Ad
Ad
Ad
Pages: « 1 2 3 [4]   Bottom of Page
Print
Author Topic: Did the Nikon D800 change the world?  (Read 12518 times)
BJL
Sr. Member
****
Offline Offline

Posts: 5124


« Reply #60 on: May 13, 2012, 01:28:52 PM »
ReplyReply

Are you quite sure? Are you sure you haven't got that arse about face?
I agree that it is also quite common these days for technological innovations to arrive from the bottom up; that is why I said "can effect", not "does effect". For example, decisions by Canon and Nikon about what technology to offer in 35mm format gear interacts with what is offered in APS-C formats from the same makers.

Oh but the weather is getting better, and it is past time to go outside!
Logged
KLaban
Sr. Member
****
Offline Offline

Posts: 1671



WWW
« Reply #61 on: May 13, 2012, 01:41:57 PM »
ReplyReply

Oh but the weather is getting better, and it is past time to go outside!

Good shooting to you  Wink
Logged

Ray
Sr. Member
****
Online Online

Posts: 8878


« Reply #62 on: May 14, 2012, 08:01:53 AM »
ReplyReply

I just did an exercise comparing the DXOMark D800 results with a few MFDBs of similar pixel count. I think the closest in pixel count would be MFDBs such as the P645D, P40+ and P45+.

Those are all around 40mp, but if cropped to the 3:2 aspect ratio of the D800, the pixel count should be almost identical and therefore the resolution from all sensors compared should be the same, excluding any significant differences in lens quality, especially if we use the D800E.

I have to admit I was a bit surprised. I expected at least one or more of these DBs to have some qualitative advantage over the D800 in at least one or more parmeter, such as SNR at 18% for example. In fact, on all the graphs the D800 was either the equal, marginally better, or in the case of DR, significantly better.

There was just one instance of the P40+ having an insignificant 1dB better SNR at its less sensitive base ISO of 46. Needles to say, the D800 DR is way better, ranging between 1.3 and 2.5 stops better than all 3 DBs, depending on ISO.

The P40+ is a 2009 camera and the 645D, 2010. That we can get this level of performance from a $3,000 DSLR just 2 or 3 years after these very expensive MFDBs were announced, is very remarkable.
Logged
douglasf13
Sr. Member
****
Offline Offline

Posts: 547


« Reply #63 on: May 14, 2012, 01:19:24 PM »
ReplyReply

I just did an exercise comparing the DXOMark D800 results with a few MFDBs of similar pixel count. I think the closest in pixel count would be MFDBs such as the P645D, P40+ and P45+.

Those are all around 40mp, but if cropped to the 3:2 aspect ratio of the D800, the pixel count should be almost identical and therefore the resolution from all sensors compared should be the same, excluding any significant differences in lens quality, especially if we use the D800E.

I have to admit I was a bit surprised. I expected at least one or more of these DBs to have some qualitative advantage over the D800 in at least one or more parmeter, such as SNR at 18% for example. In fact, on all the graphs the D800 was either the equal, marginally better, or in the case of DR, significantly better.

There was just one instance of the P40+ having an insignificant 1dB better SNR at its less sensitive base ISO of 46. Needles to say, the D800 DR is way better, ranging between 1.3 and 2.5 stops better than all 3 DBs, depending on ISO.

The P40+ is a 2009 camera and the 645D, 2010. That we can get this level of performance from a $3,000 DSLR just 2 or 3 years after these very expensive MFDBs were announced, is very remarkable.

  In those terms, even some of the better aps-c cameras out there are remarkable, at this point.  In DxO's tests, the K-5 and NEX-7's DR tests higher than the 645D, and it isn't all that far off in most of the other categories.
Logged
BJL
Sr. Member
****
Offline Offline

Posts: 5124


« Reply #64 on: May 14, 2012, 04:08:35 PM »
ReplyReply

I have to admit I was a bit surprised. I expected at least one or more of these DBs to have some qualitative advantage over the D800 in at least one or more parmeter, such as SNR at 18% for example. In fact, on all the graphs the D800 was either the equal, marginally better, or in the case of DR, significantly better.
Ray,
   you should be surprised, because the DXO numbers seem to violate some basic physics in the realm near base ISO sensitivity, where noise at 18% is dominated by photon shot noise. The explanation is something that I mentioned in another recent post, and tried to explain to you a while ago: this is one case where the misguided DXO calibration of sensitivity is completely unjustified, and effectively pushes the SNR 18% curves of the DMF back about two stop left of where they should be.

