DXOmark ranks DB image quality well below DSLR!

[Mort54]

I shoot about 50% in a studio and 50% on location.  I think in a studio MFDB wins by a mile, but on location it can be tricky.  
Basically, I'm at a cross road between a D3X and a used MFDB, and it sounds like given the choice you would go for a D3X...

If you really really really need to print big, and your revenues will benefit from printing big, then MFDBs make a lot of sense. However, it's all too easy to be seduced by those gorgeously detailed files, and to convince yourself you need MFDB when you really don't. Only you can answer that question. If I were you, I'd rent a D3X (or a 1DsIII) and shoot the same subjects that you shot with the MF system, and then print them side by side at the same print sizes you're currently selling to customers. Unless you are printing really big, I doubt you're going to see a huge difference, if any. And after all, it's prints you sell, not the files.

I haven't bought a D3X yet because, like you, I'm put off by Nikon's pricing strategy. A D3X costs way less than I paid for my P45+ system, of course, but I don't want to reward Nikon for what I consider to be a horrible marketing decision. Plus I already have the P45+ system for my high MP needs, so I'm reluctant to purchase something else that satisfies the same purpose. Nevertheless, knowing what I know now about how D3 files compare against  P45+ files, and the inconvenience of using two systems, and the inferiority of MF autofocus, metering, lenses, live view, etc, I think a D3X would be a better match for my needs than the P45+ system. If a D700X with the D3X sensor in it ever appears, it'll be a no brainer, and that's basically what I'm waiting for (although it's getting harder and harder to resist the D3X :-)

[Panopeeper]

Usable DR is a lot less than what you come up if you measure just the DR as technically defined and there are big differences in how these numbers shift by camera

It depends on the definition of dynamic range adopted for the measurement. The engineering (scientific) definitions is IMO total waste of time and bandwidth relating to photography, so is DxO's 18% etc. measurement.

When I am measuring the noise/dynamic range, I don't presume any specific limit of acceptance of noise. For example the 5D2 noise graphs can be used by selecting the level of accepted noise on the y axis, and the x value shows the dynamic range, depending on the ISO (I have a collection of captures showing how the nosie of let's say 40% looks).

Of course, noise reduction is a subject on its own.

[Graeme Nattress]

This is BS. "Digital gain" in the current context is an oxymoron.

So you don't like the term - then suggest a more appropriate one and be positive about it rather than heckle and call BS. This can be a very useful and educational discussion....

Graeme

[EricWHiss]

The Phase Ones create nominally 16bit raw data. In fact, the pixel values are 16bit wide, i.e. they go up to 65535. However, about two bits of that are not only useless, they are actually detrimental; if the raw processing requires so much pushing, that the low-order two bits become apparent in the result, then the photographer becomes unhappy with the camera (back).

Gabor,
Can you tell us what kind of camera(s) you use, what you shoot, and if you make prints? Don't think it doesn't matter.  We all have our tricks on handling the files and experience doing real work is very valuable - much more so than pushing the files around in a program that measures noise.  Noise is subjective - there's nice noise that mimics what we are used to with film and there's ugly noise.     Don't forget that a lot of people add grain to files to get the look they want.   Noise is also added in up-sampling programs now.  Does that make you crazy or what?
Eric

[Mort54]

I do STRESS however, that at smaller sizes like A3, you'd be very hard pressed to tell the difference. I only see teal value in MFDB if you're going to print big. For all other work, especially wildlife, DSLRs are way superior.

That's the key. I totally agree with being hard pressed to see a difference at A3. That's true even with my D3 files compared to my P45+ files. At A2 I can more readily see a difference, but both prints still look fantastic. Someone thinking about getting a MFDB system should ask themselves if this difference is enough to justify the extra cost and inconvenience of going MFDB. For some people, the answer will be yes, for others it'll be no. If printing bigger than A2 helps the bottom line, the answer becomes more and more obvious in favor of MFDB the bigger you print.

