Ad
Ad
Ad
Pages: « 1 ... 4 5 [6] 7 »   Bottom of Page
Print
Author Topic: D800E -Truly Amazing Low-Light Performance  (Read 22677 times)
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7703


WWW
« Reply #100 on: November 24, 2012, 07:02:26 PM »
ReplyReply

Hi,

In my view your interpretation of DR is wrong.

The other question is tricky, the best answer I can give is a lot if you combine sharpening, masking and noise reduction.

Best regards
Erik


From the Full SNR curve, you can read DR and SNR18%.
Is that right  Huh

Another question, with LR or DXO, what's the SNR possible improvement (in dB) without significant loss of resolution ?!
Logged

jeremypayne
Guest
« Reply #101 on: November 24, 2012, 07:12:20 PM »
ReplyReply

Everything except the actual measurements of the RAW data from a bunch of meticulous scientists with PhDs, at DXO.

After some further investigation ...

How would you interpret this data, Ray?

http://home.comcast.net/~NikonD70/Charts/PDR.htm
Logged
allegretto
Sr. Member
****
Offline Offline

Posts: 530


« Reply #102 on: November 24, 2012, 08:50:20 PM »
ReplyReply

I appreciate all the knowledge to be gained about digital images here as there are so many here with so much to share. But of course comparing graphs is far more boring than actually testing for oneself. Further, though manufacturers try to maintain strict control, it has been my experience with very sophisticated electronic equipment that there is some sample to sample variation. So while comparing DxO, senseorgen and comcast data is intriguing, it may be time to evaluate for myself.

I have a D4 and love it. It really takes pictures in the dark. But am contemplating a D800e or perhaps a D600 for a lighter weight alternative and the advantage to having a lens already mounted. Beside that, the 800e would allow me to be a "Big Cropper" when a long tele is unavailable. (is this a great time to be a photographer or what?).

So took a series of photos with the lens cap on, all 1 sec exposures at every ISO rating, all in LR 4.2. When contrast is pushed to 100 and 1:1 there is very little noise up to 3200 (as in almost none), at 6400 I can see a bit, but noting really until 25200 and even then not much. Only when we go above that is there much to see. 50K is still pretty good, but 102K and 204K is noticeably more.

If exposure is pushed the noise is far more evident but not sure how much is LR artifact so maybe that's not a good comparison.

Could some of you more experienced and educated folks tell me how to establish where the "knee" is where adding ISO is detrimental? And perhaps how to discern the point where it becomes "ISO Max" as far as image quality goes?

Thanks, as always...
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #103 on: November 24, 2012, 10:37:06 PM »
ReplyReply

From the Full SNR curve, you can read DR and SNR18%.
Is that right  Huh


Thierry,

The Full SNR curve is revealing but I get the impression that it applies spefically to the pixel without regard to sensor size, so for example, the Full SNR curve for the D800 will be almost identical to the Full SNR curve for the D7000, but in photographic images of the same scene the D800 will exhibit noticeably better SNR.

In your example in attached image showing a vertical red line on the right indicating DR, you've missed out the necessary red line on the left of the graph to indicate the range. Dynamic Range is a range, from the brightest highlights that contain detail, to the darkest shadows that contain detail.

In the Full SNR log graph you have shown, 0.1% grey on the horizontal axis, and less, would represent the deep shadows, and this is where the men are separated from the boys.  Grin

The rather large pixel of a camera like the Canon 5D has an SNR of about 5dB at 0.1% grey, whereas the much smaller pixel of the Nikon D800 has an SNR of about 16dB at 0.1% grey, a vast improvement. The fact that the D800 sensor has many more pixels than the 5D, increases that advantage even further at equal image size.

Quote
Another question, with LR or DXO, what's the SNR possible improvement (in dB) without significant loss of resolution ?!

I've got no idea, Thierry. When I process my images, I use my eyeballs to determine what is acceptable noise and acceptable resolution.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #104 on: November 24, 2012, 11:12:37 PM »
ReplyReply

After some further investigation ...

How would you interpret this data, Ray?

http://home.comcast.net/~NikonD70/Charts/PDR.htm

The data seem to be broadly in line with DXOMark's results, Jeremy. Shadow noise that is photographically acceptable will inevitably be a subjective opinion, but I think most people would agree that noise at the engineering limit of DR, such as noise in the 14th stop for the D800, or noise in the 11th stop for the canon 5D, is unacceptable.

