Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: NikosR on January 23, 2009, 09:43:10 am
-
Interesting essay:
http://www.diglloyd.com/diglloyd/free/Push...acks/index.html (http://www.diglloyd.com/diglloyd/free/PushingTheBlacks/index.html)
-
Interesting. My one question would be, did he test the actual ISO for ETTR such that we'd have apples to apples comparisons for noise? IOW, if the "correct" exposure assumed for the Canon was actually down, more noise. Could it be that in addition to the strong possibility the Nikon is producing a cleaner capture that its suggested ISO and exposure is more accurate and affecting shadow noise?
-
Interesting. My one question would be, did he test the actual ISO for ETTR such that we'd have apples to apples comparisons for noise? IOW, if the "correct" exposure assumed for the Canon was actually down, more noise. Could it be that in addition to the strong possibility the Nikon is producing a cleaner capture that its suggested ISO and exposure is more accurate and affecting shadow noise?
That's a good question. He says he will provide full test specs in the pay-read part of his site
Another, less good, question would be whether performing the shadow lift in the raw converter would have lessened the gap between the 2 cameras.
-
Against a bunch of cameras (but I don't have access to a 5D mkII or anything else high-end and released in the last year), I'm seeing this by eye on tricky landscape subjects - the D3x is just amazingly clean at low ISO, and pulls detail out of the shadows extremely well. It really DOES look like a H3DII/31 (or other low-end MF back) file. Note Phase One's "who's afraid of the D3x" (my title, not theirs) promotion where they throw in up to FIVE lenses with purchase of any of their lower-end systems. They wouldn't be throwing in five lenses (and notably, not offering any promotion on the P65+, which has image quality in a different league) if they weren't scared...
-Dan
-
I paid for Lloyd Chamber's DAP, and I can report that in his "Pushing the Blacks" review he carefully shot all the comparison cameras at the same exposure value (shuttle speed and aperture), and with the same lens and lighting.
Actually anybody has little scientific training should have learned the very same comparing these DxO measurement curves (go to the far right and click the "Full SNR" tab, and put the both on logarithmic scales):
http://www.dxomark.com/index.php/eng/Image...abase/Nikon/D3X (http://www.dxomark.com/index.php/eng/Image-Quality-Database/Nikon/D3X)
http://www.dxomark.com/index.php/eng/Image.../EOS-5D-Mark-II (http://www.dxomark.com/index.php/eng/Image-Quality-Database/Canon/EOS-5D-Mark-II)
What the curves in these two charts tell us? The D3x is the first DSLR that produces nearly the linear response of an ideal system. It extends the 3dB/EV slop from the mid-tones well into the shadows region, unlike all the others that bend over to 6dB/EV, because of "dirty" readout pathes.
If you don't want to read the numbers out of the curves, here are my readings rounded up to the nearest half dBs (underscores added here to help -- the server remove the white spaces to make this kinds of text table impossible even with a fixed width font):
DxO Measured SNR (Signal to Noise Ratio) in dBs:
Gray Camera +--------------- ISO ----------------+
Level Model | _100 _200 _400 _800 1600 3200 6400 |
----- ----- | ---- ---- ---- ---- ---- ---- ---- |
0.1% _D3x__ | 14.5 11.0 _6.5 _2.0 ---- ---- ---- |
0.1% _5DII_ | _7.0 _6.5 _5.5 _3.5 ---- ---- ---- |
0.1% _D700_ | _N/A 10.5 _9.0 _6.5 _1.5 ---- ---- |
1.0% _D3x__ | 26.5 23.0 20.0 16.5 12.5 _9.0 _4.0 |
1.0% _5DII_ | 24.5 23.0 21.0 18.0 15.0 11.0 _7.0 |
1.0% _D700_ | _N/A 26.5 23.5 21.0 17.0 12.5 _7.5 |
Since I am busy at this moment, let me simply say at twice the pixel area the D700 is expected to have 3dB SNR advantage across the full board, which start to be the case at 1% gray level but not below, and quote what I said this morning in another forum:
"The new D3x price is not a bubble. And the dealers are not going to file bankruptcy because their D3x inventory are just sitting on the shelf and gathering dust. Actually, the major US online houses, B&H, Adorama, and J&R, are all out of them at this moment.
Suddenly everything came together, and everything makes sense now. The DxO measurements, and Lloyd's professionally done tests published last night. When shadown details are pushed up by few stops ("Pushing the Blacks" is Lloyd's title), the amazingly clean D3x low ISO shadow noise beats at least that of one medium format (MF) digital back, the $15,000 Mamiya DL28. The D3s noise at ISO 400 is superior to the 5D Mark II at ISO 100, or 50, and if you pay for Lloyd's DAP to view the whole study (I did), you can make an argument that as a matter of fact the D3's noise at ISO 800 is still better than the 5DII at ISO 100, in deep shadows.
All these confirms the DxO numbers are real, the 13 stops dynamic range (almost 14 stops when scaled down). Even at 18% gray and above and high ISO values the other models (D3 and the Canons) start to catch up, it is in the darker shadows where the D3x really shine, shining alone, in a league of its own in the DSLR world, and become very close to the theoritical limits governed by the rules of physics (which means there will be not much room for possible improvements).
Enjoy the reading for now (and dig into the DAP if you can for more amazing details), while I am writing a long detailed scientific analysis (with a lot of math, unfortunately, to justify what I am saying) titled "When the Numbers Meet Eye" and try to get it published in a better and more rational online place.
http://diglloyd.com/diglloyd/free/PushingT...ndex.html" (http://diglloyd.com/diglloyd/free/PushingTheBlacks/index.html")
-
Hi,
I would say that Mr. Chambers observation on the very good DR of the D3x is consistent with the DxO-mark.
Best regards
Erik
Against a bunch of cameras (but I don't have access to a 5D mkII or anything else high-end and released in the last year), I'm seeing this by eye on tricky landscape subjects - the D3x is just amazingly clean at low ISO, and pulls detail out of the shadows extremely well. It really DOES look like a H3DII/31 (or other low-end MF back) file. Note Phase One's "who's afraid of the D3x" (my title, not theirs) promotion where they throw in up to FIVE lenses with purchase of any of their lower-end systems. They wouldn't be throwing in five lenses (and notably, not offering any promotion on the P65+, which has image quality in a different league) if they weren't scared...
-Dan
-
Hi,
Please put a link to your article also on this place. Although I agree that this forum is not always rational, it is still one of the best fora about photography IMHO.
I'm quite impressed with Nikon's achievement, especially as it's probably based on the same basic sensor design as used in the Sony Alpha. As an observation the Sony Alpha 900 seems also be pretty good on DR in the low ISO range, even if it is not even close to Nikon D3x.
Best regards
Erik
Enjoy the reading for now (and dig into the DAP if you can for more amazing details), while I am writing a long detailed scientific analysis (with a lot of math, unfortunately, to justify what I am saying) titled "When the Numbers Meets Eye" and try to get it published in a better and more rational online place.
http://diglloyd.com/diglloyd/free/PushingT...ndex.html" (http://diglloyd.com/diglloyd/free/PushingTheBlacks/index.html")
-
I am not very sure that the P45+ back, still retails for $30000 without the extended service package, is considered "lower-end". One thing I read is that P45+ (Kodak sensor) easily handles long (one hour and longer) exposures in the low light, while the P65+ (Delsa "full full frame" sensor) can only do a few minutes before things are buried in the noise.
If you add up the B&H price (and the rumored prices for the new lenses such as the new 45mm and the new 150mm) for the five more expensive new lens choices (get your 80mm else where) from the pool, plus the free new Phase One camera body, the total value is at or slightly exceeds $20000. This leaves under $10000 for the back itself (while the truth is you still need to have $30000 in your deep pocket).
Against a bunch of cameras (but I don't have access to a 5D mkII or anything else high-end and released in the last year), I'm seeing this by eye on tricky landscape subjects - the D3x is just amazingly clean at low ISO, and pulls detail out of the shadows extremely well. It really DOES look like a H3DII/31 (or other low-end MF back) file. Note Phase One's "who's afraid of the D3x" (my title, not theirs) promotion where they throw in up to FIVE lenses with purchase of any of their lower-end systems. They wouldn't be throwing in five lenses (and notably, not offering any promotion on the P65+, which has image quality in a different league) if they weren't scared...
-Dan
-
LEPING,
As far as I can tell, since I'm also a DAP subscriber, Lloyd has not yet put up the noise comparison you're referring to on DAP. He promises it will be there soon. How the heck did you read it?
Secondly, if he really used the same exposure / speed for both cameras compared in the linked essay, that would not guarrantee optimal ETTR as digitaldog was quick to point out.
I have no problem believing that the D3x is the best dSLR at this point in time, nor that it competes with the low end MFDBs. I'm just wondering about how the particular test was conducted and how you know about it since it hasn't been published yet.