If A2 were my final goal (and in most cases that's what I print), I think the money would be better spent on upgrading to the best primes available, making use of live view whenever possible to achieve more precise critical focus, and maybe even having the DSLR AA filter removed by a company like MaxMax. I think if someone did all those things, he'd even be hard pressed to see a difference at A2 between a D3X and P45+, and it would be way less expensive than buying a P45+ system.

[photolinia]

If you really really really need to print big, and your revenues will benefit from printing big, then MFDBs make a lot of sense. However, it's all too easy to be seduced by those gorgeously detailed files, and to convince yourself you need MFDB when you really don't. Only you can answer that question. If I were you, I'd rent a D3X (or a 1DsIII) and shoot the same subjects that you shot with the MF system, and then print them side by side at the same print sizes you're currently selling to customers. Unless you are printing really big, I doubt you're going to see a huge difference, if any. And after all, it's prints you sell, not the files.

I haven't bought a D3X yet because, like you, I'm put off by Nikon's pricing strategy. A D3X costs way less than I paid for my P45+ system, of course, but I don't want to reward Nikon for what I consider to be a horrible marketing decision. Plus I already have the P45+ system for my high MP needs, so I'm reluctant to purchase something else that satisfies the same purpose. Nevertheless, knowing what I know now about how D3 files compare against  P45+ files, and the inconvenience of using two systems, and the inferiority of MF autofocus, metering, lenses, live view, etc, I think a D3X would be a better match for my needs than the P45+ system. If a D700X with the D3X sensor in it ever appears, it'll be a no brainer, and that's basically what I'm waiting for (although it's getting harder and harder to resist the D3X :-)

Agree - I do not need 39Mp.  I shot the same subject with a D700 + Nikkor 24-70 at about 45-50mm and H3D-II 39 with an 80mm.
Looking at the files on a computer screen without blowing up the images I now have trouble figuring out which one is which...

The huge difference I felt were during the retouching of the portraits - with the 39mp files it was a LOT simpler and more convenient to retouch the eyes and skin and the few times when I decided to do agressive crop after the shoot and was able to do it with a 39Mp files, but not with the 12M...

Of course, an H3D also looks a lot more impressive than a D700...

I had also decided to boycut the D3X and wait for a D700X, but I'm rethinking it now...

-ilya

[Panopeeper]

We all have our tricks on handling the files and experience doing real work is very valuable - much more so than pushing the files around in a program that measures noise

This is not the same issue as the dispute over the use or uselessness of 14bit vs 12bit; the 16bit issue of the Phase Ones is not the question of noise. The two low order bits are not usable (not not useful) in the image reproduction.

[Panopeeper]

So you don't like the term - then suggest a more appropriate one and be positive about it rather than heckle and call BS

This is not the question of liking or not. Analogue ISO gain "reveals" image information captured by the sensor but not utilized at lower ISOs; that's the point of having them.

Multiplying the pixel values does not contribute to the useful image information. The current adoption of the fake ISOs with the DSLRs is a sad affair. There is absolutely no need to multiply the pixel values in order to make the resulting image brighter. MFDBs demonstrate this: the fake ISO is nothing but a tag.

It is important to note here, that the "real gain" of the analogue gain is limited: the higher analogue ISOs of almost all DSLRs don't deliver, i.e. their effect is worse than the digital adjustment.

[Graeme Nattress]

The "fake ISO" of DSLRs is done wrong though - they just apply a gain in post A-to-D digital signal before recording it, which brightens the image and throws away a stop. If  they'd left it alone and just metadata tagged +1EV or whatever, all would still be in the RAW and all would be good.

[cjmonty]

Ive extensively used two Canon 21mp bodies, and now shoot with a Phamiya P45 refurb with assorted Mamiya glass.  With Canon i used Camera Raw or Lightroom, with Phase I use C1 4.7.

There is no comparison in image quality.  The Phase files are on a whole other level of clarity.  And not just in terms of 21MP vs 39MP relative size.

A Phase-derived image looks like my old 4x5 prints in describing detail.  The Canon prints should not be printed past 200% base resolution.