The following graph indicates that the D800e has a 'Photographic DR' of 11.5 stops, about 3 stops down from the engineering DR which is described at DXOMark as over 14 stops. The Canon 5D has a PDR of about 8.25 stops which is close to 3 stops down from the DXO rating of a fraction over 11 stops.

The only surprise is the difference between the D7000 and the D800E. At equal image sizes (print mode on the graphs), the D800 has an advantage of close to 1/2 a stop, according to DXOMark, whereas the PDR graph shows an advantage of over one full stop.

Whatever the explanation for such a difference, it would tend to fit with the surprise I experienced when processing the underexposed shot of the moth.

Perhaps someone more technically orientated than my self can provide an explanation. I'm really just a simple guy trying to get the best visual quality from the scenes I capture.  Grin

Logged
jeremypayne
Guest
« Reply #105 on: November 24, 2012, 11:38:10 PM »
ReplyReply

The data seem to be broadly in line with DXOMark's results, Jeremy.


I meant about the RX100 ... it isn't quite the same as DXOMark it would seem.
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7703


WWW
« Reply #106 on: November 25, 2012, 12:04:27 AM »
ReplyReply

Hi,

I essentially agree with Ray's explanation. 'Photographic DR' has a different interpretation from the normal one used in engineering. DR is normally based on SNR=1 condition. My guess is that it is at this level that signal is discernible from noise. My understanding/presumption is that 'Photographic DR' is simply based on another threshold, like SNR=8, that is signal 8 times stronger than noise.

Now, the normal DR is defined as FWC/read noise. There is another factor affecting noise and that is shot noise. Shot noise is a property of light. If you collect more light, the noise will be less. When you move the threshold for DR to stronger SNR ratios, it also mean that shot noise will be more dominant.

Making either the sensor larger or increasing the full well capacity of the pixels will reduce shot noise.

If you are really interested in noise behavior, this article is an excellent read: http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

Best regards
Erik



The data seem to be broadly in line with DXOMark's results, Jeremy. Shadow noise that is photographically acceptable will inevitably be a subjective opinion, but I think most people would agree that noise at the engineering limit of DR, such as noise in the 14th stop for the D800, or noise in the 11th stop for the canon 5D, is unacceptable.

The following graph indicates that the D800e has a 'Photographic DR' of 11.5 stops, about 3 stops down from the engineering DR which is described at DXOMark as over 14 stops. The Canon 5D has a PDR of about 8.25 stops which is close to 3 stops down from the DXO rating of a fraction over 11 stops.

The only surprise is the difference between the D7000 and the D800E. At equal image sizes (print mode on the graphs), the D800 has an advantage of close to 1/2 a stop, according to DXOMark, whereas the PDR graph shows an advantage of over one full stop.

Whatever the explanation for such a difference, it would tend to fit with the surprise I experienced when processing the underexposed shot of the moth.

Perhaps someone more technically orientated than my self can provide an explanation. I'm really just a simple guy trying to get the best visual quality from the scenes I capture.  Grin


Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #107 on: November 25, 2012, 12:32:50 AM »
ReplyReply

On second thoughts, I think I've got it!  Wink

DXOMark on their DR graph are describing the outer limits of the dynamic range at each ISO, whereas the Photographic DR describes the SNR over a range from photographically acceptable shadows, at a chosen minimum SNR of 20dB, to the non-blown highlights. Over that entire range, the improved SNR of a sensor such as the D800, which is a bit more than twice the size of the D7000 sensor, should be on average a bit over 3dB better, which translates to a bit over one stop better.

In parts of the scene where SNR is less than 20dB, which might be considered as photographically unacceptable, the SNR advantage of the D800 dimishes to the point where, at the engineering limit or around 14 stops, it is a bit less than 1/2 a stop better than the D7000, as DXOMark describe it.

Everyone agree?  Grin
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #108 on: November 25, 2012, 01:13:18 AM »
ReplyReply

I meant about the RX100 ... it isn't quite the same as DXOMark it would seem.


Oh, I see! You mean this oddity where the RX100 seems to have the same PDR at ISO 125 as it does at ISO 80, and DR is worse at ISO 100 than it is at ISO 125 when one would expect it to be at least as good if not marginally better.

Sorry! I have no explanation for this. Maybe Bart or Bill Claff can help. Since you no longer have the camera, it's not possible for you to check this for yourself, but Thierry seems to have done so. This is what he wrote.

Quote
So, I've made manual shots (@f3.5) of a colorchecker, 1st 1/4s @ISO80, 2nd 1/5s @ISO100, 3rd 1/6s @ISO125.

