR naysayers." As someone who prints a lot myself, perhaps I can offer some insight.

Assuming you print at native resolution, printing does not average the original amount of information into something less. A print at native resolution represents the same, original, fully detailed image information at a higher density. That higher density may be anywhere from three to six times greater than the 100ppi of my screen (or even more dense, for those who use a 72ppi screen). That is very different than downsampling, which is destructive to information. Generally speaking, I downscale my images for display online. I print at native size at home, or perhaps slightly enlarged, and I might upscale (at a lower PPI) for large canvas prints from a lab. Depending on the amount of cropping, I might print as low as 8x10 with a small amount of downscaling (no where near a two-fold reduction for proper averaging, though), although most of the time it is 11x17", 12x18", 13x19" (the latter being my most common print size, usually at 300ppi).

There are other problems with DXO calling their rated DR "Print DR", though. Assuming you are using a godly form of paper, such as Innova FibaPrint Gloss, which has a dMax of over 2.7, you might be able to get 7 stops or so from a print. Your average fine art print paper has a dMax randing from around 1.3 to 1.5 on average to 1.75 or so for some of the more recent higher-end fine art papers. That gets you maybe 5-6 stops of DR. Most papers also don't have particularly high white points. Again, Innova FibaPrint bleached or bright white glossy papers have a pretty high brightness, and the Moab Lasal papers might be some of the brightest papers available with the highest L* I've ever seen (although they achieve it via OBAs, which require proper lighting properly produce the papers full DR.) Despite that, none of those papers, even when printed with top of the line pigment inks like Lucia or UltraChrome, will offer much more than around 7 stops of DR.

So that takes us back to the definition of DR. I'm happy to accept that DXO has a purely mathematical interpretation of DR, the ratio between white point (maximum saturation) and black point (noise floor). Again, though, I am not sure it is a useful or realistic definition of what dynamic range is. When one thinks about the value of dynamic range in digital photography, the first thing that usually comes to mind is the ability to recover useful detail from deep shadows. I say from the shadows, as I think any photographer who uses digital knows that it is critical to preserve the highlights, as once they are clipped, detail is well and truly gone. Since detail generally "fades" into noise on the shadow end, an improvement in photographic dynamic range has the benefit of improving one's ability to "recover", or pull up, detail out of the deep shadows.

Now, in this respect, I don't think anyone who has been involved in this debate will honestly dispute the fact that Sony Exmor sensors offer more dynamic range. That has never been in dispute...its a simple fact, clear as day for anyone who has seen or directly used a Nikon RAW image from any camera with an Exmor sensor. The dispute on record here, if I may define it according to my own views as well as that which I've read from other DXO DR naysayers, is this:

What value does DXO PrintDR (the mathematically derived ratio between white point (maximum saturation, FWC) and black point (electronic noise floor)) have in a real-world context?

From the standpoint of simply moving the black point in a downsampled image, the only thing that occurs is shadows become darker. One LOSES information during the process of downsampling, so the primary benefit of having additional DR in the hardware no longer applies. In the context of viewing images on a computer screen, primarily done via the web, having a deeper black point might be valuable. Computer screens generally support a much deeper black point than actual prints on paper (particularly prints on high quality fine art paper), although none actually support 14 stops of DR regardless, and the average consumer screen is only 6-bit, so roughly the same DR as a print.

When it comes to real print, assuming one is printing at native size, or an upsampled image, original detail is preserved or slightly softened, but none of it is lost due to downsampling. Regardless, assuming one even does significantly downsample a D800 image so they can print at 8x10", even printed on the highest dMax papers on the market with the brightest L* rating, your going to get HALF the DR you should supposedly be getting from DXO's 14.4 stop Print DR rating. If we assume you tweak the white and black points, curves, and levels in Photoshop to manually and ideally compress all that extensive 14.4 stops of DR into the 5-7 stops of DR your paper is capable of, then the additional mathematical DXO DR (darker black point) offered by downsampling is still of no benefit. It might actually make it more difficult to compress a greater range of shadow tones into the limited dynamic range of your paper, resulting in some funky tonality.