-
I have no problem believing that the D3x is the best dSLR at this point in time, nor that it competes with the low end MFDBs. I'm just wondering about how the particular test was conducted and how you know about it since it hasn't been published yet.
Neither do I. I'm just wondering if we know that the ideal exposure was indeed used for the captures. In my tests, ETTR can make that significant difference in noise.
-
Hi Nikos,
I asked him with the same kind of questions, and he replied in his emails that the exposures were exactly the same, and the amount of "push" were exactly the same too. So probably he fixed the exposures rather than searching the "optimal ETTR" for the individual cameras.
In the DxO tests we can see they have found the "true ISOs" were very close for the D3x and the 5DII, but of course the full well capacities may be significantly different so that for the same 18% gray reading the highlight headrooms can differ. But this requires much more complicated studies, I think. And shooting the same speed and aperture things are matched in the practical sense to me -- would you say XXX's noise behavior is better when we add 2/3 stop of exposures, in comparison to another model at the meter reading?
From reading the detailed DxO test technology descriptions I think it is the same way they derived their numbers. If you look at my table, at 0.1% gray D3x produced an SNR of 6.5 at ISO 400 and 5DII produced 7 at ISO 100. This is quite in line with what Lloyd reported. At ISO 400 ideally the D700 should have achieved an SNR of 9.5 (6.5+3), but it is only 9.0 indicating the read noise is significant. At ISO 200 the D700 is even worse (-0.5dB), representative to almost all the sensors prior to D3x. And, since the D3x's true base ISO is near 80, one can always choose to shoot at ISO 80 without sacrifice the highlights.
He also promised the full test result contents in the DAP in a couple of more days. If you have other questions you can ask him directly. We are both in the SF Bay area, and I have met him twice over the years in public meetings.
Thanks,
Leping
LEPING,
As far as I can tell, since I'm also a DAP subscriber, Lloyd has not yet put up the noise comparison you're referring to on DAP. He promises it will be there soon. How the heck did you read it?
Secondly, if he really used the same exposure / speed for both cameras compared in the linked essay, that would not guarrantee optimal ETTR as digitaldog was quick to point out.
I have no problem believing that the D3x is the best dSLR at this point in time, nor that it competes with the low end MFDBs. I'm just wondering about how the particular test was conducted and how you know about it since it hasn't been published yet.
-
One should also bear in mind that the DXOmark tests of ISO sensitivity place the D3X and 5D2 as having almost identical ISO sensitivities at the cameras' rating of ISO 100. There should be no ISO sensitivity trap to fall into here.
It seems clear that both Sony and Nikon in their 24mp models have concentrated on optimising performance at base ISO rather than at ultra-high ISO. We all must have images of contrasty scenes that have unacceptably noisy shadows. I certainly have, from my 5D. If the scene is really contrasty and I try to lift the shadows, I invariably see traces of banding as well as general image degradation in the deepest shadows. An extra 1&1/2 to 2 stops of DR would fix that.
However, it's interesting that the increased DR of the D3X, compared with the 5D (never mind the 5D2), gradually gets smaller as ISO increases. By ISO 800, the D3X is no better than the 5D, regarding DR. If one downsamples the both images to an 8x12' print at 300 ppi (thus sacrificing the D3X resolution advantage over the 5D), then the D3X DR is about 1/2 a stop better than the 5D at ISO 800, but no better at ISO 1600. Interesting, eh!
-
Wow, Lloyd just modified his post, and added explanations to his exposure choice, including considerations to ETTR.
http://diglloyd.com/diglloyd/free/PushingT...acks/index.html (http://diglloyd.com/diglloyd/free/PushingTheBlacks/index.html)
-
Wow, Lloyd just modified his post, and added explanations to his exposure choice, including considerations to ETTR.
It appears to be a fair and rigorous test.
It should be interesting to see how Michael felt about his loan D3x in Antartica.
Cheers,
Bernard
-
From reading the detailed DxO test technology descriptions I think it is the same way they derived their numbers. If you look at my table, at 0.1% gray D3x produced an SNR of 6.5 at ISO 400 and 5DII produced 7 at ISO 100. This is quite in line with what Lloyd reported. At ISO 400 ideally the D700 should have achieved an SNR of 9.5 (6.5+3), but it is only 9.0 indicating the read noise is significant. At ISO 200 the D700 is even worse (-0.5dB), representative to almost all the sensors prior to D3x. And, since the D3x's true base ISO is near 80, one can always choose to shoot at ISO 80 without sacrifice the highlights.
Leping thanks for the clarification. I believe DxO are using their derived 'true' ISOs in their charts and not their shooting ISOs. Anyhow this might be just a question of purely acedemic interest.
-
The full review is in Lloyd's DAP now:
"I’ve just published my DAP report on noise behavior of the Nikon D3x, Nikon D3, Canon 5D Mark III and Canon 1Ds Mark III.
See yesterday’s blog entry for an introduction; the full report covers four cameras in greater breadth and depth."
http://diglloyd.com/diglloyd/blog.html (http://diglloyd.com/diglloyd/blog.html)
Thanks,
Leping
-
This is mostly unrelated. And I do understand not everyone uses a Mac. But for those who do, I just found Lloyd's (totally free) comprehensive guide for optimizing Mac performance, especially running Photoshop and other photographic applications (such as the Canon DPP).
http://macperformanceguide.com/index.html (http://macperformanceguide.com/index.html)
-
That was a pretty interesting piece.
If I am reading it right, the 5DMkII is hands down the best value in serious DSLR imaging (in fact, Lloyd directly says this). It completely eclipses its true competitor, the D700.
However, the test shows the D3x can get better detail in some lowlight cases.
I would hope so at 3x the price
-
FWIW, I'm a subscriber to Lloyd's site, and his A900 review was a bit of a joke in some ways. He got his A900 and lenses from a rental place, and then complained that none of his shots were in focus, but he didn't have the time to do AF adjustments, because he expects things to work right out of the box....from a rental company. Why even take the time to do the review? Plus, he used Sony's consumer/crappy supplied software as the RAW converter, which is one of the worst converters known to man. Granted that's Sony's fault, but many conclusions about the A900 shouldn't be made by reading his review. Clearly, he is a bit too entrenched in his current systems, which his latest title line confirms: "Noise performance of early 2009’s premier digital SLRs." Apparently, the A900 is not on that list.
My only point to this rant is this. Lloyd's tests, like many others, should be taken with a grain of salt.
-
ok ,s i read through this thread now and looked at the blog (i won't subscribe....) and looked at the dxo website.....i am very happy with canon, but a 13stop DR is not something to joke around with.....but i still don't have a clue if this is actually true or not....i am not sure i understand the arguments pro or con (and i sure as hell don't get all the technical babble....)...
why is it so hard to get this straight? we all know that most Dbacks have more DR then DSLRs, this is measurable and visable....so why not here? or is it just that some people do not WANT to see it?
the blog shows me nothing....i can get superclean shadows with my G10 from different converters, (just don't ask about the highlights....) and i generally don't judge cameras by online jpegs....i guess paying the subscription and downloading the raw files would tell me a bit more, but why are there people who did just that and still say it does not add up to an extra 2 stops?
i guess i will have to rent it for a day and do my own tests....
a camera with near DMF quality at 100 (at least in terms of DR) and almost 5DII quality at 800 sounds too good to be true....
i still cannot get over the fact that nikon did not build in a sensor cleaner....that is just nuts to me....the missing video might not be a big deal now, but the question is if canon can put it into a 2700$ body, why can't nikon fit it in for 8000$....just does not make sense....
imagine how well it would sell with a full frame HD mode with 13stops DR! this would blow anything available out of the water....at any price.....
-
It completely eclipses its true competitor, the D700.
Not quite, Jack. I think you might be in danger of becoming a Canon fanboy .
Lloyd claims his tests are pretty much in line with DXOmark's results. Below is DXO's DR page comparing these two cameras.
Panopeeper would probably claim that the graph at the top is more indicative of the true DR of these cameras, showing a 2/3rds stop DR advantage for the D700. However, when both images are downsampled for an 8x12" print at 300 ppi, the DR is about the same for both cameras, although the D700 still retains about 1/3rd of a stop advantage at base ISO.
However, the 5D2 certainly eclipses the D700 regarding resolution.
You should also not forget that some cameras have other features that contribute to image quality in addition to basic sensor performance. I'm particularly impressed with the D700's ability to autobracket up to 9 consecutive shots with a choice of exposure interval of 1/3 to 1 stop. This feature is tremendously useful for merging to HDR to increase DR. In my experience a lot of the criticism of the HDR process is due to people trying to merge an inadequate number of shots with an exposure interval that is far too great for the 2 or 3 shots taken.
Another tremendously useful feature of the D700 is the facility to set auto ISO in manual mode. You choose the aperture and the shutter speed. The camera chooses the ISO for correct exposure. It has always puzzled me why Canon have not bothered to provide such a facility considering they have provided cameras with low noise at high ISO for a number of years now.
