DXOMark just tested the D3X

Started Jan 15, 2009 | Discussions
rhlpetrus Forum Pro • Posts: 25,860
Re: Here I am

Iliah Borg wrote:

I also would like to see your findings

We will do better in due time, we will provide a tool to do such
measurements in an accurate manner and compare those measurements
taken by different testers.

I'll be looking forward to that, but just one question at this point: you don't see the 13.7 EV of DR or you don't see any improvement over the A900?

-- hide signature --

Renato.
http://www.flickr.com/photos/rhlpedrosa/
OnExposure member
http://www.onexposure.net/

Good shooting and good luck
(after Ed Murrow)

 rhlpetrus's gear list:rhlpetrus's gear list
Leica D-Lux (Typ 109) Nikon 1 V1 Nikon D7000 Nikon 1 Nikkor 10mm f/2.8 Nikon AF Nikkor 35mm f/2D +4 more
rhlpetrus Forum Pro • Posts: 25,860
Re: Thanks Frank!

Iliah Borg wrote:

DxO do not measure and present the visual impact of noise. Same noise
numbers can correspond to fine grain agreeable and ugly blotchy
noise. Noise numbers are different for files white-balanced for
different colour temperatures, they do not test those however. Noise
should be measured separately for luma and chroma, that is also
skipped.

Is that discernible at RAW level? I thought chroma noise was actually a post conversion issue. I mention that since I've seen some rather different results for some cameras when using different RAW conversion softs.

It matters how many gradations of grey are recorded in the last stop.
Some of the tested cameras make the last stop barely separable from
the previous while others allow for the practical use of the last
stop. For example, if you need to open shadows on A900 it is better
to sacrifice some DR but shoot as ISO 400 to ISO 640. DxO tests won't
show you this because they do not have the criteria for the
photographic usefulness of the last stop.

Colour measurements they do are a poor man's substitution of sensor
spectral characteristics, and are unusable as such, or as a tool to
compare sensors. The more meaningful numbers are numbers of metameric
faults by band, say, how many metamers are in the hues that represent
skin, foliage, sky.

Mathematics and physics are to be applied to the subject of
photography, results should be meaningful and reported in the
language of photography; and here DxO tests are lacking.

Agreed on that, but I think DxO is tryong to keep things at the pure sensor level, i.e., before any rendering is performed. To evaluate what you mention you need some reference for post-rendering evaluation, a different subject, no?

-- hide signature --

Renato.
http://www.flickr.com/photos/rhlpedrosa/
OnExposure member
http://www.onexposure.net/

Good shooting and good luck
(after Ed Murrow)

 rhlpetrus's gear list:rhlpetrus's gear list
Leica D-Lux (Typ 109) Nikon 1 V1 Nikon D7000 Nikon 1 Nikkor 10mm f/2.8 Nikon AF Nikkor 35mm f/2D +4 more
Daniel Browning Senior Member • Posts: 1,058
Metameric faults and chroma noise

Iliah Borg wrote:

Colour measurements they do are a poor man's substitution of sensor
spectral characteristics, and are unusable as such, or as a tool to
compare sensors. The more meaningful numbers are numbers of metameric
faults by band, say, how many metamers are in the hues that represent
skin, foliage, sky.

Agreed. It would be really nice if those were published more often.

Noise should be measured separately for luma and chroma, that is also
skipped.

Read noise is the same for all pixels, so any difference between luma/chroma can only be attributed to the converter.

Photon shot noise, on the other hand, can vary by color channel, due to the spectral response shaping by the CFA.

-- hide signature --

Daniel

Iliah Borg Forum Pro • Posts: 25,835
Re: Metameric faults and chroma noise

Noise should be measured separately for luma and chroma, that is also
skipped.

Read noise is the same for all pixels

That is correct (most of the cases, that is if the paths for the pixels under differently coloured pixels is the same).

so any difference between
luma/chroma can only be attributed to the converter.

Actually, in a sense this is not entirely true. Take 2 shots with different cameras under the same light, do the binning, convert to Lab and you will get meaningful numbers to compare.

-- hide signature --
m_appeal Veteran Member • Posts: 3,434
Do these numbers mean you'll see a real difference between D3X

and A900 at base ISO? I doubt it...

Dan Wells Contributing Member • Posts: 571
Re: Do these numbers mean you'll see a real difference between D3X

DxO always seems optimistic to me, but their great advantage is that they're consistently optimistic (by about two stops, to my eye). I don't see any 12.84 stops (DxO's number once you uncheck the "print size conversion", which complicates their algorithm, out of a D3x (more like 11 or a bit less), but I never got 10.79 stops out of a 1Ds mkII either (somewhere between 8 and 9). DxO does accurately capture what I perceive to be a close to 2 stop difference between those two cameras (both very familiar to me) in large prints. DxO numbers line up pretty closely with Imatest's most optimistic numbers (of four reported), while my real experience with cameras in the field tends to be closer to Imatest's "high" image quality, their most conservative number.

-Dan

m_appeal Veteran Member • Posts: 3,434
You are acting like this is some sort of magic proof

in the pudding of D3X's superior image quality at lower ISO.