These are the issues I have with DXO's "Print DR" statistic. It is an unrealistic number, purely mathematical in derivation, that does not seem to translate into any real-world improvement in "print". It MIGHT offer slightly better blacks when downsampled images are viewed on the web. Blacks might be just a bit richer, assuming someone actually has a properly calibrated screen with high enough bit depth to actually support it. The average home user still uses a 6-bit screen. Most serious photographers have an 8-bit screen, and some serious professionals might have thousand-dollar 10-bit screens. At best, an observer viewing a photo on the web will be able to observe about 8-10 stops of DR, although on average we are still back to 6 stops on average.

At best, DXO's downsampled DR rating should probably be called Web DR. It is not detail-preserving Photographic DR, as upon downsampling you lose detail.  It is definitely not Print DR, since a print is inherently more about color richness and gamut than white-to-black point dynamic range. The depth of blacks sometimes matters in a print, however the deeper your black point in print, the harder it tends to be to actually discern fine shadow detail. Things look richer and more contrasty, but not necessarily more detailed. Papers with a higher white point AND slightly less dense blacks tend to look better, despite having far lower dynamic range than the original photo.

So, what is the value of DXO Print DR? Realistically, practically, physcally...what do I actually gain by downsampling my full-detail RAW into a smaller-sized TIFF? For that matter, what value does DXO Print DR have if I save as a compressed JPEG for viewing on the web? Are we really just talking about a DXO weighted score, and nothing more? If so, should it really be called Dynamic Range, or is there a better term DXO could use that wouldn't come off as some kind of sketchy maneuvering (real or simply perceived) of their results in favor of a major monetary contributor?

tnargs

  • Rebel T6i
  • ****
  • Posts: 138
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #310 on: October 18, 2012, 06:05:10 PM »
jrista, I think you have just shown that DxO should be completely ignored by every intelligent photographer.

Photographic hobbyists and professionals need test measurements that are highly relevant to their realistic needs in typical photographic situations.

What they DON'T need is a scientific measurement and subsequent aggregation that is completely inconsistent with their needs as a photographer. Complete with company-specific definitions and randomly chosen normalisation points, that one has to read all the fine print to get a grasp of how on earth they came up with that number, score, or ranking. Hello DxO!

Can you imagine a food nutrition analysis website published by scientists, where they publish highest scores for palm oil and birthday cake, and lowest scores for raw fish and broccoli. After delving several levels further into their website you find other surprising and almost bizarre scores and numerics. Then after delving another 40 pages into explanatory papers and definitions, if you are sufficiently scientific in mindset, you will discover that there are no flaws in their logic or measurement methods and their system is completely consistent within itself. How useful is this website? How misleading is it?

jrista

  • Canon EF 600mm f/4L IS II
  • **********
  • Posts: 5334
  • EOL
    • Nature Photography
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #311 on: October 18, 2012, 06:34:08 PM »
jrista, I think you have just shown that DxO should be completely ignored by every intelligent photographer.

Well, I wouldn't go so far as to say DXO in its entirety should be completely ignored. My only beef with DXO is quite specifically their Print DR "score", for all the reasons I stated above. DXO does provide valuable statistics and measurements. I just have a hard time with their scoring structure being so heavily weighted in favor of a statistic that...seems entirely meaningless, and utterly lacking in real-world value.

elflord

  • 5D Mark III
  • ******
  • Posts: 693
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #312 on: October 18, 2012, 07:28:43 PM »
(And if you think my analogy with print is invalid, you need to *explain* why.  I may be wrong, but simply repeating your assertion that this is so does not show it to be the case).

I think this is a critical point, and possibly the root of the contention of the "DXO DR naysayers." As someone who prints a lot myself, perhaps I can offer some insight.