A lot of photography is about capturing the moment. One often simply doesn't have time to mess around adjusting ISO in order to get an adequate shutter speed. Controling both the aperture and shutter speed can be critical for a good shot.
[attachment=11087:DXOmark_5d2_v_D700.jpg]
-
The camera chooses the ISO for correct exposure. It has always puzzled me why Canon have not bothered to provide such a facility considering they have provided cameras with low noise at high ISO for a number of years now
Perhaps Canon do not want to sell to suckers? Perhaps they understand, that every stop in ISO reduces the DR by half stop (in average)?
-
Thanks Ray. To me the most important is to compare these two "Full SNR" charts. The charts you posted were "2D" (SNR - ISO at 18% gray level), which are cross sections of these two "3D" (SNR - ISO - Gray Level) charts that tells you much much more, such as linearity.
Since they are produced by the DxO site as "Fusion Charts" there is no labels. The left is D3x and the right is 5DII.
Thanks,
Leping
-
Perhaps Canon do not want to sell to suckers? Perhaps they understand, that every stop in ISO reduces the DR by half stop (in average)?
Nonsense. Everything is a compromise. I don't see how offering an option (you don't HAVE to use Auto ISO) is anything but a good thing. Why don't you go over to sportsshooter.com and collectively call the people over there 'suckers'. See what their reaction will be...
-
[quote name='LEPING' date='Jan 23 2009, 08:55 PM' post='254029']
DxO Measured SNR (Signal to Noise Ratio) in dBs:
Gray Camera +--------------- ISO ----------------+
Level Model | _100 _200 _400 _800 1600 3200 6400 |
----- ----- | ---- ---- ---- ---- ---- ---- ---- |
0.1% _D3x__ | 14.5 11.0 _6.5 _2.0 ---- ---- ---- |
0.1% _5DII_ | _7.0 _6.5 _5.5 _3.5 ---- ---- ---- |
0.1% _D700_ | _N/A 10.5 _9.0 _6.5 _1.5 ---- ---- |
1.0% _D3x__ | 26.5 23.0 20.0 16.5 12.5 _9.0 _4.0 |
1.0% _5DII_ | 24.5 23.0 21.0 18.0 15.0 11.0 _7.0 |
1.0% _D700_ | _N/A 26.5 23.5 21.0 17.0 12.5 _7.5 |
0.1% At ISO100 More than 7dB of difference
Very, very strange
Thierry
-
Lloyd says this on the public site:
"The examples were “pushed” using the Photoshop CS4 Highlights/Shadows feature. The amount used is equivalent to 3-4 stops, which is not much more than the light falloff with many lenses shot at wide apertures (~3 stops combined optical and sensor vignetting). Combined with features like Active D-Lighting, Peripheral Lighting correction, etc, it is perfectly realistic to “push” the shadows 2-3 stops as a matter of routine, not to mention high-contrast scenes that require very dark tones in order to record the highlights." - Lloyd L. Chambers
Really? 3 stops for vignetting? I've got an "old" 5D, and I don't think even my 24-105 f/4L is that bad. Sure, it's got relatively large photosites, but...
I can't remember the last time I pushed shadows 3-4 stops, (or 2-3), let alone "as a matter of routine." In fact usually I nudge up the blacks combined with some fill light. I'm not saying it wouldn't be nice to have the option, but the frequency of need just isn't in my experience. Or am I falling into the "I never need that (because I don't have it) trap?
Dave Chew
-
Perhaps Canon do not want to sell to suckers? Perhaps they understand, that every stop in ISO reduces the DR by half stop (in average)?
Which would you prefer? A sharp image at the sacrifice of a stop of DR, or a blurry image with lovely clean shadows?
One should also consider, when a camera has exceptionally good dynamic range, as the D700 has, there's some to spare with most scenes.
If the DXO test results, as shown in graph form on their site, are accurate, then the D700 at the actual and tested ISO of 400, has the same DR as the 5D2 at the real and tested ISO of 73 (base ISO); and that's after downsampling to an 8"x12" print size. If we use your standard, the DR of the 5D2 at ISO 73 is equal to the DR of the D700 at ISO 651 which allows for more than 3 stops faster shutter speed at the same aperture whilst still maintaining the same DR as the blurry image from the 5D2 at ISO 73.
-
Lloyd just reported that D3x is also much cleaner (no hot pixels or streaking) at long exposures (30 seconds and over) WITHOUT dark frame subtraction ("Long Exposure NR") than the 5DII, which he says come with a "dirty" sensor.
http://diglloyd.com/diglloyd/2009-01-blog....124LongExposure (http://diglloyd.com/diglloyd/2009-01-blog.html#_20090124LongExposure)
-
Nonsense. Everything is a compromise. I don't see how offering an option (you don't HAVE to use Auto ISO) is anything but a good thing. Why don't you go over to sportsshooter.com and collectively call the people over there 'suckers'. See what their reaction will be...
I fully agree. We all know the downside to pushing ISO. There's no free lunch. As Ray said perfectly, I'd rather have a sharp grainy, lower DR image then no image at all. When I go into a dim, naturally lit scene and have no intention of using ugly on camera flash versus nice but dim natural light, I'd far prefer the later and will accept the quality loss. That's why I dropped the $$ on a 35mm 1.4. In just a few weeks of shooting, I'm very pleased at the 5DMII versus the 5D it replaces in terms of its extended ISO.
-
The DxO results as well as the "professionally done" tests of Lloyd have been too suspicious for me to let them pass unchecked. Unfortunately, I can not make measurement on their raw images, for I don't have them. However, I have used the raw images published by Imaging Resources and made objective, documented measurements based on the non-demosaiced raw data.
The result proves, that the claim "two stops higher DR than the 5D2" is *ridiculous*, like some other claims. In fact, the DR of the D3X is max. 0.5 EV greater than that of the 5D2 at ISO 100. AT ISO 200 they are virtually equal, and at ISO 400 the 5D2 is already better.
A side note: I find it amusing, that someone makes a "professional test", among others involving ISO 50, obviously not knowing, that neither of the cameras has ISO 50.
Independently of the unseriosity of those claims, the D3X seems to be a great camera, according to what I see in the images (I am sure the other aspects do not negate this). Having a 0.5 EV advantage over Canon's best sensor of the moment is something Nikon can be proud of.
Anyway,
I challenge anyone to post raw files proving those claims.
-
It's hard for anyone interested to argue with you when you don't provide any supporting evidence for your claims or explain your methodology.
BTW DxO purport to have measured the following max DR values for the two cameras at their ISO 100 setting (measured before resolution normalisation):
5DII: Claimed ISO 100, Measured ISO 73, DR 11.16 EV
D3X: Claimed ISO 100, Measured ISO 78, DR 12.84 EV
At ISO 400 setting:
5DII: Claimed ISO 400, Measured ISO 285 (!!), DR 10.92EV
D3X: Claimed ISO 400, Measured ISO 337, DR 11.25EV
The 5DII is measured as having equal DR to the D3x near measured ISO 564 ( 5DII claimed ISO 800!!) with DR about 10.66EV. From then on (higher than measured ISO 564) the 5DII exhibits slightly better DR.
They say they measure DR from well saturation to S/N=1 (a noise floor which btw is probably useless from a photographic point of view but correct from an engineering point of view)
Any evidence you provide to counter DxO claims should counter the above DxO findings.
-
Hi,
Just a comment. Lloyd Chambers's tests are based on converted images and the choice of tools he use of course matters. I don't have an issue with that approach.
Your testing is based on "raw"-data, so your results tell a story which is different. Your approach tells us what information is available in the raw file, that's a valid approach and IMHO a more interesting one.
Regarding the DxO results I'm puzzled. Mr. Leping had links to these curves from DxO which I find quite revealing even if I cannot explain the difference.
http://luminous-landscape.com/forum/index....st&id=11093 (http://luminous-landscape.com/forum/index.php?act=attach&type=post&id=11093) (Nikon D3x)
http://luminous-landscape.com/forum/index....st&id=11094 (http://luminous-landscape.com/forum/index.php?act=attach&type=post&id=11094) (Canon 5DII)
My observation in the DxO figures is that SNR drops rapidly at 0.1% gray scale whereas then Nikon curves have a slower and more natural drop.
Best regards
Erik
The DxO results as well as the "professionally done" tests of Lloyd have been too suspicious for me to let them pass unchecked. Unfortunately, I can not make measurement on their raw images, for I don't have them. However, I have used the raw images published by Imaging Resources and made objective, documented measurements based on the non-demosaiced raw data.
The result proves, that the claim "two stops higher DR than the 5D2" is *ridiculous*, like some other claims. In fact, the DR of the D3X is max. 0.5 EV greater than that of the 5D2 at ISO 100. AT ISO 200 they are virtually equal, and at ISO 400 the 5D2 is already better.
A side note: I find it amusing, that someone makes a "professional test", among others involving ISO 50, obviously not knowing, that neither of the cameras has ISO 50.