What is happening there is that some DMF backs with 16-bit ADC output use that extra latitude to allow more of a safety margin in the highlights by positioning the mid-tones at an ADC output level that is further down from the (four times higher) maximum numerical level, which can be done while still keeping all the numerical level high enough that discretization noise does not harm to signal because it is still well below the noise of the analog signal: indeed, the absolute numerical levels need be no lower. DXO then declares that the "true" ISO speed is based on the placement of mid-tones in the raw file relative to maximum raw level, which is neither specified by any of the ISO12232 definitions of sensitivity nor relevant to comparisons of noise levels with equal lighting conditions and equal exposure level.

As a simplified illustration of the problem imagine a DSLR and a DMF tested with equal exposure index setting, equal f-stop and equal shutter speed with equal lighting on the test subject, and both producing an output level of 2000 for the 18% gray card mid-tones ... the catch being that for the DSLR with 14-bit output, this is level 2000 out of 16,384 whereas for the DMF back with 16-bit output, it is level 2000 out of 65,536, so that the placement relative to maximum level is 1/4 as much, so two stops lower. As a result the "DXO ISO speed" of the DMF back is two stops lower than for the DLSR: that is absurd! (And no, it cannot be explained by the need to correct for the discrepancy between t-stops and f-stops on the prime lenses that DXO uses for testing!)

But the good news is that, as you said a while ago, the ISO sensitivity values given by the cameras are also available with mouse-overs, so you can read them, adjust the curves horizontally and get a more accurate comparison. The offset is about 1 1/2 to 2 stops. When I read the graphs that way, the DMF backs do have roughly the expected advantage in 18% noise level when rescaled for pixel count in the "print" values.
« Last Edit: May 14, 2012, 05:19:55 PM by BJL » Logged
Doug Peterson
Sr. Member
****
Online Online

Posts: 2767


WWW
« Reply #65 on: May 14, 2012, 04:46:33 PM »
ReplyReply

I just did an exercise comparing the DXOMark D800 results with a few MFDBs of similar pixel count. I think the closest in pixel count would be MFDBs such as the P645D, P40+ and P45+.

Those are all around 40mp, but if cropped to the 3:2 aspect ratio of the D800, the pixel count should be almost identical and therefore the resolution from all sensors compared should be the same, excluding any significant differences in lens quality, especially if we use the D800E.

I have to admit I was a bit surprised. I expected at least one or more of these DBs to have some qualitative advantage over the D800 in at least one or more parmeter, such as SNR at 18% for example. In fact, on all the graphs the D800 was either the equal, marginally better, or in the case of DR, significantly better.

There was just one instance of the P40+ having an insignificant 1dB better SNR at its less sensitive base ISO of 46. Needles to say, the D800 DR is way better, ranging between 1.3 and 2.5 stops better than all 3 DBs, depending on ISO.

The P40+ is a 2009 camera and the 645D, 2010. That we can get this level of performance from a $3,000 DSLR just 2 or 3 years after these very expensive MFDBs were announced, is very remarkable.

I misread this post, initially I read:
"I just did an exercise comparing D800 results with a few MFDBs of similar pixel count."

I was understandably glued to my seat and, frankly, shocked at your statements. But having followed your posts for a long time I take them seriously.

Then I caught my misread that you didn't actually compare real-world photographic results, you compared dXo graphs. This is of far less interest to me.
Logged

DOUG PETERSON (dep@digitaltransitions.com), Digital Transitions
Dealer for Phase One, Mamiya Leaf, Arca-Swiss, Cambo, Profoto
Office: 877.367.8537
Cell: 740.707.2183
Phase One IQ250 FAQ
Ray
Sr. Member
****
Online Online

Posts: 8878


« Reply #66 on: May 14, 2012, 10:30:47 PM »
ReplyReply

  In those terms, even some of the better aps-c cameras out there are remarkable, at this point.  In DxO's tests, the K-5 and NEX-7's DR tests higher than the 645D, and it isn't all that far off in most of the other categories.

True! But in those cases the MFDB sometimes has an SNR at 18% advantage, which translates to smoother midtones and smoother skin tones. Also the higher resolution of 40MB and the greater sense of 3-dimensionality, or 'je ne sai quoi', as a result of the lack of an AA filter, can make a noticeable difference, when compared with 16mp sensors.

What surprises me in these comparisons I made is that those three 40mp DBs would appear to have zilch performance advantage in any respect whatsoever, not even in resolution, color sensitivity nor tonal range, and presumably when compared with the D800E, not even with regard to that mystical 3-dimensionality. They've become dinosaurs.
Logged
Ray
Sr. Member
****
Online Online

Posts: 8878


« Reply #67 on: May 14, 2012, 11:08:45 PM »
ReplyReply

I misread this post, initially I read:
"I just did an exercise comparing D800 results with a few MFDBs of similar pixel count."

