Luminous Landscape Forum
Equipment & Techniques => Cameras, Lenses and Shooting gear => Topic started by: bclaff on March 08, 2012, 10:50:44 pm
-
I have just processed the Imaging Resource CR2 files for the EOS 5D Mark III.
Most people find my Photographic Dynamic Range Chart (http://home.comcast.net/~NikonD70/Charts/PDR.htm) the best overview.
A good cross section for comparing cameras is this bar chart (http://home.comcast.net/~NikonD70/Charts/PDR_65.htm).
More technical information would include read noise (http://home.comcast.net/~NikonD70/Charts/RN_ADU.htm).
And a restatement of PDR that shows how shadow DR improves with ISO (http://home.comcast.net/~NikonD70/Charts/PDR_Shadow.htm).
You can see that the 5D Mark III compares favorably with the top Nikon models in PDR.
:)
Bill
-
Thank you for taking the time to do this.
-
Thanks for doing this - looks interesting.
It appears that the DR from the Mk II is pretty much the same as the Mk III from ISO 50 - 200, and only half a stop worse at ISO 25600.
Given that the Mk II is now 4 years old, I would have expected to see some sort of improvement in terms of DR at low ISO from the Mk III.
That it still trails the 3.5yr old D3x by almost 1.5 stops at ISO 50 is curious as well - just goes to show how good the D3x is I guess.
-
Thanks Bill
The DR from ISO 50-200 looks rather disapointing. I presume the production model will be similar.
I would thought Canon would have put more effort into this area as it was well known strength of the Nikon/Sony technology.
So DR not that great but I think noise wise from what I seen so far they have largely caught up.
-
I have just processed the Imaging Resource CR2 files for the EOS 5D Mark III
You can see that the 5D Mark III compares favorably with the top Nikon models in PDR.
:)
Bill
Maybe I'm reading the data wrong but at base ISO I see the 5DII and III the same and the D5100 (same sensor as my K5) over 1 stop more DR, is this correct?
Marc
-
I'm really not well versed in the Canon product line; but my take is that the 5D Mark II was quite good for it's time and that the 5D Mark III, although not much improved, is quite competitive today.
-
Marc,
You're reading that right.
At ISO 100 the D5100 (and D7000) are very good low ISO performers.
Regards,
Bill
-
You're reading that right.
At ISO 100 the D5100 (and D7000) are very good low ISO performers.
These topics are often discussed in forums in terms of technical ability of the engineering teams, and that may be playing some role, but my take is that cameras most often perform exactly the way the company selling them wants them to perform. In other words... what are the important parameters required to achieve the business goals that have been set for a given model?
The product planning is a tough job for camera manufacturers...
- how to prioritize, among others, the 3 following variables: resolution, low ISO DR, high ISO image quality,
- how to define the target that needs to be achieved for these metrics relative to both the want of the photographers and their actual needs
- how to please experts and the general public
- video vs still
- ...
It seems that, 6 years ago when specing the 5DII, the product planners at Canon were focusing on resolution, high ISO noise and video. It can be argued that the 5DII reached a very good balance at the time being and that its spec were a good match for both the wants and the actual needs of a large set of photographers... except landscape shooters for whom DR was lacking.
It seems that, 3 years ago when specing the 5DIII, the product planners at Canon decided that their main problem was Nikon's superior high ISO image quality and that their main differentiator was video. It seems that they didn't acknowledge the relevance of low ISO DR.
Now, is there real value in going farther than the D700/D3 in terms of low light image quality? I am sure sports shooters would answer yes... I am not sure wedding guys really need anything better although they may still want it. But I am sure pretty that landscape guys need more DR...
Cheers,
Bernard
-
I believe Japenese company's try too hard avoiding royalties ie Burr Brown's superior DAC's so I believe Canon is resisting using a Sony sensor or licensing the rights to use onboard ADC's
At any rate it just cost them a 5DIII sale I placed an order for a 800E
Marc
-
I have just processed the Imaging Resource CR2 files for the EOS 5D Mark III.
You can see that the 5D Mark III compares favorably with the top Nikon models in PDR.
