April 19, 2014, 03:54:52 AM

Author Topic: 5D3 same max dynamic range as the 5D2???  (Read 25487 times)

dilbert

  • 1D X
  • *******
  • Posts: 2370
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #105 on: March 10, 2012, 04:29:34 PM »
The 5D3 and D800 are pretty much comparable, but only at ISO800 and above.. Below ISO 800, the D800's DR is heading sky high, setting a new record.

Bill Claff, whose site this is from, is a long-time fervent Nikon guy; and these graphs are based on estimated values for both the D800 and the 5D Mk III.

Big pinch of salt in order here...

As long as the methodology used is the same across all cameras then it is valid to compare cameras with.

The numbers in his graph don't agree with DxO (for example) but the disagreement is roughly the same whether the camera is Canon or Nikon.

canon rumors FORUM

Re: 5D3 same max dynamic range as the 5D2???
« Reply #105 on: March 10, 2012, 04:29:34 PM »

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #106 on: March 10, 2012, 05:18:45 PM »
Quote
part of the read out noise stage that doesn't come into play much at high ISO is not so hot so it shows up a lot at low ISO and you get diminishing returns compared to what you'd expect looking at the high iso numbers as you go to near base ISO since you get worse read performance at lower ISOs

LTRLI: Does the amplification based on the ISO setting occur before charge read out (before noise injection due to read)?

DavidRiesenberg

  • 7D
  • *****
  • Posts: 332
    • View Profile
    • David Riesenberg
Re: 5D3 same max dynamic range as the 5D2???
« Reply #107 on: March 10, 2012, 05:22:20 PM »

actually the linear plot for DR of that type is the expected if you had a perfect read from the sensor at a certain stage of the readout, so the D800 is not applying any sort of weird curves (nor is the Canon, it curves over, which is NOT the expected, but it is not because of weird tone curves either but because part of the read out noise stage that doesn't come into play much at high ISO is not so hot so it shows up a lot at low ISO and you get diminishing returns compared to what you'd expect looking at the high iso numbers as you go to near base ISO since you get worse read performance at lower ISOs)

As far as I remember, I never saw a sensor that exhibits a kink in the curve like the D800 one. It doesn't happen with the D7000 or the Phase One (wide DR examples) at DXO. That's why in my technical-limited-knowledge logic it seems to me that Nikon felt that below ISO 800 the noise is low enough to justify some post capture adjustments.

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #108 on: March 10, 2012, 05:45:36 PM »
Quote
I think they say records and captures because the read electronics can mess up what the sensor itself captured. I believe it has been shown that the 1D4 sensor, for instance, actually captures close to 14 stops but then the readout loses a few stops and that is what gets recorded in the RAW file and it's only 11.5 stops instead of almost 14.

I'm not sure I'm explaining myself properly. My concern is that by just looking at the RAW file, are you really measuring the number of stops recorded in the actual scene that was being shot? Seems to me that to be able to do that you'd have to have a wide range of subjects with different luminosities (i.e. the multiple ND filter set up that DXO says it uses), measure those luminosities using a reliable spot meter, then look at the RAW data and see the range of luminosities for which the camera is able to:

  • Give a reasonable SNR for the darkest luminosity recorded
  • Differentiate one luminosity vs. the next on the higher (brightest) end

By just looking at the RAW file, with no actual scene information, the only way the test would be valid would be if there was a directly linear relationship between the scene luminosity & the luminosity recorded in the RAW. Which your statement about the 14 stops going to 11.5 itself seems to show isn't the case.

LetTheRightLensIn

  • 1D X
  • *******
  • Posts: 2987
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #109 on: March 10, 2012, 06:58:18 PM »

actually the linear plot for DR of that type is the expected if you had a perfect read from the sensor at a certain stage of the readout, so the D800 is not applying any sort of weird curves (nor is the Canon, it curves over, which is NOT the expected, but it is not because of weird tone curves either but because part of the read out noise stage that doesn't come into play much at high ISO is not so hot so it shows up a lot at low ISO and you get diminishing returns compared to what you'd expect looking at the high iso numbers as you go to near base ISO since you get worse read performance at lower ISOs)

As far as I remember, I never saw a sensor that exhibits a kink in the curve like the D800 one. It doesn't happen with the D7000 or the Phase One (wide DR examples) at DXO. That's why in my technical-limited-knowledge logic it seems to me that Nikon felt that below ISO 800 the noise is low enough to justify some post capture adjustments.