Independently of the unseriosity of those claims, the D3X seems to be a great camera, according to what I see in the images (I am sure the other aspects do not negate this). Having a 0.5 EV advantage over Canon's best sensor of the moment is something Nikon can be proud of.
Anyway,
I challenge anyone to post raw files proving those claims.
-
Lloyd says this on the public site:
"The examples were “pushed” using the Photoshop CS4 Highlights/Shadows feature. The amount used is equivalent to 3-4 stops, which is not much more than the light falloff with many lenses shot at wide apertures (~3 stops combined optical and sensor vignetting). Combined with features like Active D-Lighting, Peripheral Lighting correction, etc, it is perfectly realistic to “push” the shadows 2-3 stops as a matter of routine, not to mention high-contrast scenes that require very dark tones in order to record the highlights." - Lloyd L. Chambers
Now if I want to push shots I use the best RAW converter for the files and optimise there, not open in PS and use Shadows/Highlights. Or is he doing that as well? Even so.....!
The only way to realistically test cameras in my view is to use the best practice techniques for each camera, such as the best/most likely to be used RAW developer for those files and then look at the results - that has far more meaning in the real world than using each one identically, but unrealistically
I saw some comparison shots in a UK Camera mag the other day and the 5DII files looked way inferior to the D3x files but they also looked underexposed in comparison. So not surprising really.
-
This is mostly unrelated. And I do understand not everyone uses a Mac. But for those who do, I just found Lloyd's (totally free) comprehensive guide for optimizing Mac performance, especially running Photoshop and other photographic applications (such as the Canon DPP).
http://macperformanceguide.com/index.html (http://macperformanceguide.com/index.html)
He does talk some ignorant tosh though. For instance the very first thing he posts is this.
"Ergonomics: the Mac Pro is about as quiet as it gets, and all its memory, hard drives, etc are easily accessible without tools. It is esthetically pleasing also, fitting into even “hip” offices, especially in Redmond.
Massive, fast storage: The Mac Pro accommodates four internal drives for 6TB of internal storage (4 X 1.5TB) , and with a special bracket as much as 9TB, all using the built-in SATA connectors. "
Adding two extra hard drives to take my MacPro up to six from the normal max of 4 bays internally, involved removing the DVD drive and some other fiddly bits to then spend quite a while threading SATA cables through awkward gaps to plug into the obscured extra two SATA ports, which most people are unaware of. I had to also hack the end connectors of the SATA cable to allow it to fit in the sockets. I now also have no room for DVD writer, unless I use an external one.
My PC tower has 9 drives in it and 2 optical drives and isn't any bigger than the MacPro. Which is indeed a nice quiet box, except for the dreadfully loud DVD writer and was another reason I removed it. But it's decidedly limited and fiddly if you like lots of HDs. I have another 5 External drives [4 eSATAs] with loads of bulky cables and power adaptors making the neat MacPro look a lot less tidy and I still need more HDs attached . It would be a lot better if the MacPro was a bit more professional and had some options [a dirty word at Apple] like a slightly taller case if required with 2 rows of 4 HD caddies.
Going a little OT here sorry. But Lloyd doesn't make me think too highly of his 'scienticfic' testing when he gets such basic facts wrong and doesn't even know if you can run Win64 on a Mac [you can]. He just reads like a fanboi and would treat his camera tests with some scepticism as a result.
Shame as some of his other advice on the subject is actually very good.
-
My observation in the DxO figures is that SNR drops rapidly at 0.1% gray scale whereas then Nikon curves have a slower and more natural drop.
The knee in the SNR curves is the crossover between the noise being dominated by electronic read noises at the low end to being dominated by photon shot noise at the high end. The relative constancy of the slope of the curves across the dynamic range is an indication that the read noise is quite low at low ISO in the D3x. This may be a consequence of the column-parallel readout architecture of the Sony chip; the D300 has a similar property, though not quite so pronounced.
-
Exactly. Please read the DxO documentation for the three noise characterization zones, better not only this summary but all the tabs in the "Noise Characterization" topic.
If there were no read noise, both the shadow and the mid-tone zone curves would go by the same slope, 6dB per EV, so that there is no "knee". Only D3x came close to the ideal behavior, while all the other sensors fall short more or less, if you check their "Full SNR" curves out one by one.
These curves also enable to derive DR measures at any SNR definitions above 1:1 (0dB).
http://www.dxomark.com/index.php/eng/Techn...ization/Summary (http://www.dxomark.com/index.php/eng/Technologies/Noise-characterization/Summary)
The knee in the SNR curves is the crossover between the noise being dominated by electronic read noises at the low end to being dominated by photon shot noise at the high end. The relative constancy of the slope of the curves across the dynamic range is an indication that the read noise is quite low at low ISO in the D3x. This may be a consequence of the column-parallel readout architecture of the Sony chip; the D300 has a similar property, though not quite so pronounced.
-
What you have said is that basically Lloyd was right and there are solid recommendations in his Mac performance site, but you need more hard drives beyond the Mac Pro's capacity of 7, and you know things like the tricks to run Win64 on a Mac Pro he didn't.
Congratulations for being such a elite computer user requiring at least two 4-HD raws inside your desktop machine, but isn't that true for the most of the rest (needing less than 7 drives, etc.) there are still things to learn from the site, such as how to separate data and applications, and how to set up a Photoshop specific scratch volume or a RAM disk to speed things up? And is it true for them the Mac Pro is still quiet and tidy? I do competitive photography, but I do not happen to know that much about computer. Is this makes me disqualified to express my findings here in a photography forum, or my own blogs?
Seagate insisted their firmware fixes for the problematic 7200.11 drives can only be applied on a PC. I searched online, found the procedures, and had no problem to boot my Mac Pro from the ISO CD-ROM volume to flash all my 1.5T drives successfully (since now a days Macs are indeed just PCs and they can boot into DOS). Does Seagate's ignorance makes them disqualified to make hard drives?
Lloyd Chambers also offers free drive and memory stress test software (Disk Tester, etc.), that have been used in many professional review sites.
He does talk some ignorant tosh though. For instance the very first thing he posts is this.
"Ergonomics: the Mac Pro is about as quiet as it gets, and all its memory, hard drives, etc are easily accessible without tools. It is esthetically pleasing also, fitting into even “hip” offices, especially in Redmond.
Massive, fast storage: The Mac Pro accommodates four internal drives for 6TB of internal storage (4 X 1.5TB) , and with a special bracket as much as 9TB, all using the built-in SATA connectors. "
Adding two extra hard drives to take my MacPro up to six from the normal max of 4 bays internally, involved removing the DVD drive and some other fiddly bits to then spend quite a while threading SATA cables through awkward gaps to plug into the obscured extra two SATA ports, which most people are unaware of. I had to also hack the end connectors of the SATA cable to allow it to fit in the sockets. I now also have no room for DVD writer, unless I use an external one.
My PC tower has 9 drives in it and 2 optical drives and isn't any bigger than the MacPro. Which is indeed a nice quiet box, except for the dreadfully loud DVD writer and was another reason I removed it. But it's decidedly limited and fiddly if you like lots of HDs. I have another 5 External drives [4 eSATAs] with loads of bulky cables and power adaptors making the neat MacPro look a lot less tidy and I still need more HDs attached . It would be a lot better if the MacPro was a bit more professional and had some options [a dirty word at Apple] like a slightly taller case if required with 2 rows of 4 HD caddies.
Going a little OT here sorry. But Lloyd doesn't make me think too highly of his 'scienticfic' testing when he gets such basic facts wrong and doesn't even know if you can run Win64 on a Mac [you can]. He just reads like a fanboi and would treat his camera tests with some scepticism as a result.
Shame as some of his other advice on the subject is actually very good.
-
Regarding the DxO results I'm puzzled
The problem is, that they do not publish the images and the exact measurements they have taken; thus one can not reconstruct their results and critiques can not be qualified.
-
I created a demonstration for the support of my statements re the D3X and 5D2 evaluations.
a. I used the raw files from Imaging Resources, because they are suitable (more or less), and I don't have any others suitable. They were recorded in 12bit mode (why on earth); this would pose a problem *if* the DR of the D3X were really so large; however, it is not.
b. These shots are not exposed as low as I would prefer them to be. Therefor I used only the red channel, which is more than 1 EV darker than the green and about 0.8 EV darker than the blue on grey patches (under the current illumination).
c. I selected only such patches, which are enough clean and evenly illuminated; not like this: http://www.panopeeper.com/Noise/D3Xvs5D2/C...yTestSheet.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/CrappyTestSheet.GIF)
d. It is not necessary to have any high exposure when measuring the DR; only the dark end is interesting.
e. The 5D2 has a small disadvantage in the measured ISO (less than 1/3 EV); this affects the noise characteristics, but not the dynamic range. The test images have compensated for that difference by the shutter time.
f. I highlighted with red one or two numbers over the image. The last number in the upper raw per color group (identified as DR) represents the average pixel intensity on the selection, measured from saturation downwards. The last number of the color group in the second row (identified as NP) is the noise in the salection, measured as standard deviation, expressed as percentage of the average intensity. 100 times the reciprocate of this number is the SNR. NOTE: this number has nothing to do with "stop".
