Pages: [1] 2   Go Down

Author Topic: 1/3 stop ISOs  (Read 17933 times)

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« on: September 09, 2006, 09:42:23 am »

After the Canon 30D came out, I inspected some of its RAW files, and found that the new 1/3-stop ISOs were nothing more than implicit EC-adjustments to the old 100/200/400/800/1600 20D series.  To make a log story short, ISO 125 has as much shadow noise as ISO 640, the extra ISOs all lose 1/3 stop of highlights; half of them lose it completely and the other half shifts it to the shadows.  The results are spiked and gapped histograms.

Someone asked me about the 5D after I posted my conclusions in DPR.  They linked to some 5D files that I inspected, where I found the histograms to be unaltered for the in-between ISOs.  The thread got revived again, and the 5D came up again, and someone linked to some blackframes from the 5D at all ISOs.  Histograms looked the same (although *all* had a tiny stretch with gaps about every 15.5 RAW values), but the blackframe noise, unfortunately, had some very bad news ... 160 is higher than 200, 320 is higher than 400, etc.  The 100/200/... series was the best; the 160/320/... group were the worst.  My conclusion at this time is that the 5d uses analog amplification for the extra ISOs, but it is futher amplifying a subset of the ISO amplifications, to get the rest.

My question is, do any of the cameras out there with "in-between" ISOs that actually have "in-between" noise, or are they all trying to deliver technology that they don't really have?
« Last Edit: September 09, 2006, 09:43:42 am by John Sheehy »
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
1/3 stop ISOs
« Reply #1 on: September 09, 2006, 10:07:22 am »

John,
That's useful information to know. Maybe you could give us all a tutorial on how to use Dcraw and Iris so we can check out these results for ourselves   .
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #2 on: September 09, 2006, 10:22:46 am »

Quote
John,
That's useful information to know. Maybe you could give us all a tutorial on how to use Dcraw and Iris so we can check out these results for ourselves   .
[a href=\"index.php?act=findpost&pid=75918\"][{POST_SNAPBACK}][/a]

This is very easy; just load the blackframe into IRIS ("Load RAW file"), open the command window, and type "stat".  Another window will pop open with the stats.  If you want to see the noise, it centers about the 128 level, so set the thresholds to something like 255 and 0, to see it in low contrast, or set them to something like 158 and 108 to raise the contrast.  The thresholds should stick as you load other RAW files.
Logged

macgyver

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 510
1/3 stop ISOs
« Reply #3 on: September 09, 2006, 12:41:49 pm »

So, basically, shooting that event at ISO 1000 last night was a bad idea?
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
1/3 stop ISOs
« Reply #4 on: September 09, 2006, 07:07:36 pm »

Quote
So, basically, shooting that event at ISO 1000 last night was a bad idea?
[a href=\"index.php?act=findpost&pid=75931\"][{POST_SNAPBACK}][/a]


If Ive understood the point John is making, it was probably a bad idea if you were using a 5D, but not necessarily if you were using a 20D or 30D. When using high ISOs it's generally more important to expose fully to the right for a low noise shot.

With a 30D it would appear there is no difference between a shot at ISO 800, underexposed by 1/3 stop, and the same shot at the same shutter speed and aperture at ISO 1000.

However, if you were using a 5D, noise at ISO 1000 would be marginally worse than noise at ISO 800, using the same shutter speed and aperture.
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #5 on: September 09, 2006, 10:40:12 pm »

Quote
So, basically, shooting that event at ISO 1000 last night was a bad idea?
[a href=\"index.php?act=findpost&pid=75931\"][{POST_SNAPBACK}][/a]

Not especially.  There is no higher ISO with lower blackframe noise.  The 5D blackframe noise "lemons" are 125, 160, 250, 320, and 640.  500 is close, but not as bad.  Of course, this is blackframe noise, which affects mainly the shadows.  Shot noise is dependent purely on the exposure of the sensor; what the aperture and shutter-speed allowed to strike it.  Going up 1/3 or 2/3 stop to an ISO with better blackframe noise, however, won't increase shot noise all that much.

For the Canon 30D, ISO 1000 is a lemon; lotsa noise, and 1/3 stop of RAW highlights clipped away.
Logged

Eric Myrvaagnes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 22814
  • http://myrvaagnes.com
    • http://myrvaagnes.com
1/3 stop ISOs
« Reply #6 on: September 09, 2006, 11:38:31 pm »

Quote
Not especially.  There is no higher ISO with lower blackframe noise.  The 5D blackframe noise "lemons" are 125, 160, 250, 320, and 640.  500 is close, but not as bad.  Of course, this is blackframe noise, which affects mainly the shadows.  Shot noise is dependent purely on the exposure of the sensor; what the aperture and shutter-speed allowed to strike it.  Going up 1/3 or 2/3 stop to an ISO with better blackframe noise, however, won't increase shot noise all that much.