Oh that kink, I was ignoring that, I thought you meant something else. I think the said that he made a small mistake with the ISO400 reading on the camera, that would explain some of it. Or maybe in how the ISO got rated, once you go above base ISO and compare things it's a little trickier because of that. I doubt it's anything in the camera itself.

The 7D does get a bit of a kink in the SNR charts because it has an unusual form of gain, as opposed to offset, vertical banding that can add a little to the SNR and it has more influence at a certain point along the ISO scale, I think.

LetTheRightLensIn

  • 1D X
  • *******
  • Posts: 2987
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #110 on: March 10, 2012, 07:00:11 PM »
Quote
I think they say records and captures because the read electronics can mess up what the sensor itself captured. I believe it has been shown that the 1D4 sensor, for instance, actually captures close to 14 stops but then the readout loses a few stops and that is what gets recorded in the RAW file and it's only 11.5 stops instead of almost 14.

I'm not sure I'm explaining myself properly. My concern is that by just looking at the RAW file, are you really measuring the number of stops recorded in the actual scene that was being shot? Seems to me that to be able to do that you'd have to have a wide range of subjects with different luminosities (i.e. the multiple ND filter set up that DXO says it uses), measure those luminosities using a reliable spot meter, then look at the RAW data and see the range of luminosities for which the camera is able to:

  • Give a reasonable SNR for the darkest luminosity recorded
  • Differentiate one luminosity vs. the next on the higher (brightest) end

By just looking at the RAW file, with no actual scene information, the only way the test would be valid would be if there was a directly linear relationship between the scene luminosity & the luminosity recorded in the RAW. Which your statement about the 14 stops going to 11.5 itself seems to show isn't the case.

Unlike film, the digital sensors have linear capture from the scene.
The 14 to 11.5 is because read noise makes the bottom few stops meaningless mess of randomness.

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #111 on: March 10, 2012, 07:25:21 PM »
Quote
It means that SNR would be better up in the realm where photon noise dominated read noise. But it does relatively little, in comparison, for the performance in the deep shadows compared to the read noise. So it can be that they have the same SNR in the deepest shadows near black but that one has better SNR say at semi-dark gray and brighter.


Ah, all you're saying is that the absolute difference in SNR will be less for low signals vs. high, as is clear in this simple linear plot here of the SNRs for QE 1 vs QE of 0.5, assuming a read noise of 6e-:



This always begged the question in my mind: if you increase QE dramatically for a sensor without increasing full well capacity, don't you run the risk of blowing out the sensor for any given f-ratio & exposure combination compared to an older sensor with lower QE & same full well capacity? I realize that the increased QE sensor would have a lower gain for any f-ratio/exposure combination, but blown out is blown out... this always confused me.

Thanks.

canon rumors FORUM

Re: 5D3 same max dynamic range as the 5D2???
« Reply #111 on: March 10, 2012, 07:25:21 PM »

neuroanatomist

  • CR GEEK
  • *******
  • Posts: 12779
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #112 on: March 10, 2012, 08:24:45 PM »
This always begged the question in my mind: if you increase QE dramatically for a sensor without increasing full well capacity, don't you run the risk of blowing out the sensor for any given f-ratio & exposure combination compared to an older sensor with lower QE & same full well capacity? I realize that the increased QE sensor would have a lower gain for any f-ratio/exposure combination, but blown out is blown out... this always confused me.

Sure.  'Full well capacity' is measured in electrons, after conversion from photons. If QE goes up, you get more electrons per photons input, and the well fills faster.  It's probably not a practical problem, though, since in most normal shooting situations a lower gain (ISO) would solve the problem. 
EOS 1D X, EOS M, and lots of lenses
______________________________
Flickr | TDP Profile/Gear List

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #113 on: March 10, 2012, 08:47:23 PM »
neuroanatomist, or anyone: Though I asked this previously, I think it got lost in all the back-and-forth -- is 'gain' at the pixel-level applied during charge reading? I.e. does the ISO setting determine the 'conversion gain' (┬ÁV/e-)? Meaning, if you have a very low signal, is increasing this gain advantageous b/c it will help offset the read noise to be injected & b/c of the limited precision of reading?

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #114 on: March 10, 2012, 11:36:15 PM »
jrista:
Quote
Do you mean "charge reading" or "conversion of charge to a digital unit"?

Not sure what happened to your post, but I'm answering anyway... I mean during 'charge reading'. If it's just during 'conversion of charge to digital unit', then LTRLI's statement "part of the read out noise stage that doesn't come into play much at high ISO is not so hot so it shows up a lot at low ISO and you get diminishing returns compared to what you'd expect looking at the high iso numbers as you go to near base ISO since you get worse read performance at lower ISOs" wouldn't make much sense (if I understood what he was trying to say correctly).