EKaffehr Forum Member • Posts: 75
Re: for what it's worth... 3Dx and H3D-31

Hi,

Just a question. The 3Dx obviously needs more sharpening than H3D-31. What is your impression of image quality comparing 3Dx and H3D-31 with optimal sharpening and possibly moiré removal on the H3D-31?

Best regards
Erik
--
Erik Kaffehr

photoforfun Veteran Member • Posts: 6,084
Re: Detail /// thanks for posting this...

Hi Mr. Languillier,

I was testing a friends D3X and D3 side by side( http://www.fotografie.fr/n3-test4-e.htm ) but we never had the opportunity to concentrate on a real DR sample, we were mostly concerned about noise and detail compaison. After having seen your sample we will very soon update our little review.

The sample you posted is very well chosen and the DR provided by the D3x seems incredible. Thanks for taking the time and share.

I made a link to this thread on my webside-forum with a link to your original picture, hope you don't mind. I send you an email as well.

Thanks and kindest regards,
Stany
http://www.fotografie.fr/

I prefer one really good picture in a day over 10 bad ones in a second...

bloomoose
bloomoose Veteran Member • Posts: 3,224
S5 is unique ...

... it excels above middle grey up to white. You can recover blown highlights that are lost on all other DSLRs including the D3X. I really don´t care too much about noise-free shadows, it´s above middle grey where most of my images live.

That´s why DXO numbers really don´t tell you anything about photography.
Seems like a sensor tech thing, and everyone takes it as gospel.
--
Greetings from Germany,
Pam

Check out my images @
http://www.fotografie.fr/n3-galleryPMeier-1.htm

 bloomoose's gear list:bloomoose's gear list
Canon G9 X Sony RX100 Nikon D750 Sony a7R II Nikon AF Nikkor 105mm f/2D DC +7 more
jean bernier Veteran Member • Posts: 3,181
Re: for what it's worth... 3Dx and H3D-31

EKaffehr wrote:

Hi,

Just a question. The 3Dx obviously needs more sharpening than H3D-31.
What is your impression of image quality comparing 3Dx and H3D-31
with optimal sharpening and possibly moiré removal on the H3D-31?

Sorry, I can't add more to what I already stated. I had not much time to work with the two cameras and to optimize settings, and I did not have the best tools to convert the raw files (View NX for the D3X)

Phocus-for-PC was published just recently, but did not work properly on my system, so that I did not have access to the demoiré feature that was demonstrated at the dealer's studio on an apple. So I'm not posting samples, as I'm sure to get critics such as "invalid badly executed test, why did you not...blablabla" Nevertheless, I managed to get excellent jpg's out of the raw files, good enough to judge noise and sharpness on full lenght portraits, and push the sharpening to the maximum acceptable on both cameras. The H3DII-31 has visibly more sharpness, though not a quantum leap, but this comes at a huge price in money as well as convenience, in my opinion, to the extent that it looks to me somewhat like a loosing proposition. I can see a D3X added to my arsenal in the near future. As to the Hasselblad, the 39 mpix looks like a more interesting proposition.

The D3X files can take a huge amounts of sharpening as I said, in order to reduce the AA filter effect on sharpness, whithout falling apart, and comes close to the image quality of the H3DII-31.

If you look at image size, the D3X is 4032 x 6048 and the H3DII-31 is 4872 x 6496, not such a big difference in linear measurements.

The hasselblad sub-par AF system was a deal breaker for us. It certainly works, as long as the conditions are met for critical focussing: little to no focus/reframe, centered subjects, or very careful manual focus. Some scenarios present no particular problems, such as group shots, many landscapes, full lenght portraits with the camera at waist level, etc...
Just food for thoughts...

Best regards
Erik
--
Erik Kaffehr

-- hide signature --

Jean Bernier

All photographs are only more or less credible illusions

Artichoke
Artichoke Forum Pro • Posts: 12,479
I think Nikon picked up some Fujifilm magic

and put it into the D3X

bloomoose wrote:

... it excels above middle grey up to white. You can recover blown
highlights that are lost on all other DSLRs including the D3X. I
really don´t care too much about noise-free shadows, it´s above
middle grey where most of my images live.

I skipped the S5, but used the S3 a great deal

shooting with my D3X reminds me of the S3 & while I think the S3 may still have more useable DR when shot RAW, the D3X is very close & forgiving, particularly at low ISO
also similar to the S3, the jpg engine is superb in the D3X

shot at Neutral settings with noise reduction at minimal settings & sharpening turned off, the D3X produces a very resilient 10 MB jpg file that takes a great deal of PS manipulation without breaking up
given the beta state of NX2, this is particularly important & welcome

from my testing, shooting jpgs with the D3X imposes very little loss in file quality & is suitable for 95 % of my needs

That´s why DXO numbers really don´t tell you anything about photography.
Seems like a sensor tech thing, and everyone takes it as gospel.

could not agree more
DXO's ratings are held in too high regard

this is particularly true for measures like tonality and DR which are much more difficult to measure
--
--
pbase & dpreview supporter
DPR forum member since 5/2001
http://www.pbase.com/artichoke

 Artichoke's gear list:Artichoke's gear list
Agfa ePhoto 1680 Fujifilm FinePix S3 Pro Leica M8 Leica M9 Nikon D3X +13 more
bloomoose
bloomoose Veteran Member • Posts: 3,224
Re: I think Nikon picked up some Fujifilm magic

And into the D700/D3

In October I did a side-by-side comparison of D3 and S5 and posted images on a German forum as a quiz. Only in one case could one user identify which was which.