All shots are properly exposed (ETTR) and need only a slight correction (max 0.1Ev) to obtain exactly same exposure in LR4.2.

Results: RAW with same exposure exhibit more noise @ISO100 and even more @ISO125, so DXOMark DR results are correct !


Logged
thierrylegros396
Sr. Member
****
Offline Offline

Posts: 672


« Reply #109 on: November 25, 2012, 05:56:41 AM »
ReplyReply

Very interresting thread !!

Thanks Erik and Ray !
Logged
jeremypayne
Guest
« Reply #110 on: November 25, 2012, 08:19:01 AM »
ReplyReply

Oh, I see! You mean this oddity where the RX100 seems to have the same PDR at ISO 125 as it does at ISO 80, and DR is worse at ISO 100 than it is at ISO 125 when one would expect it to be at least as good if not marginally better.

It would seem to contradict the DXO data ... and would seem to be in-line with the specs reporting a 125-6400 true ISO range.

I will prolly get another ... But not until I find out if AMEX will replace mine ... Which I'm thinking they will. 
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2838



« Reply #111 on: November 25, 2012, 08:27:00 AM »
ReplyReply

From the Full SNR curve, you can read DR and SNR18%.
Is that right  Huh

Another question, with LR or DXO, what's the SNR possible improvement (in dB) without significant loss of resolution ?!

Your interpretation of DR is not correct. The vertical line at 100% shows the SNR at sensor saturation or clipping in the ADC for various ISOs. The DR is the ratio between the brightest part of the image (100%) and the darkest part of the image with a given SNR (the noise floor). At base ISO the engineering DR as determined by DXO places the noise floor at a SNR of 0 dB or 1:1 (green vertical bar). The percent saturation at this noise floor with your example is 0.0291% and the DR is 100%/0.0291% or 11.75 stops. If you place the noise floor at 12 dB (4:1) as shown by the red bar, the percent saturation at that SNR is 0.336% or 8.22 stops.

The methodology for these determinations is explained by Emil. The difficult part is with log interpolation of the graph value.

Regards,

Bill
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #112 on: November 25, 2012, 08:43:22 AM »
ReplyReply

It would seem to contradict the DXO data ... and would seem to be in-line with the specs reporting a 125-6400 true ISO range.

I will prolly get another ... But not until I find out if AMEX will replace mine ... Which I'm thinking they will. 

Not really. DXO don't show test results for the intermediate ISOs. They test base ISO then at increments of one stop from ISO 100. If you remove the PDR result at ISO 125, the shape of the graph approximates DXOMark's graph for DR. The jump in DR performance at both ISO 80 and ISO 125, or the dip in performance at ISO 100, looks very odd to me.
Logged
thierrylegros396
Sr. Member
****
Offline Offline

Posts: 672


« Reply #113 on: November 25, 2012, 12:24:45 PM »
ReplyReply

Many Thanks Bill !!

It's more clear now Wink
Logged
LKaven
Sr. Member
****
Online Online

Posts: 832


« Reply #114 on: November 25, 2012, 05:06:49 PM »
ReplyReply

Has anyone taken time to study the thermal noise characteristics of the D800/E and how best to mitigate them?  In my view, this is the only thing standing in the way of this camera being usable at the highest gain settings. 

A simple black frame subtraction will go a long way, and LENR is mandatory.  What I think though is that LENR should be a full-time option on this camera.  I see abundant thermal noise at ISO6400 and a handheld shutter speed (1/80th). 
Logged

BJL
Sr. Member
****
Offline Offline

Posts: 5164


« Reply #115 on: November 26, 2012, 10:41:30 AM »
ReplyReply

You're probably right in that respect, BJL. The vast majority of photography is now done with an iPhone or iPad, followed by the so-called P&S camera which in turn is followed by the cropped-format DSLR. The full-frame DSLR seems to be a relative  rarity.
Let me clarify my comments, in the off chance that you genuinely misunderstood me to be talking about photography done with phones and other truly pocketable cameras: what I intended to convey is that a roughly six or seven stop subject brightness range covers the great majority of the sort of photography of interest to readers of these forums. To be a little more specific, this SBR historically covered the great majority of film photography in any common film format and camera type, even if one only considers photography done with SLR's or larger formats. This is made fairly clear by reading the discussions of film photography capture and printing technique in numerous texts on the subject: tools and techniques like graduated ND filters are there, but not needed for the great majority of cases (and seem particularly needed with color transparency film for which six stops of SBR is optimistic). As a mild extrapolation, It seems from my observations that the move to digital cameras has not greatly changed the preponderance of scenes with "normal" SBR.