Example: in http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00100_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00100_b.GIF) the intensity is -8.62 EV, the noise is 19.48%, which corresponds roughly to SNR=5.
g. The D3X raw pixel values are not linear (nor are the 5D2's, but they are compensated for), therefor the very dark patches appear to be darker than they really are; I have not highlighted them.
Finally, the measurements:
ISO 100
a. Medium dark patch: the intensities are virtually identical: -6.92 EV vs -6.97 EV. The D3X patch is cleaner; noise: 6.59% vs. 7.56% (very low, visually not perceivable).
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00100_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00100_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00100_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00100_a.GIF)
b. Dark patch: the advantage of the D3X is now visually perceivable; the noise is only 15.4%, vs. 19.48% of the 5D2, at the intensity -8.62 EV.
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00100_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00100_b.GIF;)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00100_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00100_b.GIF;)
The question is, how much difference in EV does this difference in noise represent. To find it out, I searched for a patch in the same 5D2 image with closely comparable noise. I found one with noise 15.55% (very close to the 15.4%), at the intensity -8.29 EV. This means, that the 5D2 requires 8.62-8.29=0.33 EV more light to have the same level of noise as the D3X. Measurements on different spots show like or slightly larger difference, and it can be somewhat greater in even darker spots.
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00100_c.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00100_c.GIF)
Here it would be interesting to see darker patches. Unfortunately, the community of this forum is more interested in flaming than in providing usable raw files.
ISO 200
The difference is gone on the medium dark patch (actually, the D3X sghows slightly higher noise, but negligable).
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00200_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00200_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00200_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00200_a.GIF)
The darker patch shows a slight advantage of the D3X:
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00200_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00200_b.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00200_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00200_b.GIF)
ISO 400
The 5D2 has clearly lower noise than the D3X, on the darker patch as well:
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00400_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00400_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00400_a.GIF; (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00400_a.GIF;)
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00400_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00400_b.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00400_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00400_b.GIF)
ISO 800
The gap is widening to the advantage of the 5D2:
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00800_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00800_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00800_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00800_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO00800_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO00800_b.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO00800_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO00800_b.GIF)
ISO 1600
The D3X is clearly not a high ISO camera.
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO01600_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO01600_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO01600_a.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO01600_a.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/N...ISO01600_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/NikonD3X_Noise_ISO01600_b.GIF)
http://www.panopeeper.com/Noise/D3Xvs5D2/C...ISO01600_b.GIF (http://www.panopeeper.com/Noise/D3Xvs5D2/Canon5DMkII_Noise_ISO01600_b.GIF)
UPDATE: semicolons removed from the image URLs
-
If there were no read noise, both the shadow and the mid-tone zone curves would go by the same slope, 6dB per EV, so that there is no "knee". Only D3x came close to the ideal behavior, while all the other sensors fall short more or less, if you check their "Full SNR" curves out one by one.
These curves also enable to derive DR measures at any SNR definitions above 1:1 (0dB).
http://www.dxomark.com/index.php/eng/Techn...ization/Summary (http://www.dxomark.com/index.php/eng/Technologies/Noise-characterization/Summary)
I see now what you meant by 3dB/EV on the D3x.
There is something I always wonder about the Full SNR curves in DxO Mark, even if I consider them very valid conceptually:
1. What does the sign 'Gray scale' in the X axis mean? Doesn't the X-axis on those plots represent any undemosaiced and unprocessed RGB RAW values?
2. The 100% mark in the right end, is camera's sensor saturation point, or the maximum RAW value in the sensor's bit scale? I mean: for the 5D a 12-bit camera, the 100% represents the value 4095 or the real saturation point of that camera, which is 3692 at ISO100?
BR
-
Lloyd just reported that D3x is also much cleaner (no hot pixels or streaking) at long exposures (30 seconds and over) WITHOUT dark frame subtraction ("Long Exposure NR") than the 5DII, which he says come with a "dirty" sensor.
Even when NR is disabled, Nikon applies noise reduction to long-exposure raws designed specifically to remove hot pixels. (Incidentally, it also removes stars, which is frustrating for us astrophotographers).
If he wanted to see what Nikon raw is like without noise reduction, he should *enable* dark frame subtraction in the camera (enabling this causes the camera to disable the other hot pixel noise reduction), take an exposure, then turn off the camera after the light frame (but before the dark frame can finish). This will cause the dark frame subtraction *and* the hot pixel noise reduction to both be skipped. (The cost, of course, is a bunch of manual labor every time you take a light frame.)
-
take an exposure, then turn off the camera after the light frame (but before the dark frame can finish). This will cause the dark frame subtraction *and* the hot pixel noise reduction to both be skipped. (The cost, of course, is a bunch of manual labor every time you take a light frame.)
And is the NEF file properly saved into the memory card when doing this 'operation'? that sounds a bit strange.
-
I understand and appreciate your elaboration (your long post on the next page where many if not all the links are not working). However to me you did not study and really understand the two 3D charts I posted which made your results largely irrelevant.
On the DxO exact measurements: if you were patient enough and mouse over the DxO charts (on their web site not my extracted images) you will see the actual data values pop out in small yallow squares.
Read their technical documents and understand what their numbers really mean.
The problem is, that they do not publish the images and the exact measurements they have taken; thus one can not reconstruct their results and critiques can not be qualified.
-
I created a demonstration for the support of my statements re the D3X and 5D2 evaluations.
a. I used the raw files from Imaging Resources, because they are suitable (more or less), and I don't have any others suitable. They were recorded in 12bit mode (why on earth); this would pose a problem *if* the DR of the D3X were really so large; however, it is not.
With all due respect.... I stopped reading here... how on earth are you hoping to be able to convince anyone that testing with 12 bits file is going to give representative results for a camera that claims to be the first one able to make useful use of 14 bits raw files?...
Is that approach more "scientific" than those proposed by DxO and Lloyd?
Cheers,
Bernard
-
With all due respect.... I stopped reading here...
Well...my post is not meant as an introductory course in the understanding of raw image data.
-
However to me you did not study and really understand the two 3D charts I posted which made your results largely irrelevant
To me you did not study and really understand my demonstration, which make your reasult largely irrelevant (sorry for the wrong URL's; an internet savy person would have noted the strange semicolon at the end of the URLs).
Now, that I repaired the links, you should have no difficulty to relate my statements to actual measurements. Note: these are measurements reproducable for anyone from the used images - in contrast to DxO's and Loyd's numbers, which I can type in in any program as "proof".
-
What you have said is that basically Lloyd was right and there are solid recommendations in his Mac performance site,
No I said he talked some rubbish and some good stuff, but as he started with the rubbish.....
but you need more hard drives beyond the Mac Pro's capacity of 7,
It's ostensibly 4 unless you want to start doing hacking cables and digging around in your Mac, removing optical drives and then it's 6 SATA drives.
...and you know things like the tricks to run Win64 on a Mac Pro he didn't.
Uh yeah, run Bootcamp and insert W64 DVD into Mac DVD drive!! But shush don't tell anyone else.
Congratulations for being such a elite computer user requiring at least two 4-HD raws inside your desktop machine,
I'm elite now! That's good to know, I shall go away and practice my sneering in a downwards manner.
Nothing elitist about accumulating data. 3 HDS for storage is pretty damn small these days particularly if you mirror for safety and archive. Which is a 3 drives already. But your archive should be elsewhere really, so only 3 in machine. Which leaves 1 bay for a scratch disc, 1 bay for OS and Progs and 2 for storage. Not a lot really. Certainly not elitist to need more.
but isn't that true for the most of the rest (needing less than 7 drives, etc.) there are still things to learn from the site, such as how to separate data and applications, and how to set up a Photoshop specific scratch volume or a RAM disk to speed things up?
How do you know what is good advice, accurate advice or fanboi wittering?
And is it true for them the Mac Pro is still quiet and tidy?
Not when the DVD spins up and one reason why I took it out.
I do competitive photography, but I do not happen to know that much about computer. Is this makes me disqualified to express my findings here in a photography forum, or my own blogs?
Nope, never said that was the case.
-
Well...my post is not meant as an introductory course in the understanding of raw image data.
Well... I am not sure why you even bother commenting on this issue without having proper testing material (14 bits images). I'll read your stuff once you have done your homework and tested (14 bit files) what has to be tested and not something else (12 bits files).
In case you respect Thom Hogan more than the Field medal winners/MIT professors belonging to the DxO scientific board, he is claiming that he sees one extra stop DR in the D3x compared to any other Nikon DSLR at base ISO (see this updated review just published - http://www.bythom.com/nikond3xreview.htm) (http://www.bythom.com/nikond3xreview.htm)). Considering that the D3 is also considered to have more DR at base ISO compared to any Canon DSLR you get claims that are pretty close to those made by both DxO and Lloyd.