Mind you, I processed through ACR with the beta profiles, but came close to jpgs. The scenes were contrasty, but not too much.

But I´ll be glad to see some good samples of good dynamic range in D3X pictures.

Artichoke wrote:

and put it into the D3X

-- hide signature --

Greetings from Germany,
Pam

Check out my images @
http://www.fotografie.fr/n3-galleryPMeier-1.htm

 bloomoose's gear list:bloomoose's gear list
Canon G9 X Sony RX100 Nikon D750 Sony a7R II Nikon AF Nikkor 105mm f/2D DC +7 more
sidelight Senior Member • Posts: 1,013
Shadow Performance: S/N and Tonal Range at 10% Gray Level

Iliah Borg wrote:

DxO do not measure and present the visual impact of noise. Same noise
numbers can correspond to fine grain agreeable and ugly blotchy
noise. Noise numbers are different for files white-balanced for
different colour temperatures, they do not test those however. Noise
should be measured separately for luma and chroma, that is also
skipped.

It matters how many gradations of grey are recorded in the last stop.
Some of the tested cameras make the last stop barely separable from
the previous while others allow for the practical use of the last
stop. For example, if you need to open shadows on A900 it is better
to sacrifice some DR but shoot as ISO 400 to ISO 640. DxO tests won't
show you this because they do not have the criteria for the
photographic usefulness of the last stop.

Colour measurements they do are a poor man's substitution of sensor
spectral characteristics, and are unusable as such, or as a tool to
compare sensors. The more meaningful numbers are numbers of metameric
faults by band, say, how many metamers are in the hues that represent
skin, foliage, sky.

Mathematics and physics are to be applied to the subject of
photography, results should be meaningful and reported in the
language of photography; and here DxO tests are lacking.

I was interested in the point made above regarding number of gray levels in the shadows, so I took the following data from DxOMark and tabulated it below for the D3X, A900, and S5Pro. I took the S/N from the Full SNR plots at 10% gray and the Tonal Range values are from the normalized print plots. I ranked the sensors by S/N values at lowest ISO, and then you can also see the number of gray levels the S/N range is divided into for the shadow region. I also included the values for the A900 at ISO400 since the post above mentioned better performance in the shadows at 400. I didn't see this but may have misunderstood. Below the table is DxOMark's summary of noise variance for shadows, midtones, and highlight regions.

I conclude from this data that the S5Pro has slightly higher S/N than the D3X but has significantly fewer gray levels available and therefore less smooth tonal transitions in the shadows than either the D3X or A900. The D3X has slightly higher S/N than the A900 and somewhat more gray levels available in the shadows, so should have ever so slightly deeper black before hitting the noise floor and a little smoother tonal transitions. I have no idea whether these differences are noticeable in prints. I suppose that might depend on the paper and the printer. I didn't include any color data.

DxOMark S/N @ Low Light for D3X. A900, and S5Pro

At 10% Gray Level and Lowest ISO

...............................Tonal........Gray
...............S/N...........Range.......Levels.......Levels/dB
..............(dB)...........(bits)*

S5Pro.......40.............8.29..........313............7.83

D3X.........37..............9.48..........714............19.3

A900.......34..............8.98..........505............14.9
ISO100

A900.......29..............8.28..........311............10.7
ISO400

Tonal Range is from the normalized print plot.

From DxOMark, Technologies, Noise Characterization, Summary:

Sensor noise variance is signal-dependent and comes from three main independent sources:

-Dark (electronic) noise, which is a constant, depending on the quality of the electronics, and which influences the quality of images in shadows.

-Photonic noise, which is a quantum phenomenon whose variance is proportional to the signal.

-Pixel response non-uniformity, which is due to electronic inhomogeneity of the sensor, and whose variance is proportional to the square of the signal.

Expressed as an equation, the variance corresponding to an output signal can be approximated by

Sigma^2 (x) = (y^2 * sigma^2 dark) + (y*x) + (k^2*x^2)

As a consequence, the SNR also has three regimes:

Shadows (Sigma^2 (x) ~ y^2 * sigma^2 dark): the SNR increases 6dB for every EV and loses 6dB for each doubling of the ISO setting.

Midtones (Sigma^2 (x) ~ y*x): the SNR increases 3dB for every EV and decreases by 3dB for each doubling of the ISO setting.

Highlights (Sigma^2 (x) ~ k^2*x^2): the SNR is constant and does not depend on the ISO.
--
David

Keyboard shortcuts:
FForum MMy threads