Assuming you print at native resolution, printing does not average the original amount of information into something less. A print at native resolution represents the same, original, fully detailed image information at a higher density. That higher density may be anywhere from three to six times greater than the 100ppi of my screen (or even more dense, for those who use a 72ppi screen). That is very different than downsampling, which is destructive to information. Generally speaking, I downscale my images for display online. I print at native size at home, or perhaps slightly enlarged, and I might upscale (at a lower PPI) for large canvas prints from a lab. Depending on the amount of cropping, I might print as low as 8x10 with a small amount of downscaling (no where near a two-fold reduction for proper averaging, though), although most of the time it is 11x17", 12x18", 13x19" (the latter being my most common print size, usually at 300ppi).

If you're comparing two different cameras, it doesn't make sense to compare dynamic range at two different resolutions (and therefore different noise/resolution tradeoffs)

As long as your sampling to some known target resolution (whether it's 2400x3000 or 3300x5100  or 3900x5700), you need take into account that the higher resolution camera will benefit from either more downsampling or less upsampling.

For any given print size, as long as dynamic range per pixel is equal, you always get either better dynamic range (lower black point) or more detail (resolution) from a higher megapixel camera.

Quote
From the standpoint of simply moving the black point in a downsampled image, the only thing that occurs is shadows become darker.

No. The point at which the SNR is 0db moves down. So does the point where the SNR is  5db, 10db, etc. The shadows don't become darker, they become less noisy

Quote
One LOSES information during the process of downsampling,

I'm not sure which information you're talking about "losing". A 40 megapixel camera sampled to 10 megapixels doesn't have less "information" than the image from a 10 megapixel camera.

The point is to put everyone on the same playing field.

If you're looking at a measure in isolation, perhaps the screen score is easier to understand. But if you're comparing sensors with different resolutions, you need to make some adjustment to account for the fact that the higher resolution sensor has more pixels and therefore more photographic detail.

I think the point about only 6 stops of dynamic range being available for paper is a bit of a digression -- if you don't need more than 6 or maybe 8 stops for some safety margin, why would you care about dynamic range ? Any modern DSLR has more than that.

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #312 on: October 18, 2012, 07:28:43 PM »

jrista

  • Canon EF 600mm f/4L IS II
  • **********
  • Posts: 5334
  • EOL
    • Nature Photography
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #313 on: October 18, 2012, 09:57:52 PM »
The only thing I can gather from the Pro-DXODR camp's arguments so far, is that we are not really comparing images on a dynamic range basis with DXO's Print DR statistic. We are comparing images on a noisiness basis. Black points and white points and noise may make for a neat and simple definition of DR, but as far as I can gather, DR is not what what is of interest here. The amount of noise, and how that noise presents...explicitly in relation to other images from other sensors...is what is of interest here. If the sole purpose of downscaling a large megapixel image to the size of a smaller megapixel image (or, for that matter, the inverse) is to see which camera is NOISIER, then I have no issue. I don't dispute the necessity of that, although I've never called the improvement in IQ when downsampling a large image to a small image an improvement in dynamic range.

I do dispute that you actually gain anything beneficial by downscaling a high megapixel image to a smaller size, in a real-world context (as differentiated from a "clean-room" test and compare context, within which normalizing noise for the purposes of subjective IQ comparison DOES make sense). I dispute the simple notion of scaling at all to determine a sensor's dynamic range, given that the point of having MORE dynamic range is to preserve detail under a wider range of tones, and in the case of digital photography allow that detail to be "recovered" via post-process exposure adjustments. Adjustments that will only really work properly when editing a RAW...which intrinsically cannot be scaled.



I'm not sure which information you're talking about "losing". A 40 megapixel camera sampled to 10 megapixels doesn't have less "information" than the image from a 10 megapixel camera.