So again, these people have tested the camera with suitable 14 bits raw files, you have not, yet you keep making sweeping statements on the D3x vs 5DII.
The most reasonnable thing to do would be to try to find suitable data, test again and report back to us. I will then consider your inputs as one of the information sources and weight it fairly relative to the others... final comment from me on this matter.
Cheers,
Bernard
-
And is the NEF file properly saved into the memory card when doing this 'operation'? that sounds a bit strange.
That is how it works on previous Nikon DSLR; I don't have a D3X to test for myself. It is strange and laborious, just like the many other workarounds photographers are foced to use if they want to get the highest quality images (e.g. UniWB, negative EC instead of ISO 1600+, MLU in liveview, etc.).
-
Well... I am not sure why you even bother commenting on this issue without having proper testing material (14 bits images)
[
I see I have to reveal this to you: Nikon is offering not only 12bit, but LOSSY 14 bit, even worse, LOSSY 12 bit as well.
If you think that the 12bit recording is inferor to the 14bit, you should create a thread about that subject, and report your factual findings to Nikon. Isn't it a bande of suckers, selling a camera fo $6000-8000 or whatever, and fooling the customers into using seriously inferior options?
I for myself would not use 12bit, nor lossy compression, but I am not a Nikon owner.
Back to the subject: the measurements on 12bit do not indicate, that any big change is expectable with greater bit depth.
In case you respect Thom Hogan more than the Field medal winners/MIT professors belonging to the DxO scientific board, he is claiming
I don't give a fig for his or others' credentials; I am writing about sensors, not about testers. I wonder, why you are not questioning their measurements: you are accepting numbers presented in nice charts without any proof.
I know almost nothing of LLoyd, but what I do know is more than enough: if he is testing these cameras at ISO 50 and ISO 6400, then he knows a LOT less about these cameras than I do without ever having held them in my hands.
-
When there are doubts as to the relevance of scientific tests or reviews of camera performance, one simply has to carry out one's own tests. This is what I did after initial reports on the performance of the D3 claimed it had over a stop (and maybe up to 2 stops) lower noise at high ISO than any other camera on the market at the time.
I was not able to hire a D3 (they were in such short supply at the time) so instead made arrangements with the Nikon agent in Bangkok to compare their demonstration copy of the D3 with my own 5D, in the store. Unfortunately, they wouldn't let me borrow their D3 and take it outside for thorough testing because they had only one copy for demonstration purposes.
Nevertheless, I was able to satisfy myself that at ISO 3200 and above, the D3 had approximately 1/3rd stop to 1/2 stop noise/DR advantage over the 5D.
However, I didn't bother and indeed didn't have the time to compare both cameras at base ISO. If I had done so, I would no doubt have discovered that the main advantage of the D3 with respect to noise and DR was not at high ISO but at base ISO where I now believe it's at least one stop better than the 5D.
-
I see I have to reveal this to you: Nikon is offering not only 12bit, but LOSSY 14 bit, even worse, LOSSY 12 bit as well.
If you think that the 12bit recording is inferor to the 14bit, you should create a thread about that subject, and report your factual findings to Nikon. Isn't it a bande of suckers, selling a camera fo $6000-8000 or whatever, and fooling the customers into using seriously inferior options?
I for myself would not use 12bit, nor lossy compression, but I am not a Nikon owner.
This isn't you Gabor, is it? Tell me that somebody hacked your login/password and is trying to damage your reputation with this non sense?
How is the availability of lower image quality options in a camera relevant when discussing the best output it can produce? Why did you stop there and didn't you comment on the stupidity of Nikon providing the option to shoot in medium resolution jpgs? Using the same logic you are using above, you could have written "If Nikon provides medium resolution jpgs, it must mean that the quality of a medium resolution jpg file is as good as that of a 14 bits raw".
You know full well that 12 bits/lossy options with the D3x can be useful sometimes because of the smaller file sizes and higher frame rate they boast. The availability of these options says nothing about the quality gap between them and the best available 14 bits mode.
Back to the subject: the measurements on 12bit do not indicate, that any big change is expectable with greater bit depth.
How on earth are you able to make assumptions on the behaviour of a 14bits file from a 12 bits file? How rigorous and scientific is this approach?
I don't give a fig for his or others' credentials; I am writing about sensors, not about testers. I wonder, why you are not questioning their measurements: you are accepting numbers presented in nice charts without any proof.
I stopped believing in conspiracies a few years back. I have no reason to doubt the credibility of these information sources, and you have so far not brought any evidence to the table indicating that they could be wrong. You are measuring something different.
All I am saying is, please do an apple to apple comparison by measuring the same thing they are measuring.
If you are not interested in doing this, please avoid commenting on the quality of their results.
Cheers,
Bernard
-
How is the availability of lower image quality options in a camera relevant when discussing the best output it can produce? Why did you stop there and didn't you comment on the stupidity of Nikon providing the option to shoot in medium resolution jpgs?
Bernard, you too are posting and reading on other forums as well, for example DPReview. Thre is a thread in the D3 forum called "NEF Compression". Several members were expressing their firm conviction, that compression means always loss; like if you intend on editing your best bet is not to compress. I read many other threads stating, that the lossy compression does not cause any visible loss (ask Emil Martinec about that).
In short: expecting the customers making such decisions is far too much.
However, I reiterate: I am a proponent of lossless compressing in 14bit depth.
How on earth are you able to make assumptions on the behaviour of a 14bits file from a 12 bits file? How rigorous and scientific is this approach?
This is quite simple: look at the involved pixel values and the standard deviations.
It IS possible, that the D3X shows somne extra advantage in the very-very deep shadows, but
1. do you find it useful, when the camera is more noisy in the "normal shadows" but less noisy in the "very deep shadows"?
2. only a few of us know, thay all Nikon cameras are cutting off the worse part of the very dark regions; that is the reason, that you see the horizontal or cross-hatch pattern with the 5D2 bu not with any Nikon. This is the question of raw processing.
Anyway, there is something I don't understand here. You do have the D3X; you developed a template for the DR measurement. Why don't YOU make suitable shots and publish the raw files instead of this fruitless discussion? Then we could see the difference between the noise in 12bit and 14bit, as well as the very deep shadows, instead of speculating about that.
-
Anyway, there is something I don't understand here. You do have the D3X; you developed a template for the DR measurement. Why don't YOU make suitable shots and publish the raw files instead of this fruitless discussion? Then we could see the difference between the noise in 12bit and 14bit, as well as the very deep shadows, instead of speculating about that.
Gabor,
Yes, I do own a D3x, but I have not developped any template for DR measurement.
I do personnaly not doubt the excellence of the D3x DR at low ISO and frankly speaking, my life is already complex enough those days that I am not willing to embark on such a project. Sorry about that.
What I could perhaps do is send you a few 14 bits nefs using yousendit if you provide me with a suitable email address through PM.
Cheers,
Bernard
-
Yes, I do own a D3x, but I have not developped any template for DR measurement
Then I don't remember who made the template for Ray. Sorry for the mixup.
I do personnaly not doubt the excellence of the D3x DR at low ISO and frankly speaking, my life is already complex enough those days that I am not willing to embark on such a project. Sorry about that
This is perfectly all right. In fact, your happiness with the camera should not depend on any evaluation but on what it is doing for you. Evaluations/reviews are meant for before purchasing, not afterwards.
I purchased my 40D after many customers reported their findings, but before DPReview reviewed it. Then I have never read the review; what for?
What I could perhaps do is send you a few 14 bits nefs using yousendit if you provide me with a suitable email address through PM
Thanks for the offer, but
1. I can use only particular images: containing some smooth, non-textured, unicolored, dark, not curving surfaces of very even illumination in the very dark shadows, in different shades. Otherwise measuring the noise is not much worth,
2. I have no personal interest in this question, I have been doing this for those, who might be interested and were misled (IMO) by the mentioned reviews;
3. my email box quota is not enough for even a single file (max. 15 MB); yousendit is the way to go anyway.
However, if you have some "spare" raw files with 1/3 stop ISOs, particularly if clipping occured, only in 14bits, that would be nice to have, so that I can determine the saturation levels, and if those ISOs are fake (like the 5D2's) or associated with some analog gain. Though I don't know why you would be using these ISO steps.
Finally, for the case you don't know this yet: ISO 50 is non-existent and ISO 3200 is fake, so there is no reason to use them with raw.
-
This is totally untrue. Nikon clamps the DC level at the center of the noise distribution is a well known fact, not only a few know, but this is does not mask or bury any pattern noise in the shadows and much above that level. Plus, the extraordinarily clean deep shadows is a brand new phenomenon to the DSLR sensors, as the Nikon's own D3/D700, which also cut the true black noise by half, do not match the cleanness, as well documented in both DxO measures and visual test results.