For the Canon 30D, ISO 1000 is a lemon; lotsa noise, and 1/3 stop of RAW highlights clipped away.
[a href=\"index.php?act=findpost&pid=75951\"][{POST_SNAPBACK}][/a]
So for a 5D it sounds as if one should stick with 100, 200, 400, and if necessary 800 and 1600.
Logged
-Eric Myrvaagnes (visit my website: http://myrvaagnes.com)

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #7 on: September 09, 2006, 11:54:26 pm »

Quote
So for a 5D it sounds as if one should stick with 100, 200, 400, and if necessary 800 and 1600.
[a href=\"index.php?act=findpost&pid=75954\"][{POST_SNAPBACK}][/a]

Well, staying away from 800 and 1600 is another issue; I was just talking about inconsistencies in blackframe noise, which is important if you want maximum DR.  Of course, someone who shoots high-key images only, well-exposed, will not likely encounter problems.

I use ISO 800 and 1600 almost exclusively with my 20D.  Rarely gets to see the light of 400.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
1/3 stop ISOs
« Reply #8 on: September 10, 2006, 12:39:11 am »

Quote
The 5D blackframe noise "lemons" are 125, 160, 250, 320, and 640. [a href=\"index.php?act=findpost&pid=75951\"][{POST_SNAPBACK}][/a]

So what you're saying is this downgraded intermediate ISO performance does not extend beyond ISO 640, with the 5D. Is this correct?
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #9 on: September 10, 2006, 07:38:17 am »

Quote
So what you're saying is this downgraded intermediate ISO performance does not extend beyond ISO 640, with the 5D. Is this correct?
[a href=\"index.php?act=findpost&pid=75958\"][{POST_SNAPBACK}][/a]

The natural curve is steep enough at the higher ISOs that the extra amplification doesn't put the noise out of the expected range.

Here are the values:

100 2.0
125 2.4
160 2.9
200 2.0
250 2.5
320 3.0
400 2.2
500 2.7
640 3.3
800 3.0
1000 3.7
1250 4.6
1600 4.45 (the test file was 4.6 overall, but it looked like it had USM moter noise at the top of the frame; the unaffected areas measured 4.45).
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
1/3 stop ISOs
« Reply #10 on: September 10, 2006, 09:46:57 am »

John,
It's interesting that noise at ISO 200 is no greater than at ISO 100. What about dynamic range? Is there any advantage to using ISO 100, outside of a requirement to use a slow shutter speed for blurring effects etc?
Logged

jani

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1624
    • Øyet
1/3 stop ISOs
« Reply #11 on: September 10, 2006, 05:34:59 pm »

Quote
My question is, do any of the cameras out there with "in-between" ISOs that actually have "in-between" noise, or are they all trying to deliver technology that they don't really have?
I can easily gain access to a 1D MkII for testing, and if you can specify the testing methodology, I'll be happy to provide the raw files.
Logged
Jan

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #12 on: September 10, 2006, 10:41:32 pm »

Quote
John,
It's interesting that noise at ISO 200 is no greater than at ISO 100.

That's blackframe noise.  The shot noise is still 41% higher at ISO 200 than 100 for any given RAW level.  However, shot noise is not very strong relative to signal at the lower ISOs, not until you get down into the shadows.  The blackframe noise is stronger than the shot noise for the very deepest shadows.

The fact that they both come out to 2.0 might just be a rounding issue.  OK, I just multiplied both blackframes by 100 first.  Then, the deviations are 195.5 and 202.5.

Quote
What about dynamic range? Is there any advantage to using ISO 100, outside of a requirement to use a slow shutter speed for blurring effects etc?
[a href=\"index.php?act=findpost&pid=75981\"][{POST_SNAPBACK}][/a]


The way I think of dynamic range, there should be little benefit to 100, but there are certain ranges of upper shadows that are going to have more visible shot noise.  IOW, they are visibly equal at the extremes - in the deepest shadows and in the highlights; however, 200 will be a little noisier in the brighter shadows.