The rest of you post seems to be a single example of a "scene of unusually high SBR"  which I mention in your quote from me above (and a second case in this thread of severely underexposing the subject matter of main interest!) So we agree that such situations exist: my point was that measurements of relevance to the imge quality achieved in the great majority of one's photography are overall more important that measurements relevant only in a small fraction of extreme situations (or errors or afterthoughts in choice of exposure levels). If your photography involves numerous cases of such exposure difficulties, then I can accept that SNR in parts of the scene that would normally be displayed in deep, low detail shadow regions are more important to you than they are to most of us. For my purposes though, I am still far more interested in measurements that indicate how a camera will handle my intended subject matter when roughly correct exposure level is chosen.
Logged
BJL
Sr. Member
****
Offline Offline

Posts: 5164


« Reply #116 on: November 26, 2012, 11:53:23 AM »
ReplyReply

Please note that the ISO standard gives only one figure for ISO and that figure is based on sensor saturation.
Not at all: that is a myth, partly encouraged by DxOMark's quirky use and selective quotation of the relevant ISO standard, number 12232.

In reality the current ISO standard 12232:2006 specified five ways to measure and report a digital camera's sensitivity:

- One based on signal saturation. (This seems primarily intended for measuring minimum usable exposure index and maximum usable exposure levels, and is related to "base-ISO speed" rather than being suited to callibration of the EI settings on a camera, and is what DxOMark's uses at all exposure index levels.)

- Two, called Snoise40 and Snoise10, that measure low-light handling, and are based on the maximum exposure index levels that give SNR levels of 40:1 and 10:1 respectively under specified conditions. (These are more akin to the ISO definition of film speed, but are rarely used, and seem designed to assess maximum usable exposure index, rather than for calibration of EI settings.)

- Two newer ones added in the 2006 update of 12232, "Standard Output Sensitivity" and "Recommended Exposure Index", which relate to the level placement in output after conversion to an sRGB format like JPEG. (These are the only ISO standards that seem directly intended for use in calibrating the EI settings on cameras, and are what most or all camera makers use for this purpose, in part due to a requirement to do so from the Japanese industry association CIPA, which originated these two standards before their adoption by the ISO.
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7703


WWW
« Reply #117 on: November 26, 2012, 02:09:30 PM »
ReplyReply

Hi,

Thanks for putting things right!

Best regards
Erik


Not at all: that is a myth, partly encouraged by DxOMark's quirky use and selective quotation of the relevant ISO standard, number 12232.

In reality the current ISO standard 12232:2006 specified five ways to measure and report a digital camera's sensitivity:

- One based on signal saturation. (This seems primarily intended for measuring minimum usable exposure index and maximum usable exposure levels, and is related to "base-ISO speed" rather than being suited to callibration of the EI settings on a camera, and is what DxOMark's uses at all exposure index levels.)

- Two, called Snoise40 and Snoise10, that measure low-light handling, and are based on the maximum exposure index levels that give SNR levels of 40:1 and 10:1 respectively under specified conditions. (These are more akin to the ISO definition of film speed, but are rarely used, and seem designed to assess maximum usable exposure index, rather than for calibration of EI settings.)

- Two newer ones added in the 2006 update of 12232, "Standard Output Sensitivity" and "Recommended Exposure Index", which relate to the level placement in output after conversion to an sRGB format like JPEG. (These are the only ISO standards that seem directly intended for use in calibrating the EI settings on cameras, and are what most or all camera makers use for this purpose, in part due to a requirement to do so from the Japanese industry association CIPA, which originated these two standards before their adoption by the ISO.
Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #118 on: November 27, 2012, 06:31:32 AM »
ReplyReply

Let me clarify my comments, in the off chance that you genuinely misunderstood me to be talking about photography done with phones and other truly pocketable cameras: what I intended to convey is that a roughly six or seven stop subject brightness range covers the great majority of the sort of photography of interest to readers of these forums. To be a little more specific, this SBR historically covered the great majority of film photography in any common film format and camera type, even if one only considers photography done with SLR's or larger formats. This is made fairly clear by reading the discussions of film photography capture and printing technique in numerous texts on the subject: tools and techniques like graduated ND filters are there, but not needed for the great majority of cases (and seem particularly needed with color transparency film for which six stops of SBR is optimistic). As a mild extrapolation, It seems from my observations that the move to digital cameras has not greatly changed the preponderance of scenes with "normal" SBR.