I could have been misusing the Canon gear, or have gotten two bad cameras, or should not have processed the files in Lightroom, but in my working experience some people are better served by the Canon and Nikon gear, and some are better served by MFDBs.  

However, neither is unequivocally better.

There are some very good reasons to stick with Canon

1- Not everyone prints large enough so the Canon dough-ball looking pixels become apparent.  When the prints are small, an invisible pixel is an invisible pixel.  If I didn't require a certain amount of detail in my work, I would have happily kept my Canon.  And several thousand dollars.

2- Much of the deficiency with Canon gear may be in the lenses, which when sharp, don't seem that sharp.  However, the whole problem is probably the AA filter.  As far as I can tell.  The description detail in a Canon image is rather fugly.  However, I have only had my work to judge this on... I would love to see a scientific comparison of Canon primes vs Mamiya vs Rodenstock/Schneider digitals in Helical mounts all compared on one sensor.

3- If I shot action, needed lowlight, or needed any real photographic versatility, I would happily have kept my the Canon.  And several thousand dollars.

[Panopeeper]

The "fake ISO" of DSLRs is done wrong though - they just apply a gain in post A-to-D digital signal before recording it, which brightens the image and throws away a stop. If  they'd left it alone and just metadata tagged +1EV or whatever, all would still be in the RAW and all would be good.

There is no justification for what the DSLR manufacturers are doing.

Some stick to the numerical range of the bit depth and fill that by multiplication; for example the Sony A700 multiplies the raw values at ISO 200 as well. In effect the camera fakes 12bit depth, when it uses in fact only less than 3200 levels.

The Nikon D300 stretches the range of red and blue by about 20% to "reach" the green range, instead of applying different white levels

The Nikon D200 stretches the red by 18.5% and the blue by 17%; the effect is, that the green clipping level (not a fixed value) is lower than the red and blue.

Canon is the worse: the clipping level of the newer models depends on the ISO, but still the fake ISOs are achieved by multiplication instead of by different white levels.

All this with the only effect of reducing the dynamic range. I would not believe it if I did not prove it myself.

[EricWHiss]

This is not the same issue as the dispute over the use or uselessness of 14bit vs 12bit; the 16bit issue of the Phase Ones is not the question of noise. The two low order bits are not usable (not not useful) in the image reproduction.

Come on  - the two lowest bits represent 0, 1, and 2 values out of a total of 65536.  Who gives a cr*p about that? Most people adjust the black point higher anyhow even if they push up the shadows.  The problem is chromatic noise and blob type of noise and banding up in the 1/4 tones of the image, not those two last bits.  And back to the original topic - I am not certain you can take DXO's numbers across from DSLR to MFDB like apples to apples.  


As far as DR - thanks for clarifying how DXO is defining DR.  I did not know they were using anything other than the technical definition as the numbers they report seem too high overall to be anything else.

As to Fake ISO -  actually I understand why a camera company would do this - its just easier to operate the camera this way with exposure metering and exposure checking on the histogram.   If I were making a camera I'd probably do the same thing.

[Panopeeper]

Come on  - the two lowest bits represent 0, 1, and 2 values out of a total of 65536

and 3 :-( (i.e. four values out of 65536). On the other hand, this is not an issue of bits, that is only a convenient way of expressing such aspects.

Who gives a cr*p about that?

Let's keep things straight.

1. I did not criticize the image quality of the Phase One cameras. In fact I regard those cameras very highly.

2. I stated, that the 16bit depth is in practice only 14bit. You became outraged reading that, and now you are saying "who gives a cr*p about that".

So, pls make up your mind against what and for what you are arguing, i.e. if those bits are useful or not, or perhaps not even worth of arguing about.

Most people adjust the black point higher anyhow even if they push up the shadows

You can't adjust the black level in terms of raw pixel values. Nevertheless, you are right in that those pixels are not used in most cases. I wrote before: The two low order bits are not usable (not not useful) in the image reproduction.

The problem is chromatic noise and blob type of noise and banding up in the 1/4 tones of the image, not those two last bits

Some of the banding (or all of it) is caused by those bits (i.e. by the far too great depth of the data, which represents tiny sensor related issues, not image information).

