Nikon D3X DXOMark Numbers Blow the Sony A900 Away!

[David Kilpatrick]

I dug out my 2007 report on the Sekonic meter which offers a kind of measurement of camera ISO sensivity, which you use to program the meter to match your camera:

http://www.dphotoexpert.com/2009/01/18/ ... libration/

I've put this up on dPhotoexpert, as although it uses illustrations from KM5D and Alpha 100, it's not really that relevant to Alpha alone. The software calculates effective ISO values on the basis of an 18% grey target, which is very different from the basis used by DxO. It should be pretty easy to replicate what their software does with a simple Excel spreadsheet incorporating a few calculations which are, fortunately, beyond me - means I don't waste my time creating one!

David

[Greg Beetham]

I dug out my 2007 report on the Sekonic meter which offers a kind of measurement of camera ISO sensivity, which you use to program the meter to match your camera:

http://www.dphotoexpert.com/2009/01/18/ ... libration/

I've put this up on dPhotoexpert, as although it uses illustrations from KM5D and Alpha 100, it's not really that relevant to Alpha alone. The software calculates effective ISO values on the basis of an 18% grey target, which is very different from the basis used by DxO. It should be pretty easy to replicate what their software does with a simple Excel spreadsheet incorporating a few calculations which are, fortunately, beyond me - means I don't waste my time creating one!

David

David I went over there and read that article at normal speed....and decided that I must have missed some crucial point and went back and read it again slowly to try and get the gist of what was being done and why. Tell me if I've got this right, I think the meter doesn't actually "test" the camera ISO against some industry standard or anything, it seems too be more an elaborate exposure meter that can be calibrated to match your camera ISO deviation/s, and then presumably give adjusted exposures....and from that you can draw (good enough for the bush) conclusions about the accuracy of the ISO's in a given camera.
Greg

[David Kilpatrick]

It does work by measuring dynamic range as well as the grey midpoint, but it is as you suggest entirely a way of calibrating a meter to match the real ISO speed of each camera setting. There is no way you can use it recalibrate the camera, or even transfer the values. You just find out that when you set 100 on your camera, the software is telling you it's really 133, and then you program that into the meter so when you set 100 on the meter it is also really 133 (etc). You can use the software to tell you the true ISO of your camera and raw converter, or your camera's in-cam JPEGs.

David

[Greg Beetham]

Ah so, but I guess you could always run a bit of a test with a decent ambient light meter as well/instead, like point the camera at the sky away from the sun in S mode ISO100 and compare the f/stop with the light meter @ the same shutter speed setting ...too get an idea of how accurate the camera is...etc. but the only problem there might be the meter doing an f/stop readout in digital 0.01 increments, far finer than the camera scale, there is the anaolg scale though in third f/stops..that might be good enough.
Greg
ps I also read the "Why the D3X will still sell" article..a damn good read actually.

[David Kilpatrick]

But, as the Russian raw converter project guys (Iliah Borg etc) point out, there is a huge flaw in the method outlined. The flaw is that the three colour channels have different sensitivities, determined by the filter set and by individually tuned channel gain in early stage conversion. So the colour of light needed to make all four Bayer components - R, G1, G2, and B - max out at 4096 (or 4096 less fixed headroom) - may not be daylight, or tungsten, or any other standard light source. The gain and sensitivity differences as seen in the final RGB channels mean that effective digital camera ISO becomes dependent on the spectral energy profile of the light used.

Borg has started shooting his outdoor work on the Alpha 900 using an 85 (tungsten conversion) amber filter because he thinks it's as close as you can get, off the shelf, to a slightly more magenta filter pack needed to equalise the A900's channel sensitivity and kill the excess gain applied to the R channel. By doing this, he's been able to reduce noise levels.

I guess it makes sense if you consider that high ISO low light shooting is most likely to be done by tungsten light. So Sony make the filter 'native' to tungsten light, and daylight exposures are suffering as a result.

If Iliah Borg's findings are correct - and I think they are although in the past I found the reverse applied with the KM5D, it paid to convert tungsten to daylight using a blue filter - then really DxO, and others, should be trying to identify the native Kelvin + LB/CC equienergy response of the sensor, and stating the ISO for this, for daylight, and for tungsten.