If that's what you mean, BJL, I can only presume it's an assumption based upon the fact that the vast majority of photographic scenes have been captured with devices not capable of capturing even 6 or 7 stops of DR, if one includes 35mm slide film and early photographic B&W plates.

It stands to reason, if one is a photographer trying to make a living with a device that captures no more than 6 stops of photographic dynamic range (PDR), that the vast majority of results will not contain more than 6 stops of DR.

The Graduated Neutral Density filter was always a rather inadequate device bcause the line of graduation would rarely even approximately match the significantly varying, wiggly line that often separates a bright sky from a much darker foreground.

Whilst merging multiple images with different exposures to match what the eye sees when one views a sunlit scene, is now much easier in the digital darkroom, it's not something which lends itselft to critically good results without taking the pains of using a tripod when shooting the limited number of scenes which are truly static, and without taking the trouble to make a good job of the tone-mapping, which is just a fancy word for the processing-compression required of any photographic capture to fit the more limited DR of screen or print.

Whilst I would agree that the majority of scenes that  most people actually shoot may not require a camera with a greater PDR than than 6 or 7 stops, I would suggest that's because the majority of shots that people take in general are snapshots of themselves and/or friends standing in front of some interesting or famous artifact, ruin, building, sign etc.

On such occasions, if there happens to be a bright sky in the background, it will be blown. But that's of little consequence since the main interest for them is themselves in association with the famous structure they are partially obscuring.

It seems to me that serious landscape photographers, and let's not forget this is a landscape-orientated site, have always been concerned about the limited dynamic range of their cameras. Didn't Ansel Adams devise the 'Zone System' in an attempt to tackle this issue? Isn't the subject of ETTR one of the most frequently discussed technical issues that's raised again and again, presumably because people sense that their cameras may not be adequate to capture the full PDR of many scenes that they find interesting, unless they maximise exposure without blowing essential highlight detail, and even then perhaps not adequate?

Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8914


« Reply #119 on: November 27, 2012, 08:20:02 AM »
ReplyReply

Not at all: that is a myth, partly encouraged by DxOMark's quirky use and selective quotation of the relevant ISO standard, number 12232.

In reality the current ISO standard 12232:2006 specified five ways to measure and report a digital camera's sensitivity:

- One based on signal saturation. (This seems primarily intended for measuring minimum usable exposure index and maximum usable exposure levels, and is related to "base-ISO speed" rather than being suited to callibration of the EI settings on a camera, and is what DxOMark's uses at all exposure index levels.)

- Two, called Snoise40 and Snoise10, that measure low-light handling, and are based on the maximum exposure index levels that give SNR levels of 40:1 and 10:1 respectively under specified conditions. (These are more akin to the ISO definition of film speed, but are rarely used, and seem designed to assess maximum usable exposure index, rather than for calibration of EI settings.)

- Two newer ones added in the 2006 update of 12232, "Standard Output Sensitivity" and "Recommended Exposure Index", which relate to the level placement in output after conversion to an sRGB format like JPEG. (These are the only ISO standards that seem directly intended for use in calibrating the EI settings on cameras, and are what most or all camera makers use for this purpose, in part due to a requirement to do so from the Japanese industry association CIPA, which originated these two standards before their adoption by the ISO.

BJL,
No-one gives two hoots about these ancient standards of ISO. With modern DSLRs, it's sensor saturation which becomes the standard.

The only issue that is of concern to me is the consistency of DXO's approach and the consistency of their standards. Whether DXO's interpretation of the ISO standard agrees with yours, or anyone elses, is irrelevant. The only issue is the relative ISO sensitivity of the cameras that DXO test. I don't care one whit whether an ISO rating is described as ISO 100 or TWU 254.5. What counts is the sensitivity, in relation to full-well saturation, of one camera compared to another.

I've yet to see any tests which demonstrate that, when DXO claim that camera A requires twice the exposure as camera B, for full well saturation, that this is in fact incorrect.

All those interested in the ETTR concept are interested in what exposure is required for full-well saturation. It's clear that the various ISO definitions of the various manufacturers of cameras differ. According to DXO, most manufacturers exaggerate their ISO sensitivity, which is not surprising. As long as DXO are consistent in their methodology, that's all that's required, because manufacturers are clearly not consistent.
Logged
Pages: « 1 ... 4 5 [6] 7 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad