5D3: ISO 160, 320, 640 etc cleanest ISOs?

[LetTheRightLensIn]

Important thing to remember about the 160, 320 etc

"Because ISO 320 is actually ISO 400 pulled 1/3 of a stop, that means that the highlights are going to clip at exactly the same point as they would at ISO 400. The 1/3 stop pull is just making that point 1/3 stop darker than pure white. The entire image at ISO 320 is 1/3 stop darker (and may be less noisy) than the image at ISO 400, so the blacks lose detail 1/3 stop sooner, but you don't get that 1/3 stop back at the highlight end of the range -- it's still gone. Therefore, at ISO 320 you're losing a net 1/3 stop from the total usable dynamic range that you would have if you were shooting at ISO 400."

http://shootintheshot.joshsilfen.com/2010/05/13/canon-hd-dslr-native-iso/

but they do have less noise..

That is wrong, with ISO320 you gain 1/3 stop in the darkest shadows and lose 1/3 in highlights and get 1/3 better SNR across the board since it gets exposed to light for 1/3 stop longer than ISO400. ISO320 and 400 have the exact same dynamic range and they are equivalent, if you use EC +1/3 ISO400 it is the same as using ISO320 and if you use ISO320 EC -1/3 it is the same as using ISO400 (the file contents are stored a touch differently in what range of numbers they use, but the true end result is the same). That is for RAW.

For jpg and video it might be different since they chop off some of the DR that RAW has and depending upon what tone curve and how they chop things it's possible ISO320 and 400 might not be the same and that using EC +/-1/3 might not be able to make them the same. I never looked into it too much myself. Some claim the 160,320, etc. are better for jpg/video since they claim the top 1/3 of highlights that ISO160/320 chop off get chopped at ISO200 and 400 so you don't lose that relatively speaking but you do gain 1/3 in shadows, which may seem even larger since you might rise above the worst of banding, and SNR, not sure if that is true, but some have said that. For RAW shooting that is definitely not the case though.

 

[THX723]

Referring to Mr. Claff's Photographic Dynamic Range chart:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%201D%20Mark%20IV
(note: select EOS 5D Mk III)

The analysis methodology appears sound imho, however regardless of the absolute accuracy, it is the delta that is of interest here and with that, it clearly supports the multiples of 160 theory (at least for the 5D Mk3).

Put another way in terms of read noise:
http://home.comcast.net/~NikonD70/Charts/RN_ADU.htm

again clearly shows iso160, 320, 640, etc. having the least noise.

 

[jrista]

All Canon cameras most definitely have NATIVE ISO 100, 200, 400, 800, etc. stops. The reason noise is cleaner on the -1/3rd stop ISO settings is because Canon uses a Base/Push/Pull approach to ISO. As an example, ISO 320 is actually ISO 400 with a -1/3rd stop "pull" that underexposes the ISO 400 shot by one third of a stop, then "recovers" the underexposure with a bit of extra analog gain. That "pushes down" noise at ISO 320, which is why it looks cleaner...but, its at the cost of an additional third-stop of dynamic range.

You can find a complete analysis of Canon's method of achieving the various ISO settings at the following link:

http://photography-on-the.net/forum/showthread.php?t=1081982

Core ISO levels up to ISO 1600 (which, in the case of the 5D III, might now be ISO 12800) are a base ISO setting (initial amplification) with a push or pull of 1/3rd stop of analog gain. Beyond the maximum core ISO setting, things tend to get a bit nastier, and you might have initial amplification, analog gain, digital gain, and metadata settings that tell the RAW processor to boost even more in post.

Personally, one of the things I was really hoping Canon would improve on with the 5D III and 1D X was their approach to ISO. Its annoying to alternatively lose a third stop of DR for "non-native" ISO settings. It would be much nicer to simply have electronic amplification right off the pixel directly to the appropriate ISO. Thats what Sony Exmor sensors do, and they seem to do a pretty good job of it until higher ISO settings. In the grand scheme of things, though, it doesn't really matter a wit to IQ...having shot Canon cameras for a number of years now, lower-ISO noise is largely indistinguishable across settings when you properly expose. It may only manifest as a minor problem if you need multiple low-light shots from a sequence that involved differing lower-ISO settings to be very consistent...in which case you might notice slight differences in noise characteristics between frames. On an individual shot basis were no consistency is necessary, the only real drawback is that 1/3rd stop DR loss for non-native (100, 200, 400, etc.) settings.