Take close look at the crops in the Lloyd's open article again. The Canon's pattern noise does not only happen at nearly the pure dark noise level but also way above it, far above where Nikon biased down to zero.
http://www.diglloyd.com/diglloyd/free/Push...acks/index.html (http://www.diglloyd.com/diglloyd/free/PushingTheBlacks/index.html)
Actually, today, Lloyd just posted new comparisons in the DAP (pushing underexposed foreground shadow details by two stops in RAW conversion by setting two stops of exposure compensation, in real landscape situation at ISO 100). There is this free mouse over comparison, which demonstrates the outcomes, but the paid section offers much more details including the full size images. If, as you claimed, Nikon just buried all the deep shadow details, there would be nothing to push to be compared with the Canon.
http://www.diglloyd.com/diglloyd/2009-01-b...0126D3x_vs_5DM2 (http://www.diglloyd.com/diglloyd/2009-01-blog.html#_20090126D3x_vs_5DM2)
I just wish I can have Lloyd's permission to copy his conclusion words here.
2. only a few of us know, thay all Nikon cameras are cutting off the worse part of the very dark regions; that is the reason, that you see the horizontal or cross-hatch pattern with the 5D2 bu not with any Nikon. This is the question of raw processing.
-
Gabor,
Iliah Borg warned you not to use imaging resource images for a noise comparison due to reasons of target quality and shooting methodology
http://forums.dpreview.com/forums/read.asp...essage=30792855 (http://forums.dpreview.com/forums/read.asp?forum=1021&message=30792855)
http://forums.dpreview.com/forums/read.asp...essage=30795654 (http://forums.dpreview.com/forums/read.asp?forum=1021&message=30795654)
Target quality (noise inherent in the target) was mentioned also by THom Hogan in the same thread.
Additionally, DxO state that according to their findings 14bits do make a difference in the case of the D3X
http://www.dxomark.com/index.php/eng/DxOMa...r-the-Nikon-D3X (http://www.dxomark.com/index.php/eng/DxOMark-Sensor/DxOMark-reviews/DxOMark-review-for-the-Nikon-D3X)
There's no doubts in my mind that imaging resource targets were not shot with you analysis in mind. I would tend to trust more someone who shoots for the test (DxO) rather than an analysis a posteriori of the resuts of two independent shoots.
However you chose to ignore all of the above and use imaging resource images shot at 12-bit, to try and demonstrate that DxO are wrong.
Do you honestly believe you have demonstrated this? Your rant about compression (how on earth did you manage to get this in the discussion?) tells me that you don't.
(I want to be clear that I believe you might well be right. However any self respecting scientist would not go public in the aggressive way you have done so, based on the quality of your data. If I was feeling so strongly about proving somebody wrong, I would beg borrow or steal to be able to test under a controlled environment. Otherwise I would just express my doubts and shut up).
I will give you this though. DxO publish numbers and findings without revealing their testing methodology. Thus it is difficult to argue about the way they are testing).
(PS. For anyone inclined to do this, the above dpreview thread contains an interesting exchange between Gabor and member bobn2 which, leaving test image validity aside, boils down to how and if Gabor considers in his measurements the difference in read noise (as opposed to shot noise) which by many accounts is the factor which differentiates the D3X at very low EV levels and seems to differ substantially between 12bit and 14bit captures. For reference one can use DxO's logarithmic FULL SNR diagrams here:
http://www.dxomark.com/index.php/eng/Image...abase/Nikon/D3X (http://www.dxomark.com/index.php/eng/Image-Quality-Database/Nikon/D3X)
http://www.dxomark.com/index.php/eng/Image.../EOS-5D-Mark-II (http://www.dxomark.com/index.php/eng/Image-Quality-Database/Canon/EOS-5D-Mark-II)
Just select the Full SNR diagrams and press the Logarithmic tab
)
PS2. One final word from me in this thread. Quoting Michael, when your eyes disagree with the numbers trust you eyes... In my case, I can only use my eyes indirectly since I have no access to a D3X. Because of this, and until shown differently, I will choose to trust Lloyd's eyes as evidenced in his visual tests which, in any case, seem to be quite in line with at least one set of measurements.
-
Hi NikosR,
Thank you for your contributions -- I really enjoyed reading them.
To me DxO had actually quite adequately documented their testing methodology, in the more than 20 pages in their Technologies section:
http://www.dxomark.com/index.php/eng/Technologies (http://www.dxomark.com/index.php/eng/Technologies)
For instance, their noise measurement protocols. They are the only tester claimed to follow the ISO standards, as far as I know. Some of the descriptions and formulas are not very inviting for quick reading there, but they did try to clarify what they measure and how the numbers are defined and derived. I am sure there are rooms for clarification, organization, and comprehensiveness though.
http://www.dxomark.com/index.php/eng/Techn.../Noise-protocol (http://www.dxomark.com/index.php/eng/Technologies/Testing-protocols/Noise-protocol)
Also, as I did, you will find your links (to the dxomark site) did not lead to where you wanted. They do not provide direct links to most of their charts (but you can link to the two charts I extracted and posted in this thread).
Best regards,
Leping
I will give you this though. DxO publish numbers and findings without revealing their testing methodology. Thus it is difficult to argue about the way they are testing.
-
Also, as I did, you will find your links (to the dxomark site) did not lead to where you wanted. They do not provide direct links to most of their charts (but you can link to the two charts I extracted and posted in this thread).
Indeed. Just press on the Full SNR tab in the links provided and choose Logarithmic.
It's obvious that there is a unique characteristic in the D3X curves (as compared to other cameras) which, not knowing better, I can only attribute to very low read noise.
Leping, thanks for pointing me to their noise protocol write up. I had missed that although I had gone over some of their other write ups in that section.To a novice like me, their methods sound much more controlled than measuring off files produced by imaging resource.
(PS. One interesting thing I noted is that they are measuring all channels for noise. Gabor argues, not unconvincingly, in the aforementioned dpreview thread, that it should not matter which channel you're measuring in the raw data. I'm wondering and the only explanation I can give, if indeed the channel makes a difference, is the potential of metamerism effects or non-linear wavelength response in sensors leading to difference in shot noise depending on wavelength )
PS2. I've just gone over Lloyd's latest comparison in DAP in which he extracts details out of the shadows in a very high contrast image by pushing in the raw converters by 2 stops. I don't care if somewhat different results might have been achieved by different selection of raw converters. The difference between the 2 cameras can only be descibed by 'Wow!'. Since for a digital capture to be a photograph it has to go through the demosaicing phase, what does it matter if the difference in noise in the raw files can be measured as 0.5EV or 1.7EV? Just looking at Lloyd's comparison does not leave anyone with a doubt about the GREAT difference in low level detail that can be extracted in the two cameras and the quality differences resulting from the 2 stop push.The difference is well beyond pixel-peeping territory. It's about the difference I see between my D70 and my D700. It jumps up and smacks you in the face. I hope I'm not infringing any copyright here since the same can be seen in his free overview.
-
Then I don't remember who made the template for Ray. Sorry for the mixup.
Ray's template is very basic, but accurate enough for him. If image of very contrasty scene from camera A, with ETTR exposure, appears to have the same noise in the deepest shadows as image from camera B, when image from camera B is overexposed by one stop, then camera A has has 1 EV greater DR than camera B.
Nothing could be simpler. I don't know what all the fuss is about.
-
Nonsense. Everything is a compromise. I don't see how offering an option (you don't HAVE to use Auto ISO) is anything but a good thing.
I agree, that would be an awesome feature. I really love the Auto ISO on the 5DMII which I didn't have in my previous 5D. An option as you describe would be useful in some situations.
-
Bernard,
Doesn't the D3x has two readout modes? A fast 12-bit and a slower 14-bit?
Best regards
Erik
This isn't you Gabor, is it? Tell me that somebody hacked your login/password and is trying to damage your reputation with this non sense?
How is the availability of lower image quality options in a camera relevant when discussing the best output it can produce? Why did you stop there and didn't you comment on the stupidity of Nikon providing the option to shoot in medium resolution jpgs? Using the same logic you are using above, you could have written "If Nikon provides medium resolution jpgs, it must mean that the quality of a medium resolution jpg file is as good as that of a 14 bits raw".
You know full well that 12 bits/lossy options with the D3x can be useful sometimes because of the smaller file sizes and higher frame rate they boast. The availability of these options says nothing about the quality gap between them and the best available 14 bits mode.
How on earth are you able to make assumptions on the behaviour of a 14bits file from a 12 bits file? How rigorous and scientific is this approach?
I stopped believing in conspiracies a few years back. I have no reason to doubt the credibility of these information sources, and you have so far not brought any evidence to the table indicating that they could be wrong. You are measuring something different.
All I am saying is, please do an apple to apple comparison by measuring the same thing they are measuring.
If you are not interested in doing this, please avoid commenting on the quality of their results.
Cheers,
Bernard
-
I agree, that would be an awesome feature. I really love the Auto ISO on the 5DMII which I didn't have in my previous 5D. An option as you describe would be useful in some situations.