I may be totally wrong.

But I'm basing my notion from this comment in the chapter on CMOS Image Sensors in "Image Sensors & Signal Processing for Digital Still Cameras" (assuming these are active, not passive, pixel sensors): "The active pixel concept in which a photogenerated charge is amplified in a pixel and the amplified signal is read out has its roots in the phototransistor array image sensor. One advantage of the active pixel is its suppression of noise generated and/or injected in the signal readout path..." (pg. 145).

Then again, no idea if this 'amplification' is the one you can set by ISO... someone more knowledgeable will have to chime in.

jrista

  • 1D X
  • *******
  • Posts: 3239
  • POTATO
    • View Profile
    • Nature Photography
Re: 5D3 same max dynamic range as the 5D2???
« Reply #115 on: March 11, 2012, 01:53:48 AM »
Not really sure what happened to my post either. ISO, at least with Canon cameras, is a pretty ugly beast when you really dig into it (its not really as simple as I stated before.) Its not handled the same way at all ISO settings, and there are big changes once you get to really high ISO settings. Its not really actually handled just at one point, it can be handled at multiple points along the read and processing pipeline.

Generally speaking, a specific ISO setting is achieved via per-pixel amplifiers, the ADC, as well as post-ADC digital "gain", which is pretty much the same as adjusting exposure during post processing. In the case of pre-1DX/5DIII sensors, Canon also used a full-stop native/third-stop push\pull approach. Analog gain (which could be via amplifier or amplifier and ADC) ISO only affected full-stop ISO settings: 100, 200, 400, 800, 1600. Beyond ISO 1600, you start seeing both analog and digital gain. When it came to third-stop settings, full-stop ISO gain would be applied, followed by a 1/3rd stop push or pull exposure adjustment. ISO 125 would be ISO 100 with digital "gain" of +1/3rd stop applied. ISO 160 would be ISO 200 with digital "gain" of -1/3rd stop applied. Things get really complex when it comes to higher ISO settings. For example, ISO 5000 would be:

+4 stops "clean" analog gain (pixel amplifier, I believe...never found any specific details)
+1 stops "dirty" analog gain (modified ADC gain, I believe...never found any specific details)
-1/3 stops "nasty" digital "gain" (reduction of EV of post-ADC digital image)
+1 stops "pretend" "gain" (setting in metadata that tells RAW processor that more digital exposure increase is necessary)

So, the initial four stops of clean analog gain by the pixel amplifier would conform to what you've read in your book. Modern CMOS sensors usually include nose reduction circuitry to make sure that the direct pixel read is as clean as possible (usually CDS, or correlated double sampling, to limit dark current noise...used in both Canon and Sony sensors; sometimes additional measures are taken to minimize fixed pattern noise caused by deviations in the performance of each pixels transistors relative to each other, etc.), so that when the pixer amplifier jacks up ISO according to the camera setting, its jacking up an electron count read as close to the real photon conversion count as possible. That would only be the case with full-stop ISO settings up to ISO 1600 on a Canon sensor, however.

No one can say yet how Canon's new sensors handle ISO settings. One of the things I personally really want to know is whether analog gain, either at the pixel or via ADC or something else, is done for every ISO setting. The whole push/pull third-stop approach is rather annoying, as it always chops off/shifts at lest 1/3rd stop DR at those ISO settings. It also causes a very funky progression of noise levels. Full-stop noise is "normal", push stop noise is "excessive" (often worse than full-stop ISO two to three settings higher), and pull stop noise is "clean". ISO settings beyond 1600 (which, in the 1D X and 5D III, might actually be ISO 6400 or ISO 12800 now) pretty much get a double-whammy, with two kinds of analog gain as well as two kinds of digital exposure amplification, usually leading to a large jump in noise levels that really eat away at fine detail. It does sound like the new 1D X and 5D III sensors have incorporated new per-pixel pre-amp hardware noise reduction. Not sure exactly what kind of noise it reduces...but given that there appears to be practically no fixed patter noise in 5D III sample images, and at worst minor vertical banding noise...I'd venture a guess that it eliminates fixed pattern noise.