Still assuming that one is perpetually comparing camera sensors. Statistical knowledge about a sensor is not necessarily solely useful for the purpose of comparing them. If I buy a 40mp camera, I have no intention of comparing it to anything else. I have the intention of using it...as a 40mp camera. As such, I want the full stinkin fourty megapixels, not something less. I also want to know what my camera is capable of natively, unscaled, untainted. I want to know what my 40mp camera can do itself, not how well it might do thing X in comparison to cameras A, B, and C! In other words, I want to know its intrinsic capabilities, not its relative capabilities.

In that context, scaling my 40mp image DOWN to any size below its native size means a LOSS of information...a true and realized reduction in real image detail that used to exist, and no longer does at 10mp. It also means I have to print at a smaller size, which could be counted as a different kind of loss...but I guess that is beside the point.


The point is to put everyone on the same playing field.

Well, that is one of the points. It is not the only point of having statistical information about a camera's sensor, as noted above.

If you're looking at a measure in isolation, perhaps the screen score is easier to understand. But if you're comparing sensors with different resolutions, you need to make some adjustment to account for the fact that the higher resolution sensor has more pixels and therefore more photographic detail.

Agreed. From the standpoint of normalizing noise to compare how noisy one sensor is vs. another (which is what I gather the DXO Print DR statistic is all about based on the arguments from you an LTRLI), then sure, you should be normalizing image size. I've argued that myself in the past, although I've never called the change in noise an improvement in dynamic range before, but thats probably simply because I define dynamic range a bit differently than DXO (although admittedly in a different context as well...that of "real-world digital RAW post-processing" where most of us photographers actually live, so it is not necessarily to say DXO's definition is wrong...just meant for a different context.) I've simply called downscaling for the sake of comparing noise the normalization of noise. Something along the lines of "Normalized IQ Score" would be a much better term for this than "Print DR", though, as it really has nothing whatsoever to do with print. Print, paper, inks, ink density, lighting, etc. make actual print so far removed from what DXO is actually trying to demonstrate with this statistic that they should really rename "Print DR" to something less controversial, and something more meaningful.

I believe Imatest, technically a competitor to some of DXO's other business I guess, named their score for roughly the same thing accordingly: SQF, or Subjective Quality Factor. I believe DXO "Print" statistics are really another form of Kodak's SQF. I believe Imatest got the name right for IQ measurements as well...they simply call it MTF, rather than something more obscure like "Screen".

I think the point about only 6 stops of dynamic range being available for paper is a bit of a digression -- if you don't need more than 6 or maybe 8 stops for some safety margin, why would you care about dynamic range ? Any modern DSLR has more than that.

Sorry, but it's been my entire point all along, as my complaint is with the fact that DXO sells "Print DR" as something it is not, by its very name, even. Or, at the very least, they explain their modeling system in a way that many people have incorrectly interpreted, as people are inevitably going to think "ink on paper" when they hear the word "print". I also believe the heavily weighted use of "Print DR", if it is weighted rather than purely a measured statistic (I can't find the page right now that describes how and where DXO adds "bonus points") in their scoring mechanism is extremely misleading.

Additionally, the same statistics, assuming they are accurate enough in the broader context (which in the case of Print DR, it is not...it has a very narrow and specific usage solely in the context of comparing sensors scored by DXO), can be used to learn about what a sensor offers in and of itself outside of any context of comparison. Such as how much exposure latitude do I REALLY, ACTUALLY have when I am pushing shadows around with my RAW images in a RAW editor. Or, say, what the sensor offers when ACTUALLY PRINTING IT AT NATIVE SIZE (so no scaling or averaging of any kind)? DXO's Print DR statistic is insufficient to answer those questions, while Screen DR, effectively being hardware measurements as they are otherwise untainted by software, IS sufficient to answer those questions. Hence my reliance on Screen DR to describe intrinsic capabilities in contexts where relative capabilities are meaningless.