I have just dug out a box of No85 filters, I'll be trying this and seeing if there is any difference except the annoyance of seeing everything through an amber filter and losing 1.5 stops of light.

David

[David Kilpatrick]

I like how you poo-poo the Borg's on some things but now you are trying to use them as backup info!

Not aware of having poo-pood the Borgs, though they have made a negative comment about me somewhere. I think Raw Photo Processor (by Andrey Tverdokhleb, not apparently by Iliah though he is clearly involved) is unusable for any practical purpose, although I have donated to it and value the research they have been doing. Since RPP is freely downloadable and the donation is voluntary, it's easy enough to try. I may make another donation rather than email Andrey to unlock some advanced features (an option he offers). I have used it for investigating files but the interface is seriously obscure and operation is very slow. I also found the colour rendering very dull (I would not suggest RPP as the ideal Sonolta raw converter!).

Most of Iliah's input to dPreview is geared to promoting RPP. I think there is a more ambitious plan behind it all anyway, they clearly have in mind the creation of a commercially viable raw processor, or the licensing of some of work to a camera maker. If everything they have put into RPP and the http://www.libraw.org" onclick="window.open(this.href);return false; project was taken up and studied properly by whoever creates Sony IDC, it would be a good thing. We might end up with decent own-maker raw converters.

One software developer does appear to have the same approach - Iridient, with Raw Developer. Iliah doesn't really like people using RD instead of RPP. RD is a fully usable, fast, user-friendly program where RPP is not.

David

[Bengeo]

Interesting new article here: http://www.luminous-landscape.com/essay ... bers.shtml.

Andy

[Dusty]

Interesting new article here: http://www.luminous-landscape.com/essay ... bers.shtml.

Andy

I don't want to entirely resurrect this thread, but the author seems to make most of the point, but doesn't name it. That point is that is that sometimes the whole is greater than a sum of it's parts. DXOMarks also, as he notes, doesn't take into account resolution, which begs the question, if you don't take that into consideration, how do measure overall image quality?

Give we the best 35mm and 35mm film, and then give me an 8x10 camera and mediocre film, and lets see which shots look better. Some will same that's not the same thing, but it is. The 8 x 10 records more information than the 35mm. More recorded information means better results.

Dusty

[01af]

... how do [they] measure overall image quality?

They don't.

-- Olaf

[01af]

In the context of an ongoing discussion on the Luminous Landscape user forum about DxOMark's latest ratings for some medium-format digital backs' (MFDB) sensors and whether the current 35-mm full-frame DSLRs can or cannot outperform MFDBs, Michael Reichmann made an interesting remark about the Sony A900 and the Nikon D3X which he both used on his latest trip to Antarctica earlier this year.

He says he cannot see any difference whatsoever between the prints made from A900 files and those made from D3X files, at print sizes up to 24" × 36" (60 × 90 cm) and at any ISO settings up to ISO 1600/33°. Furthermore he says that at medium print sizes 35-mm full-frame DSLRs are pretty close to MFDBs but in large prints the difference is obvious, clearly favouring the MFDB. See his post here; the discussion as a whole (not really worth reading) starts here.

So the lesson to learn is this: DxOMark numbers have no connection to what we see as image quality. When shopping for a digital camera, they are grossly misleading, so better ignore them.

-- Olaf

[01af]

When comparing the Sony A900 to the rest of the current 35-mm full-frame DSLR crowd then noise usually is considered the A900's weakest point.

In this context, the test results of PopPhoto.com (the website of the U.S. photo magazine 'Popular Photography') are interesting. According to their tests of the A900 and the D3X, the A900's noise actually is less than or equal to the D3X's---at ISO 100/21° to ISO 1600/33° (with the exception of the ISO 800/30° setting---it seems as if in the D3X some kind of in-camera noise reduction kicks in at ISO 800/30°). At ISO 100/21°, the A900's noise is even less than the D3X's at ISO 50/18°.

At ISO 3200/36°, the D3X has a significant but not tremendous advantage over the A900. Only at ISO 6400/39° the difference is huge, in favour of the D3X---here, the A900's noise is rated 'unacceptable' while the D3X's noise still is 'moderate.' To see for yourself, click here (A900) and here (D3X).

