Another view DxO

[ErikKaffehr]

Hi,

I decided to post this as a new topic so it will generate more discussion.

My view is that we should not look at the DxO mark as a figure of merit but the individual measurements. If we do that and compare the Phase One P45+ with the Sony Alpha it is quite obvious that the P45 has an advantage of say 3-4 dB in all areas at minimum sensivity. The sole reason that the Sony Alpha rates higher is that it has higher usable sensivity, and who would doubt that?

One thing that DxO mark seems to indicate clearly that higher ISOs on the P45+ are not for real, no more amplification just underexposure.

The DxO mark quasi ignores that fact that the P45+ plus has about 60% more pixels, so if the A900 is good for A1 enlargements than the P45+ would be good enough for almost A0 and still offering a better image quality than the A900 at the pixel level as long as we are shooting at optimal ISO.

Maybe DxO is a good thing, if we just ignore the single figure of merit and look at the real data.

If we compare the P45+ with Nikon D3x the Nikon is a little better on noise (1 dB at minimum ISO) but the pixel ratio is still the same.

One area where the Nikon shines in DxO test is DR (Density Range). It is less than obvious to me how they have achieved that especially as it seems to be widely assumed that Nikon uses same sensel technology and readout method as the A900. I had a guess that Nikon makes 14-bit readouts by doing a double readout using different on chip preamp settings, but people much more knowledgeable than me have pointed out on this forum that doing that would leave tell-tale signs all over in the "raw" file. A simple truth is that we cannot represent a very large DR in prints. So even if the Nikon files have a real DR advantage if all candidates involved have a DR which is good enough the difference may not be visible in print.

My conclusion is that DxO mark may be entirely relevant when we look at the measurement data at "pixel level". On the other hand I cannot compare pixels between say Phase One P45+ and Nikon D3x, of which I have neither, nor do I have the necessary experience. Michael Reichmann has all the stuff and experience needed, so I have a great trust in his judgments.

One area that I hope that Michael Reichmann will discuss is the difference between different cameras in relatively small prints. Michael Reichmann has himself stated quite clearly that differences between MFDBs and DSLRs are visible in small prints. I obviously don't have the MFDB experience but experiments i have made with the equipment I have sort of indicated that I need to print big to be able to see difference which are very much visible in pixel peeping.

Best regards
Erik

[attachment=11358:Noise_cmp.jpg]

[Panopeeper]

One thing that DxO mark seems to indicate clearly that higher ISOs on the P45+ are not for real, no more amplification just underexposure

This is not the case - just this point makes the review suspect for me (I posted it already pn another thread).

[ErikKaffehr]

Gabor,

Can you point to that posting?

I got the impression from other postings that MFDBs don't really have different ISOs which would coincide with DxO figures. It seems that your measurements tell a different story, considering that you are one of the few who actually measure things, this is quite interesting.

I got the impression that you are getting more and more critical of the DxO data, not just the DxO marks but also the underlying data?


Best regards
Erik

This is not the case - just this point makes the review suspicious for me (I posted it already pn another thread).

[NikosR]

A simple truth is that we cannot represent a very large DR in prints. So even if the Nikon files have a real DR advantage if all candidates involved have a DR which is good enough the difference may not be visible in print.

Yes we can demonstrate a DR difference in print. It's called dynamic range compression and its what all photographic printers have been doing longer than I care to remember (since BW film has a greater dynamic range than paper). This is what tone curves do. How clean the shadows are when lifted while still maintaining detail is where you will have to look to see it evidenced.

Another way to put it. HDR, which indirectly increases the captured DR by combining many captures with different exposures, can be represented in print. Dynamic range compression is used again (often misused pictorially to produce those garish 'HDR looking' pictures).

There's no such thing as good enough DR. For some scenes this notion might exist, but nature in general hits us with much greater dynamic range than we can handle so any advance in that respect is welcome.

[Panopeeper]

Can you point to that posting?

Not worth of it, I wrote only so much there as here.

I got the impression from other postings that MFDBs don't really have different ISOs which would coincide with DxO figures

This was so for a while, but now it is not a common characteristic of MFDBs. Some Phase One models do have analog gain, others not; I know of one, which has analog ISO 50 and 100, the higher speeds are digital.

The P45+ does have analog ISO, up to 800.

I got the impression that you are getting more and more critical of the DxO data, not just the DxO marks but also the underlying data?

There is no trend there, I have nothing against their methods. However, there are some issues where I know that their publication is incorrect. Unfortunately they do not publish their raw files and the exact measurements, like Imaging Resources does.

[BJL]

Maybe DxO is a good thing, if we just ignore the single figure of merit and look at the real data.

Agreed; that is what I suggest in another thread: look at each measurement separately. That done, I even agree to normalizing for resolution difference, as this then gives a measure of what you would see in comparisons of equal sized prints. So long as resolution is added as a fourth item on the list of measurements! It is when these three measures, scaled to be independent of resolution, are then lumped into a number that is treated as an overall measure of virtue that the wheels come off!

I would even add a few more items to my score sheet:
- size and weight
- cost
But these do not need to be measured by DxO, and perhaps resolution does not either: pixel count or "pixels per picture height" is probably a sufficiently precise measure, since I suspect that variations in "resolution per pixel" are mostly so small as to be in the realm of the false precision that Michael rightly criticizes.