Perhaps my complaint is simply with how DXO sells, names, uses, and refers to their statistics. If they had used the terms "MTF" or "Measurements" and "SQF" in place of "Screen" and "Print" respectively, I probably would have never had a problem. Labeling their overall dynamic range score as the "Landscape" score (immediately putting it in a context where actual camera USERS might literally try to photograph scenes with more than 14 stops of real-world dynamic range), then stuffing in a number that (potentially greatly) inflates the intrinsic hardware capabilities, is excessively misleading. At the very least, SQF is explicitly defined as a way to produce normalized comparisons for the purpose of accounting for "viewer experience"...a subjective factor, which sounds a lot like what DXO's "Print" scores are supposed to be. SQF has also been shown to be an ideal way to statistically measure image IQ produced by digital sensors (when, ironically, it wasn't so ideal for measuring IQ from film), one which has never risen so much ire as DXO's "Print DR" statistic.

LetTheRightLensIn

  • Canon EF 400mm f/2.8L IS II
  • *********
  • Posts: 4753
Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #314 on: October 19, 2012, 12:32:10 AM »
(And if you think my analogy with print is invalid, you need to *explain* why.  I may be wrong, but simply repeating your assertion that this is so does not show it to be the case).

I think this is a critical point, and possibly the root of the contention of the "DXO DR naysayers." As someone who prints a lot myself, perhaps I can offer some insight.

Assuming you print at native resolution, printing does not average the original amount of information into something less. A print at native resolution represents the same, original, fully detailed image information at a higher density. That higher density may be anywhere from three to six times greater than the 100ppi of my screen (or even more dense, for those who use a 72ppi screen). That is very different than downsampling, which is destructive to information. Generally speaking, I downscale my images for display online. I print at native size at home, or perhaps slightly enlarged, and I might upscale (at a lower PPI) for large canvas prints from a lab. Depending on the amount of cropping, I might print as low as 8x10 with a small amount of downscaling (no where near a two-fold reduction for proper averaging, though), although most of the time it is 11x17", 12x18", 13x19" (the latter being my most common print size, usually at 300ppi).

There are other problems with DXO calling their rated DR "Print DR", though. Assuming you are using a godly form of paper, such as Innova FibaPrint Gloss, which has a dMax of over 2.7, you might be able to get 7 stops or so from a print. Your average fine art print paper has a dMax randing from around 1.3 to 1.5 on average to 1.75 or so for some of the more recent higher-end fine art papers. That gets you maybe 5-6 stops of DR. Most papers also don't have particularly high white points. Again, Innova FibaPrint bleached or bright white glossy papers have a pretty high brightness, and the Moab Lasal papers might be some of the brightest papers available with the highest L* I've ever seen (although they achieve it via OBAs, which require proper lighting properly produce the papers full DR.) Despite that, none of those papers, even when printed with top of the line pigment inks like Lucia or UltraChrome, will offer much more than around 7 stops of DR.

So that takes us back to the definition of DR. I'm happy to accept that DXO has a purely mathematical interpretation of DR, the ratio between white point (maximum saturation) and black point (noise floor). Again, though, I am not sure it is a useful or realistic definition of what dynamic range is. When one thinks about the value of dynamic range in digital photography, the first thing that usually comes to mind is the ability to recover useful detail from deep shadows. I say from the shadows, as I think any photographer who uses digital knows that it is critical to preserve the highlights, as once they are clipped, detail is well and truly gone. Since detail generally "fades" into noise on the shadow end, an improvement in photographic dynamic range has the benefit of improving one's ability to "recover", or pull up, detail out of the deep shadows.

Now, in this respect, I don't think anyone who has been involved in this debate will honestly dispute the fact that Sony Exmor sensors offer more dynamic range. That has never been in dispute...its a simple fact, clear as day for anyone who has seen or directly used a Nikon RAW image from any camera with an Exmor sensor. The dispute on record here, if I may define it according to my own views as well as that which I've read from other DXO DR naysayers, is this:

What value does DXO PrintDR (the mathematically derived ratio between white point (maximum saturation, FWC) and black point (electronic noise floor)) have in a real-world context?