Basically these results are consistent with what Michael R. said about the quality of prints from the A900 and the D3X at ISO settings up to 1600/33° (see previous post). So with the exception of the ISO 6400/39° setting, noise is no weak spot of the A900 when compared to the competition. However I am somewhat surprised about PopPhoto's results regarding colour accuracy---here the D3X clearly outperforms the A900, according to their tests. Nikon colour better than Sony colour? Hmmm ...

Personally, I feel the A900's performance even at ISO 6400/39° actually is pretty good. It's better than the Dynax 7D at ISO 3200/36°, and it sure is way better than anything we were able to get from high-speed film. The fact that some other DSLRs are better still doesn't detract from that. By the way, I found an easy and fairly effective procedure to reduce noise in ISO 6400/39° raw images using ACR and Photoshop only (no noise-reduction software like Neat Image or Noise Ninja required). Here it is:

In Camera Raw, set Luminance Noise Reduction to 25 and Color Noise Reduction to 30. In the workflow options, select 3072 × 2048 (6.3 MP) as the target size if possible (you may skip this step if you absolutely need full resolution but at the reduced size noise will be suppressed even more effectively). Tweak the other settings as required, then open the image in Photoshop. Apply the noise reduction filter (Filter > Noise > Reduce Noise) with the settings Strength=7, Preserve Details=20, Reduce Color Noise=75, Sharpen Details=20. Then apply the surface blur filter (Filter > Blur > Surface Blur) with the settings Radius=40, Threshold=10. Immediately after this, fade the blur command (Edit > Fade Surface Blur) with the settings Opacity=75, Blending Mode=Color. The result will be an image with some fine-grained luminance noise which looks fairly film-like, greatly reduced colour noise, and almost no loss of fine detail. I wrapped the three Photoshop commands into an action called 'NR ISO6400.' Feel free to experiment with the particular settings.

-- Olaf

[pakodominguez]

I may, for interest, get the Sekonic kit for camera ISO calibration for a second time (I have tested this once already, but they have introduced a better target and new software since then).

David

In the USA, Phil Bradon, Ex Minolta "pro" support, is the Sekonic man. You can find a couple of videos on Youtube were he explain the process you describes here. I had the chance to see him last october at the Photo Plus Expo. A beautiful and coherent person, he follows his passions: first Minolta equipment, now light and color meters.
Look at:
http://uk.youtube.com/watch?v=_7SZ58CugpY" onclick="window.open(this.href);return false;
http://www.aftercapture.com/videos/75/S ... kflow.aspx" onclick="window.open(this.href);return false;
http://www.aftercapture.com/videos/60/I ... konic.aspx" onclick="window.open(this.href);return false;

[pakodominguez]

In Camera Raw, set Luminance Noise Reduction to 25 and Color Noise Reduction to 30. In the workflow options, select 3072 × 2048 (6.3 MP) as the target size if possible (you may skip this step if you absolutely need full resolution but at the reduced size noise will be suppressed even more effectively). Tweak the other settings as required, then open the image in Photoshop. Apply the noise reduction filter (Filter > Noise > Reduce Noise) with the settings Strength=7, Preserve Details=20, Reduce Color Noise=75, Sharpen Details=20. Then apply the surface blur filter (Filter > Blur > Surface Blur) with the settings Radius=40, Threshold=10. Immediately after this, fade the blur command (Edit > Fade Surface Blur) with the settings Opacity=75, Blending Mode=Color. The result will be an image with some fine-grained luminance noise which looks fairly film-like, greatly reduced colour noise, and almost no loss of fine detail. I wrapped the three Photoshop commands into an action called 'NR ISO6400.' Feel free to experiment with the particular settings.

-- Olaf

Thanks Olaf,

Next time I'll have the A900 I'll try your method. Do you think it will work as well with A700's files?

[01af]

Do you think it will work as well with A700's files?

Basically yes. Sure. Maybe you'd have to adjust some or all of the settings for optimal results, but the basic principle should work with any raw file that suffers from strong noise.

-- Olaf

[Dusty]

Don, I thought you had had enough and left. What happened, the other boards got tired of your ranting and kicked you off?

We were all having such a wonderful time discussing things without the personal attacks.

Why don't you go buy that $8000 D3X and then you can go haunt the Nikon boards! Or just take your meds and join in normal discussions with the rest of us.

Dusty