Bill
What this sort of analysis based on noise std dev of pixels doesn't take into account is the correlated (pattern) noise of Canon models such as the 5D2 and it seems also the 5D3 (unfortunately), which limits the latitude in processing low ISO shadows much more than the PDR figure might indicate. Recent Nikon offerings, especially those with Sony Exmor sensors, far outstrip the corresponding Canon models in this regard. I would not say that the 5D3 compares favorably with the top Nikon models in useable dynamic range; I wish it were otherwise.
-
Emil,
Very interesting observation.
It's too bad pattern noise is so hard to extract from ordinary photographs.
With proper test targets how do you think it would be best to quantify it?
Best Regards,
Bill
-
Emil,
Very interesting observation.
It's too bad pattern noise is so hard to extract from ordinary photographs.
With proper test targets how do you think it would be best to quantify it?
Best Regards,
Bill
Each row or column has an offset and a gain, so that the pixel value is (offset + lightlevel*gain).
Let's assume that the pattern noise is uniform across rows/columns. In such a case, I would take a blackframe and column (row) average it to determine the std dev of the column (row) offset. I would use a flat field and a similar procedure to determine the gain fluctuation (there one would have to be careful to exclude the corners and take a running average over windows, and use a moderate aperture, to keep vignetting etc from contaminating the data).
More generally, I would take a representative block from the center of the image and Fourier transform it; then compare the generic fluctuation of the Fourier data off axis to the data along the horizontal and vertical wavenumber axes which give the spectrum of the vertical and horizontal pattern noise, respectively. One wants to see how the typical on-axis Fourier coefficients differ from the off-axis ones, and this gives a measure of the pattern noise (again using the blackframe for the offset noise and the flat field for the gain variation). For the blackframe, there will typically be spikes in the Fourier data whose spacing indicates the number of readout channels (eg the number of spikes is the number of readout channels; it's quite obvious on Canons). The spikes are a fixed, repeatable pattern noise whereas the rest of the on-axis data gives the random component of the pattern noise.
-
Very interesting thread. Looking at the graphs, after the updated D800 chart, it seams like the 5DMKIII is far behind the D800 in terms of dynamic range.
But isn't that only half of the story? I found this on www.cambridgeincolour.com
"Most digital cameras use a 10 to 14-bit A/D converter, and so their theoretical maximum dynamic range is 10-14 stops. However, this high bit depth only helps minimize image posterization since total dynamic range is usually limited by noise levels. Similar to how a high bit depth image does not necessarily mean that image contains more colors, if a digital camera has a high precision A/D converter it does not necessarily mean it can record a greater dynamic range. In practice, the dynamic range of a digital camera does not even approach the A/D converter's theoretical maximum; 5-9 stops is generally all one can expect from the camera."
I am not so much into the technical details, however I do use a NEX 7 now for some time, and there is quite some noise in the shadows. Especially with longer exposures during low light conditions. So doesn't that mean, that even though the D800 might have a higher dynamic range, the 5DMKIII might still produce equal quality images due to lower noise levels?
-
Hi,
My understanding is that readout noise may not mostly come from ADC-s, older Nikon did use Burr Brown ADCs but still had noise characteristics similar to Canon. It seems that Sony's technology with on chip ADCs makes a difference.
I hope you will be happy with your Nikon, please share the experience.
Best regards
Erik
I believe Japenese company's try too hard avoiding royalties ie Burr Brown's superior DAC's so I believe Canon is resisting using a Sony sensor or licensing the rights to use onboard ADC's
At any rate it just cost them a 5DIII sale I placed an order for a 800E
Marc
-
I think it is very important to include pattern noise in some way in the "photographic dynamic range", or else cameras with high pattern noise (like the 5D mark 2) will get higher photagraphic dynamic range than they deserve. I've seen no measurement yet that has done that, engineering DR measurements don't say that much, only that those that win by a huge marigin is probablygreat in other aspects, but those with say 11-12 stops engineering DR can differ a lot in the subjective photographic DR due to pattern noise. Possibly there is a difference in how well colors are reproduced in shadowy areas despite similar noise levels(?), and if so that will also have an impact on photographic DR. Blotchiness of chroma noise is another aspect, but perhaps that is fairly constant regardless of sensor?