From the standpoint of simply moving the black point in a downsampled image, the only thing that occurs is shadows become darker. One LOSES information during the process of downsampling, so the primary benefit of having additional DR in the hardware no longer applies. In the context of viewing images on a computer screen, primarily done via the web, having a deeper black point might be valuable. Computer screens generally support a much deeper black point than actual prints on paper (particularly prints on high quality fine art paper), although none actually support 14 stops of DR regardless, and the average consumer screen is only 6-bit, so roughly the same DR as a print.

When it comes to real print, assuming one is printing at native size, or an upsampled image, original detail is preserved or slightly softened, but none of it is lost due to downsampling. Regardless, assuming one even does significantly downsample a D800 image so they can print at 8x10", even printed on the highest dMax papers on the market with the brightest L* rating, your going to get HALF the DR you should supposedly be getting from DXO's 14.4 stop Print DR rating. If we assume you tweak the white and black points, curves, and levels in Photoshop to manually and ideally compress all that extensive 14.4 stops of DR into the 5-7 stops of DR your paper is capable of, then the additional mathematical DXO DR (darker black point) offered by downsampling is still of no benefit. It might actually make it more difficult to compress a greater range of shadow tones into the limited dynamic range of your paper, resulting in some funky tonality.

These are the issues I have with DXO's "Print DR" statistic. It is an unrealistic number, purely mathematical in derivation, that does not seem to translate into any real-world improvement in "print". It MIGHT offer slightly better blacks when downsampled images are viewed on the web. Blacks might be just a bit richer, assuming someone actually has a properly calibrated screen with high enough bit depth to actually support it. The average home user still uses a 6-bit screen. Most serious photographers have an 8-bit screen, and some serious professionals might have thousand-dollar 10-bit screens. At best, an observer viewing a photo on the web will be able to observe about 8-10 stops of DR, although on average we are still back to 6 stops on average.

At best, DXO's downsampled DR rating should probably be called Web DR. It is not detail-preserving Photographic DR, as upon downsampling you lose detail.  It is definitely not Print DR, since a print is inherently more about color richness and gamut than white-to-black point dynamic range. The depth of blacks sometimes matters in a print, however the deeper your black point in print, the harder it tends to be to actually discern fine shadow detail. Things look richer and more contrasty, but not necessarily more detailed. Papers with a higher white point AND slightly less dense blacks tend to look better, despite having far lower dynamic range than the original photo.

So, what is the value of DXO Print DR? Realistically, practically, physcally...what do I actually gain by downsampling my full-detail RAW into a smaller-sized TIFF? For that matter, what value does DXO Print DR have if I save as a compressed JPEG for viewing on the web? Are we really just talking about a DXO weighted score, and nothing more? If so, should it really be called Dynamic Range, or is there a better term DXO could use that wouldn't come off as some kind of sketchy maneuvering (real or simply perceived) of their results in favor of a major monetary contributor?

The value of it is to compare cameras more fairly than not using it. If you print those files you'll notice that printed to the same size and viewed from the same distance the same thing happens 40MP source vs 12MP source. Or if you print both at 8MP and then put one print say 10' farther away from the other.

If you don't normalize then if you go by your screen DR all the time then you'd think that  a 20x20" print from a 36MP source cam would have a lot, lot more nasty noise than one from an 18MP camera using the same generation of technology. It doesn't make sense to compare a 36MP source as if it were the same scale as 18MP since you could always filter noise to 18MP scale or filter and downsample to 18MP and trade away your extra detail information to make it look like the same 18MP pixel level performance of the other camera. You are trying to take an advantage that can be traded various ways into a singular resolution advantage but a disadvantage in all other ways which really isn't a fair way to compare.

If you want to take full advantage of the 36MP advantage over say the 22MP of the 5D3 you can't do that and at the same time also give the extra print version boost to DR, true enough, but if you don't want to take advantage of it then you can.

canon rumors FORUM

Re: DxOMark Sensor Performance: Nikon vs. Canon
« Reply #314 on: October 19, 2012, 12:32:10 AM »