I am not certain you can take DXO's numbers across from DSLR to MFDB like apples to apples

Actually, I don't take DxO's nor anyone else's numbers at all. I prefer my own measurements, which I can document and demonstrate. If I don't have raw files suitable for such measurements, then I don't make any related statement.

[Panopeeper]

Eric, I was writing my post and have not noticed that you added to your previous post-

As far as DR - thanks for clarifying how DXO is defining DR

Hold on; I described, how I do the measurements (the graphs I linked to are mine). I don't know how exactly to interpret DxO's measurements.

As to Fake ISO -  actually I understand why a camera company would do this - its just easier to operate the camera this way with exposure metering and exposure checking on the histogram.   If I were making a camera I'd probably do the same thing

Don't be so sure you would do the same.

One of the fanciests is the Canon 5D2. I simply copy the definitions from the Rawnalyze program code. The second column in the red clipping level, the third is the green and the fourth is the blue; the fifth plays no role here. As you see, there are different clipping levels, depending on the ISO. Is that what you think makes it easier for the firmware?

ISOSat(25600,16383, 16383, 16383, 16384)
ISOSat(12800,16383, 16383, 16383, 16384)
ISOSat(6400,   16383, 16383, 16383, 16384)
ISOSat(5000,   15755, 15760, 15760, 15800)
ISOSat(4000,   15755, 15760, 15760, 15800)
ISOSat(3200,   15755, 15760, 15760, 15800)
ISOSat(2500,   15755, 15760, 15760, 15800)
ISOSat(2000,   15755, 15760, 15760, 15800)
ISOSat(1600,   15755, 15760, 15760, 15800)
ISOSat(1250,   12810, 12810, 12810, 12850)
ISOSat(1000,   15755, 15760, 15760, 15800)
ISOSat(800,   15755, 15760, 15760, 15800)
ISOSat(640,   12810, 12810, 12810, 12850)
ISOSat(500,   15755, 15760, 15760, 15800)
ISOSat(400,   15755, 15760, 15760, 15800)
ISOSat(320,   12810, 12810, 12810, 12850)
ISOSat(250,   15755, 15760, 15760, 15800)
ISOSat(200,   15755, 15760, 15760, 15800)
ISOSat(160,   12810, 12810, 12810, 12850)
ISOSat(125,   15755, 15760, 15760, 15800)
ISOSat(100,   15755, 15760, 15760, 15800)
ISOSat(50,   15755, 15760, 15760, 15800)

and then the Nikon D3X:

ISOSat(6400,4095,4095,4095,4096)
ISOSat(3200,4095,4095,4095,4096)
ISOSat(1600,4095,3830,4095,4096)
ISOSat(800, 4095,3830,4095,4096)
ISOSat(400, 4095,3830,4095,4096)
ISOSat(200, 4095,3830,4095,4096)
ISOSat(100,   4095,4095,4095,4096)
ISOSat(50,   4095,4095,4095,4096)
ISOSat(0,   4095,4095,4095,4096)

[Guillermo Luijk]

Especially the DR results make me think this. One hears repeatedly how much cleaner the shadows are with MFDBs, but this somehow doesn't translate into better DR results, counter-intuitively, I suppose because what we like and what we can measure don't match.

Are you sure those who say how much cleaner the shadows are with MFDBs (funily without specifying which MFDB nor which DSLR) really know how to evaluate noise in the shadows of their MFDBs? even if they are MFDBs selling staff who think they know what they are talking about, or they are people making a living on using these backs and thinking making a living on using their backs is the only way to have a founded opinion.

This is a 100% crop of a H3D II:


The black figure is located in the RAW range 9 stops from saturation. I am pretty sure the level of noise in the 9th stop on a Nikon D3X is the same or lower than this.

The sensor in a MFDB is larger than on a DSLR, that means with the same tecnology you can build a higher DR MFDB than the equivalent DSLR. But that does _not_ mean that all MFDBs will provide more DR than all DSLRs.