How does that work, Andrew? According to dpreview's preview of the 5D2 it has, to quote, " Auto ISO (100 - 3200) in all modes except manual". I presume that means in aperture priority mode the camera can select a reasonably fast shutter speed and ISO to suit. Does it vary the balance according to the lens focal length?
-
According to dpreview's preview of the 5D2 it has, to quote, " Auto ISO (100 - 3200) in all modes except manual". I presume that means in aperture priority mode the camera can select a reasonably fast shutter speed and ISO to suit. Does it vary the balance according to the lens focal length?
It appears to balance according to the lens. Not totally sure what's happening under the hood but for fast action, or casual shooting, the mode is really sweet.
-
Doesn't the D3x has two readout modes? A fast 12-bit and a slower 14-bit?
Hello Erik,
I am sure how they do it. I do not have any priviledged relationship with Nikon and don't have any information besides what is publicly available.
The level reached by this generation of bodies (whether they are from Sony, Canon or Nikon) is such anyway that I feel I can stop to worry too much about the internal magic and just use the camera.
In a way, the DR provided by the D3x reminds me of B&W shooting, just get the exposure approximately right and deal with it in post. Of course, you can get much better results if you nail the exposure and the quality of those perfect files viewed at 100% on screen is such that you are naturally drawn to perfection, but the fact is that in most cases the camera ends up being very forgiving.
(http://farm4.static.flickr.com/3408/3231744828_e677c8614f_o.jpg)
I assume that the same can be said about the 5DII and A900.
Cheers,
Bernard
-
Hi!
According to Thom Hogan ( http://www.bythom.com/nikond3xreview.htm (http://www.bythom.com/nikond3xreview.htm) ) the D3X shoots 5 FPS at 12 bits but only 1.8 FPS at 14 bits. A suggestion was on this forum that 14 bit readout is achieved by combining two 12 bit readouts with different preamplifiaction before the column DA converters.
This may explain some of the differences between DxO findings and Panopeepers findings. Panopeeper downloaded 12 bit data from Imaging Resource, and that could very well be shot in 12 bit mode and therefore having DR similar to A900. DxO may have used 14 bit mode in their lab shooting and therefore have extended DR.
My concern is not DR on it's own, but the discrepancy between Panopeeper's and DxO's results.
Best regards
Erik
Hello Erik,
I am sure how they do it. I do not have any priviledged relationship with Nikon and don't have any information besides what is publicly available.
The level reached by this generation of bodies (whether they are from Sony, Canon or Nikon) is such anyway that I feel I can stop to worry too much about the internal magic and just use the camera.
In a way, the DR provided by the D3x reminds me of B&W shooting, just get the exposure approximately right and deal with it in post. Of course, you can get much better results if you nail the exposure and the quality of those perfect files viewed at 100% on screen is such that you are naturally drawn to perfection, but the fact is that in most cases the camera ends up being very forgiving.
(http://farm4.static.flickr.com/3408/3231744828_e677c8614f_o.jpg)
I assume that the same can be said about the 5DII and A900.
Cheers,
Bernard
-
This may explain some of the differences between DxO findings and Panopeepers findings. Panopeeper downloaded 12 bit data from Imaging Resource, and that could very well be shot in 12 bit mode and therefore having DR similar to A900. DxO may have used 14 bit mode in their lab shooting and therefore have extended DR.
That's exactly what I have been commenting about in this thread. DxO and all the other testers who have had a D3x in their hands have been testing the 14 bits mode while Gabor has so far only worked on 12 bits samples.
Not knowing how the camera works, it is impossbile to draw any conclusion about 14 bits files from 12 bits files. This is by no means a criticism of Gabor's testing methodology.
Cheers,
Bernard
-
Bernard,
I really appreciate that comment. I felt we had a little to much agitation on this thread recently.
Best regards
Erik
That's exactly what I have been commenting about in this thread. DxO and all the other testers who have had a D3x in their hands have been testing the 14 bits mode while Gabor has so far only worked on 12 bits samples.
Not knowing how the camera works, it is impossbile to draw any conclusion about 14 bits files from 12 bits files. This is by no means a criticism of Gabor's testing methodology.
Cheers,
Bernard
-
Hi!
According to Thom Hogan ( http://www.bythom.com/nikond3xreview.htm (http://www.bythom.com/nikond3xreview.htm) ) the D3X shoots 5 FPS at 12 bits but only 1.8 FPS at 14 bits. A suggestion was on this forum that 14 bit readout is achieved by combining two 12 bit readouts with different preamplifiaction before the column DA converters.
This may explain some of the differences between DxO findings and Panopeepers findings. Panopeeper downloaded 12 bit data from Imaging Resource, and that could very well be shot in 12 bit mode and therefore having DR similar to A900. DxO may have used 14 bit mode in their lab shooting and therefore have extended DR.
My concern is not DR on it's own, but the discrepancy between Panopeeper's and DxO's results.
Best regards
Erik
Using two different 12-bit amplifications would yield telltale signs in the RAW data -- highlights would only populate the top 12 bits of the 14. I haven't analyzed D3x RAWs though, so I can't say whether this is what's going on.
As far as Gabor's results, I suspect he isn't measuring what he claims to be; I gave a detailed analysis in a parallel thread over at DPR:
http://forums.dpreview.com/forums/read.asp...essage=30809618 (http://forums.dpreview.com/forums/read.asp?forum=1021&message=30809618)
-
PS2. I've just gone over Lloyd's latest comparison in DAP in which he extracts details out of the shadows in a very high contrast image by pushing in the raw converters by 2 stops. I don't care if somewhat different results might have been achieved by different selection of raw converters. The difference between the 2 cameras can only be descibed by 'Wow!'. Since for a digital capture to be a photograph it has to go through the demosaicing phase, what does it matter if the difference in noise in the raw files can be measured as 0.5EV or 1.7EV? Just looking at Lloyd's comparison does not leave anyone with a doubt about the GREAT difference in low level detail that can be extracted in the two cameras and the quality differences resulting from the 2 stop push.The difference is well beyond pixel-peeping territory. It's about the difference I see between my D70 and my D700. It jumps up and smacks you in the face. I hope I'm not infringing any copyright here since the same can be seen in his free overview.
Good lord, as a primarily night photographer (and constantly dealing with those low values) I'm scared to subscribe now as I may end up feeling like I should have waited longer instead of buying the 5D2!
In any case, while way over my head, this discussion is interesting and I appreciate all the info those who have contributed are posting!
-
Good lord, as a primarily night photographer (and constantly dealing with those low values) I'm scared to subscribe now as I may end up feeling like I should have waited longer instead of buying the 5D2!
In any case, while way over my head, this discussion is interesting and I appreciate all the info those who have contributed are posting!
Be aware that we're talking base ISO here.
-
Be aware that we're talking base ISO here.
Yeah, I've been making heavy use of the higher ISO capabilities of the 5D2 so there's really no way I could justifiably get rid of it...but that the potential for such clean deep shadows exists is exciting.
-
IQ and dynamic range at 100 ISO seems really the best
-
According to Thom Hogan ( http://www.bythom.com/nikond3xreview.htm (http://www.bythom.com/nikond3xreview.htm) ) the D3X shoots 5 FPS at 12 bits but only 1.8 FPS at 14 bits. A suggestion was on this forum that 14 bit readout is achieved by combining two 12 bit readouts with different preamplifiaction before the column DA converters.
Another proposal (also made for the 14-bit mode of the D700) is that the same photosite charge is run through the sensor's ADC four times, each with the 12-bit ADCs of the Sony sensor, and the digital levels are summed. With the 12-bit ASC levels being integers from 0 to 4095, the sum is an integer from 0 to 16380, nicely fitting the 14-bit range from 0 to 16383.
That would average over time-random variation in ADC output for the same input. Four reads would only be expected to improve S/N by a factor of sqrt(4)=2, or a "one bit improvement", but if that is enough to go beyond the S/N ratio that 12-bits can record, the normal jump in bit depth is to 14 rather than 12, especially as that allows simply adding the four twelve-bit levels.
Some might raise the objection that A/D conversion destroys the charge in the sense capacitor at the bottom of the column. But that charge gets there from the photosite using charge gain amplification, not actual transfer of electrons, so the photosite-to-sense-capacitor transfer could be repeated. Doing that would have the advantage that the averaging also applies to the noise in the charge amplification process, which I have reason to believe is a significant source of noise in CMOS sensors.
-
That would average over time-random variation in ADC output for the same input. Four reads would only be expected to improve S/N by a factor of sqrt(4)=2, or a "one bit improvement", but if that is enough to go beyond the S/N ratio that 12-bits can record, the normal jump in bit depth is to 14 rather than 12, especially as that allows simply adding the four twelve-bit levels.
Hmm. If I recall correctly, one frequent poster to these forums has already clearly stated that of the 14 bits, 13 are clearly used, but the 14th isn't necessary to encode the signal range. If so, that makes sense considering what you write above, and vice versa.