Clean analog gain (which I believe is best achieved via per-pixel amplifiers after fixed-pattern and dark current noise are mitigated during photodiode read) is how all ISO settings should be achieved in my opinion. I believe that is indeed how Sony sensors work...at least up to ISO 1600 on current camera models (I think they also do additional ISO adjustment in the ADC at higher ISO's.)
« Last Edit: March 11, 2012, 03:00:09 AM by jrista »
My Photography
Current Gear: Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: Canon 5D III/7D II | Canon EF 300mm f/2.8 L II

KeithR

  • Guest
Re: 5D3 same max dynamic range as the 5D2???
« Reply #116 on: March 11, 2012, 06:25:26 AM »
As long as the methodology used is the same across all cameras then it is valid to compare cameras with.
Oh, I'm sure they'll be OK, but as I say, some caution is in order.

Must be honest though - I struggle to understand where Nikon/Sony can have got the thick end of two stops from.
« Last Edit: March 11, 2012, 06:27:47 AM by KeithR »

LetTheRightLensIn

  • 1D X
  • *******
  • Posts: 2987
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #117 on: March 11, 2012, 02:34:23 PM »
gotta get back to the game and then eagles but quickly, based on someone's black frame posts here (another thread):
disappointing results at ISO 100:

banding:
horizontal banding appears to be entirely removed, this is good, and will help make ISO look better and get less badly clumped chroma noise and all but vertical banding appears to be zero improved and it shows up enough that I don't think you gain any usable amount of low ISO DR back compared to the 5D2

so for low ISO I'm not sure the removal of horizontal banding alone will really help usable DR any since the vertical sticks out to the eye just as much, maybe here on there on parts of the image it might help (it ay be enough to help high iso look better though and make a difference)

DR:
5D2: 15760,1024,6.1 AU,3.8 8MP normalized ADU, DR - 11.2  , DR8MP - 11.9

5D3: 15309 (maybe the channels were still not quite blown though?),2048, 6.6 ADU, 6.4 normalized to 5D2 6.4 and then to 8MP 3.96, DR- 11.0  DRto5D2 - 11.0  DR8MP - 11.7
and let us say WP should be 15760 instead then DR8MP - 11.8
and let us even say it should be max 14bit 16383 - 11.8

so I actually get fractionally worse DR than for the 5D2, but it's for all intents, the exact same, within copy to copy variation and well within any difference you'd be able to notice

Sad to say but even the DX D7000 handily beats the latest Canon FF for DR, and not by a little. The D800, at 100% view, full 50% MP advantage may beat it by 2 stops, maybe even 2.5 usable stops.

I did not expect this at all.  :'(

I think we just have to accept that it will do nothing for the ISO 100-400 non-action shooters. It's basically a big step up in AF and possibly some degree better image quality at high ISO (hopefully enough to matter, it looks like it might be, but can't be sure yet) only. Not that those are minor things, but still, not sure how that makes it $500 more than the 6D800. I am now really, really surprised they decided to price it above the D800.
« Last Edit: March 11, 2012, 04:42:15 PM by LetTheRightLensIn »

canon rumors FORUM

Re: 5D3 same max dynamic range as the 5D2???
« Reply #117 on: March 11, 2012, 02:34:23 PM »

Abraxx

  • Canon AE-1
  • ***
  • Posts: 69
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #118 on: March 11, 2012, 03:46:50 PM »
Oh, what I feared, no real DR improvement.
If further tests confirm this, I'm not going to buy the 5D Mark III, especially not at the extreme price point Canon has chosen here in Germany.
DR improvement was crucial for me. I'm really disappointed if this turns out to be true. :'(
'Photographers are the eyes of the world'
Focus on Landscape/Nature, Architecture, Technology, Wildlife, Street

sarangiman

  • EOS M2
  • ****
  • Posts: 211
    • View Profile
Re: 5D3 same max dynamic range as the 5D2???
« Reply #119 on: March 11, 2012, 06:53:30 PM »
I forget who posted this .CR2 file (I think it was either here or on dpreview), so, sorry I'm missing credit for it, but, vertical banding is pretty poor. This is +100 Fill Light, ISO 100 (it's a crop, not full image):



That just looks like FPN that shoulda been removed in internal camera processing...

That's really going to limit usability of shadows in high dynamic range landscapes (& yes, I do shoot w/ Singh-Ray filters... but sometimes it's either not practical b/c of you're, say, shooting a cityscape without a simple horizon, or when even those filters aren't enough... e.g. when shooting long telephoto where effects of grad ND filters are minimal or, say, when shooting w/ the moon in the frame).

canon rumors FORUM

Re: 5D3 same max dynamic range as the 5D2???
« Reply #119 on: March 11, 2012, 06:53:30 PM »