BR

[Carsten W]

The black figure is located in the RAW range 9 stops from saturation. I am pretty sure the level of noise in the 9th stop on a Nikon D3X is the same or lower than this.

It is an interesting image, but you left out a lot of information. At what ISO is this? Which H3DII is it? Do you have a similar crop from a D3X to demonstrate your assertion?

[Guillermo Luijk]

It is an interesting image, but you left out a lot of information. At what ISO is this? Which H3DII is it? Do you have a similar crop from a D3X to demonstrate your assertion?

Filename: hassel.dng
Timestamp: Thu Nov 13 16:54:51 2008
Camera: Hasselblad H3D II-31
DNG Version: 1.0.0.0
ISO speed: 100

Anyway ISO is irrelevant here, since all ISOs in this back are ISO50.
I have never seen any RAW file from Nikon D3X but I did for a Nikon D3 having similar noise in the 9th stop (signed as -8EV here):



I think the quality advantage for backs is not noise, DR, colour rendition,... but just sharpness and resolution, i.e. detail. In the same way as a Leica M8 with an average sensor can produce very good quality images thanks to the lack of AA filter and superb optics, backs have the same strongholds: no AA filter and good lenses. The sensor in backs is not better than in DSLRs.

BR

[David Grover / Capture One]

Filename: hassel.dng
Timestamp: Thu Nov 13 16:54:51 2008
Camera: Hasselblad H3D II-31
DNG Version: 1.0.0.0
ISO speed: 100

Anyway ISO is irrelevant here, since all ISOs in this back are ISO50.
I have never seen any RAW file from Nikon D3X but I did for a Nikon D3 having similar noise in the 9th stop (signed as -8EV here):



I think the quality advantage for backs is not noise, DR, colour rendition,... but just sharpness and resolution, i.e. detail. In the same way as a Leica M8 with an average sensor can produce very good quality images thanks to the lack of AA filter and superb optics, backs have the same strongholds: no AA filter and good lenses. The sensor in backs is not better than in DSLRs.

BR

Converting to DNG will see a significant rise in noise levels compared to retaining our 3F format and using Phocus.  Therefore showing such an image is an unfair evaluation.  I think this shows as that is a truly appalling example of a 100 ISO shot.  What did you do to make it so unusable?

Also saying 'all ISOs in this back are ISO50' is misleading to customers.  As I have said many times before on these frustrating threads is that when a photographer sets ISO400 or whatever on his camera, he/she wants it to behave in this way and give feedback via the light meter and histogram to that effect.

How the image is then translated via the DSP in the back (which I will say again varies depending on the ISO setting) and furthermore in the processing software to the end result is of little concern to many.

ISO is probably an ancient term which should not be used with digital, but it is what we know and is what is familiar to us photographers.

Best,



David

[Guillermo Luijk]

Converting to DNG will see a significant rise in noise levels compared to retaining our 3F format and using Phocus.

Brilliant.

[Carsten W]

Anyway ISO is irrelevant here, since all ISOs in this back are ISO50.

Well, no. If you shoot at "fake" ISO 100 or 200 or 400, you are under-exposing, in effect, and thereby reducing the quality of the image. This is not irrelevant at all. Given that MF solutions are inherently much less flexible than DSLRs, there is only one reason to go the MF route, and that is to try to extract the absolute best image out of it that you can, and this means shooting at base ISO, IMO. If you do anything else, or get sloppy with processing, yes, you may as well use a DSLR. If you do go the extra mile, you should be rewarded with visibily better detail and shadows, from what I understand, but there are no shortcuts (until the S2 ).

I am interested in the A900 because it doesn't have Canon reds or Nikon greens, and the AA filter is apparently noticeably less strong than with either of those competitors, perhaps excepting the D3X, which is as expensive as a second-hand MF back in any case. Of course, there are also the Zeiss-Sony lenses to consider, which for me could be a big deal. The A900 is also meant to have very good skin tones. I am interested in what situations you are likely to miss the extra quality of the DB. So far, it seems that printing larger than A2 is one situation. Maybe there are others?