Here's an extreme thought experiment; imagine you had a camera with an ISO 100 that had strong blackframe noise, but huge pixels, and an ISO 1600 that had no readout noise.  The 1600 would have higher DR, and cleaner extreme shadows, but the 100 would be better for most of the range.    Of course, if they both had no readout noise, then the 100 would have 2 stops more dynamic range, because it collects 16x as many photons.  That's in theory, of course.  Without readout noise, the 12-bit digitization might be a significant DR bottleneck to ISO 100.
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #13 on: September 10, 2006, 10:43:26 pm »

Quote
I can easily gain access to a 1D MkII for testing, and if you can specify the testing methodology, I'll be happy to provide the raw files.
[a href=\"index.php?act=findpost&pid=76003\"][{POST_SNAPBACK}][/a]

Just a fast (1/250 or so) exposure in each ISO with the lens cap on.  Turn off AF or you may never finish experiment.
Logged

jule

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 738
    • http://www.juliestephenson.net
1/3 stop ISOs
« Reply #14 on: September 11, 2006, 10:04:51 pm »

I recieved some much appreciated and very helpful advice regarding choice of ISO in this thread ; http://luminous-landscape.com/forum/index....406&hl=best+iso , so now,  this thread has brought to my attention another factor in the equation.

Before I reinvent the wheel and stumble my way through the testing proceedures of blackframe noise, is anyone able to offer any advice on which ISO settings are "lemons" or ones to avoid using a 1dsMkII?
Thank you
Julie
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #15 on: September 12, 2006, 10:28:06 am »

Quote
Here's an extreme thought experiment; imagine you had a camera with an ISO 100 that had strong blackframe noise, but huge pixels, and an ISO 1600 that had no readout noise.  The 1600 would have higher DR, and cleaner extreme shadows, but the 100 would be better for most of the range.    Of course, if they both had no readout noise, then the 100 would have 2 stops more dynamic range, because it collects 16x as many photons.  That's in theory, of course.  Without readout noise, the 12-bit digitization might be a significant DR bottleneck to ISO 100.
[a href=\"index.php?act=findpost&pid=76024\"][{POST_SNAPBACK}][/a]

Here's an idealized chart of what you could expect with a camera with 81920-electron wells at ISOs 100 and 1600, with (5.0 std dev) and without readout noise:



This would be considering readout noise and shot noise.  I don't know what would really happen with the two lines heading southeast; quantization may make them level off, too, and I don't know exactly how shot noise works with less than 1 sample per well on average; then again it may work by the square root principal anyway.
Logged

Nill Toulme

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 738
    • http://www.toulmephoto.com
1/3 stop ISOs
« Reply #16 on: September 12, 2006, 10:57:12 am »

Here's a very basic question — What's different about the way the camera produces the full stops and 1/3 stop intermediates that produces such different results.  Why doesn't it use the same technique for both?

Nill
~~
www.toulme.net
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #17 on: September 12, 2006, 03:20:48 pm »

Quote
Here's a very basic question — What's different about the way the camera produces the full stops and 1/3 stop intermediates that produces such different results.  Why doesn't it use the same technique for both?
[a href=\"index.php?act=findpost&pid=76112\"][{POST_SNAPBACK}][/a]
The main ISOs probably have a more optimized amplification system, and the extra ones achieved by amplifying them, including their noise.
« Last Edit: September 12, 2006, 03:21:23 pm by John Sheehy »
Logged

Nill Toulme

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 738
    • http://www.toulmephoto.com
1/3 stop ISOs
« Reply #18 on: September 12, 2006, 03:54:27 pm »

Do you know if anyone has looked at the 1-series cameras on this score?

Nill
~~
www.toulme.net
Logged

John Sheehy

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 838
1/3 stop ISOs
« Reply #19 on: September 12, 2006, 08:07:34 pm »

Quote
Do you know if anyone has looked at the 1-series cameras on this score?
[{POST_SNAPBACK}][/a]

I gave the instructions for performing the test near the beginning of this thread.  Anyone with a PC can download [a href=\"http://www.astrosurf.com/buil]IRIS[/url] and test their camera.  Just make sure the exposures are all equally short.

If anyone wants to test with a Nikon, or most of the other digitals, blackframe noise is going to be on a different scale than with Canon, because Canon does not clip their RAW data to black, so Canon has both positive and negative noise in their blackframes (black is generally 128 in the RAW data).  Most other cameras have black as 0 in the RAW data, so all of the negative noise is clipped to 0.  If anyone ever starts making a comparative list of these noises between cameras, the Canon noise will need to be clipped, too, for proper comparison.
Logged
Pages: [1] 2   Go Up