I'm not 100% sure, but it does seem to me that the pattern noise of 5D mark 3 if not gone is at least considerably less than 5D mark 2 at base ISO and therefore the photographic DR should be considerably better. Surely not as good as Sony Exmor, but for us that are used to 5Dmk2 and heavily invested in Canon glass it can be a worthwhile improvement still.
A test that I would like to see is mk2 and mk3 shooting the exact same test scene at ISO100 perfectly ETTR (I mean verified in the raw file!), and then push a shadow area (preferably containing a color checker) 3 stops and show the crops side by side, resolution is almost the same so no scaling is required. I suspect that the mk3 will fair much better in such a test than these measurements indicate.
(For high ISO the tests need to be a bit different. One needs to take into account how sensitive the sensor is, because then it is about making as good image as possible with little light, and ISO1600 of one camera is not necessarily the same sensitivity as ISO1600 of another. At base ISO for tripod-mounted landscape the quantum efficiency is not at all as important)
-
Maybe I don't quite understand the data, but my very preliminary findings seem to indicate the opposite.
- Comparing the files from IR, ISO 200 vs ISO 102400
- Reading the linear data using DCRaw, internal version 9.12
- Redistributing the data as follows:
log2 ( (1 + src - min) / (65536 - min) )
I get the histograms as attached:
5D3L = 200
5D3H = 102400
The vertical bars indicate F-stops.
You can also see a representation of the RAW data partially, and what is particularly annoying is the striping in the noise. That makes RAW conversion significantly more problematic. I don't want to jump to hyperbole conclusions too quickly, but from this very preliminary information I have to consider the high-iso capability of this camera nothing other than a gimmick…
-
Each row or column has an offset and a gain, so that the pixel value is (offset + lightlevel*gain).
...
Hi Emil,
If one had a good averaged (say 100 frame average) dark frame along with a good averaged flat field frame (again 100 or so) for the particular camera parameters used (F#, etc.), couldn't one in theory eliminate this pattern noise?
Also, I've been wondering if Nikon does some of this already?, i.e. by doing their black clipping?
Chris
-
...
You can also see a representation of the RAW data partially, and what is particularly annoying is the striping in the noise. That makes RAW conversion significantly more problematic. I don't want to jump to hyperbole conclusions too quickly, but from this very preliminary information I have to consider the high-iso capability of this camera nothing other than a gimmick
The back order queue at B&H is rumored to be in excess of 10,000 for the 5D3 so obviously gimmicks sell...
-
Hi Emil,
If one had a good averaged (say 100 frame average) dark frame along with a good averaged flat field frame (again 100 or so) for the particular camera parameters used (F#, etc.), couldn't one in theory eliminate this pattern noise?
Also, I've been wondering if Nikon does some of this already?, i.e. by doing their black clipping?
Chris
Only if it was fixed pattern noise (which for instance some of the offset banding is -- the part that is regular with the periodicity of the number of readout channels). Unfortunately a lot of the banding noise in Canons is random. I know less about Nikons but I suspect it is similar, at least for non-Exmor sensors.
-
The back order queue at B&H is rumored to be in excess of 10,000 for the 5D3 so obviously gimmicks sell...
:D LOL, must be true, because I doubt they all had a change to properly handle the camera no? :P
Anyway, the camera is likely a stellar performer, but I just don't know whether the highest iso settings are going to make the discerning difference. Don't you think these are only useful when the final image is published in a significantly reduced size? And I also wonder when people are actually using these iso's?
-
Hi,
My understanding is that readout noise may not mostly come from ADC-s, older Nikon did use Burr Brown ADCs but still had noise characteristics similar to Canon. It seems that Sony's technology with on chip ADCs makes a difference.
I hope you will be happy with your Nikon, please share the experience.
Best regards
Erik
Erik
I will be glad to share my experience with the 800E
BTW the reference to BB DAC's was from my audio background
The BB PCM63 is the best audio dac ever produced
Ultra Analog cascaded 2 of them on a single die producing the best DAC module
The Classe DAC 1 uses these Ultra Analog modules and is the best high end audio DAC ever produced
The Japanese companies developed inferior delta sigma DAC's to avoid royalties to BB
Since the PCM63 digital audio has steadily declined because of the emphasis on video and mp3's
Sad, oh I forgot this is a photography forum never mind :)
Marc