I was understandably glued to my seat and, frankly, shocked at your statements. But having followed your posts for a long time I take them seriously.

Then I caught my misread that you didn't actually compare real-world photographic results, you compared dXo graphs. This is of far less interest to me.

 Grin  Grin  Grin. I would also be glued to my seat if you were to show me real-world comparisons between, say, the P45+ and the D800E. Unfortunately, my experience in the past has been, on this forum, when such comparisons are made between MFDB and 35mm, by MFDB afficionados, I usually sense some sort of strong bias going on. The DoFs may be different, or the focal lengths used are not quite equivalent, or the focussing is slightly different in both shots, or the exposures are not equal in terms of ETTR, and so on. Sometimes even, the scenes compared are different.

Whenever I compare the performance of my own cameras, which vary in sensor size and image quality to perhaps an even greater extent than a modern FF 35mm compared with some DBs, I'm always extremely careful about possible bias. I bracket exposure in 1/3rd stop intervals from obvious overexposure to obvious underexposure, to ensure that I end up comparing equally exposed shots. I use either LiveView for critical focussing, or select a target with a pattern that will produce aliasing in the viewfinder, indicating 'spot on' focussing.

Whenever I perform such comparisons, I find my results almost exactly match DXOMark's. Not actually exactly of course. One always has to allow for a small margin of error, which may be due to the methodology or may be due to manufacturing QC variations in the cameras tested. Any variation of less than 1/3rd of a stop in DR or SNR would fall into that category.

I look forward to your comparisons, Doug.  Grin
Logged
Ray
Sr. Member
****
Online Online

Posts: 8878


« Reply #68 on: May 14, 2012, 11:54:33 PM »
ReplyReply

Ray,
   you should be surprised, because the DXO numbers seem to violate some basic physics in the realm near base ISO sensitivity, where noise at 18% is dominated by photon shot noise. The explanation is something that I mentioned in another recent post, and tried to explain to you a while ago: this is one case where the misguided DXO calibration of sensitivity is completely unjustified, and effectively pushes the SNR 18% curves of the DMF back about two stop left of where they should be.

What is happening there is that some DMF backs with 16-bit ADC output use that extra latitude to allow more of a safety margin in the highlights by positioning the mid-tones at an ADC output level that is further down from the (four times higher) maximum numerical level, which can be done while still keeping all the numerical level high enough that discretization noise does not harm to signal because it is still well below the noise of the analog signal: indeed, the absolute numerical levels need be no lower. DXO then declares that the "true" ISO speed is based on the placement of mid-tones in the raw file relative to maximum raw level, which is neither specified by any of the ISO12232 definitions of sensitivity nor relevant to comparisons of noise levels with equal lighting conditions and equal exposure level.

As a simplified illustration of the problem imagine a DSLR and a DMF tested with equal exposure index setting, equal f-stop and equal shutter speed with equal lighting on the test subject, and both producing an output level of 2000 for the 18% gray card mid-tones ... the catch being that for the DSLR with 14-bit output, this is level 2000 out of 16,384 whereas for the DMF back with 16-bit output, it is level 2000 out of 65,536, so that the placement relative to maximum level is 1/4 as much, so two stops lower. As a result the "DXO ISO speed" of the DMF back is two stops lower than for the DLSR: that is absurd! (And no, it cannot be explained by the need to correct for the discrepancy between t-stops and f-stops on the prime lenses that DXO uses for testing!)

But the good news is that, as you said a while ago, the ISO sensitivity values given by the cameras are also available with mouse-overs, so you can read them, adjust the curves horizontally and get a more accurate comparison. The offset is about 1 1/2 to 2 stops. When I read the graphs that way, the DMF backs do have roughly the expected advantage in 18% noise level when rescaled for pixel count in the "print" values.


Can't make head nor tail of your post, BJL. You seem to be stuck in some antiquated methodology relating to the film era.

I mentioned in some other thread my experiments in shot noise which would be relevant to your claim that noise at 18% grey is dominated by shot noise.

It so happens that the Canon 50D has an unusual characteristic. The camera appears to produce the same image quality at ISO 100 as it does at ISO 200. DXO don't even mention test results at ISO 100 for this camera, yet ISO 100 does require double the exposure of ISO 200.

What's the effect of that extra light, one might wonder. My tests indicate that it doesn't improve DR by any significant degree, nor does it improve noise in the midtones by any significant degree.

To demonstrate my point I attach 200% crops of highlights and midtones, plus a 600% crop of midtones and lower midtones. Judge for yourself. Theory is fine. Practice nails the issue.

« Last Edit: May 14, 2012, 11:58:43 PM by Ray » Logged
Pages: « 1 2 3 [4]   Top of Page
Print
Jump to:  

Ad
Ad
Ad