 

[LetTheRightLensIn]

All Canon cameras most definitely have NATIVE ISO 100, 200, 400, 800, etc. stops. The reason noise is cleaner on the -1/3rd stop ISO settings is because Canon uses a Base/Push/Pull approach to ISO. As an example, ISO 320 is actually ISO 400 with a -1/3rd stop "pull" that underexposes the ISO 400 shot by one third of a stop, then "recovers" the underexposure with a bit of extra analog gain. That "pushes down" noise at ISO 320, which is why it looks cleaner...but, its at the cost of an additional third-stop of dynamic range.

You can find a complete analysis of Canon's method of achieving the various ISO settings at the following link:

http://photography-on-the.net/forum/showthread.php?t=1081982

Core ISO levels up to ISO 1600 (which, in the case of the 5D III, might now be ISO 12800) are a base ISO setting (initial amplification) with a push or pull of 1/3rd stop of analog gain. Beyond the maximum core ISO setting, things tend to get a bit nastier, and you might have initial amplification, analog gain, digital gain, and metadata settings that tell the RAW processor to boost even more in post.

Personally, one of the things I was really hoping Canon would improve on with the 5D III and 1D X was their approach to ISO. Its annoying to alternatively lose a third stop of DR for "non-native" ISO settings. It would be much nicer to simply have electronic amplification right off the pixel directly to the appropriate ISO. Thats what Sony Exmor sensors do, and they seem to do a pretty good job of it until higher ISO settings. In the grand scheme of things, though, it doesn't really matter a wit to IQ...having shot Canon cameras for a number of years now, lower-ISO noise is largely indistinguishable across settings when you properly expose. It may only manifest as a minor problem if you need multiple low-light shots from a sequence that involved differing lower-ISO settings to be very consistent...in which case you might notice slight differences in noise characteristics between frames. On an individual shot basis were no consistency is necessary, the only real drawback is that 1/3rd stop DR loss for non-native (100, 200, 400, etc.) settings.

160 is not 200 underexposed 1/3 stop and then given an analog boost, that doesn't even make sense.

And the 160,320, etc. ones do NOT lose any DR compared to 200,400, etc.

Again all they are are the 1/3 ISO above ISO (200,400, etc.) exposed 1/3 longer than the metering indicates and then digitally brought back down. So you get the expected 1/3 stop better SNR than 200 because it was exposed for 1/3 stop longer (NOT underexposed 1/3 stop, it gets over-exposed 1/3 stop) and you get 1/3 stop better shadows and the 1/3 stop clipped highlights.

 

[LetTheRightLensIn]

Referring to Mr. Claff's Photographic Dynamic Range chart:
http://home.comcast.net/~NikonD70/Charts/PDR.htm#EOS%201D%20Mark%20IV
(note: select EOS 5D Mk III)

The analysis methodology appears sound imho, however regardless of the absolute accuracy, it is the delta that is of interest here and with that, it clearly supports the multiples of 160 theory (at least for the 5D Mk3).

Put another way in terms of read noise:
http://home.comcast.net/~NikonD70/Charts/RN_ADU.htm

again clearly shows iso160, 320, 640, etc. having the least noise.

Weird, he gets some 1/3 down ISOs with higher DR, since everyone else has gotten 160/200, 320/400 to have the same DR for the 5D2. I wonder if he estimated or measure them.

See for instance (note that is a straight engineering DR chart, it doesn't take into account banding visual look; also it doesn't take into account how ISOs are rated camera vs camera):

 

[jrista]

All Canon cameras most definitely have NATIVE ISO 100, 200, 400, 800, etc. stops. The reason noise is cleaner on the -1/3rd stop ISO settings is because Canon uses a Base/Push/Pull approach to ISO. As an example, ISO 320 is actually ISO 400 with a -1/3rd stop "pull" that underexposes the ISO 400 shot by one third of a stop, then "recovers" the underexposure with a bit of extra analog gain. That "pushes down" noise at ISO 320, which is why it looks cleaner...but, its at the cost of an additional third-stop of dynamic range.

You can find a complete analysis of Canon's method of achieving the various ISO settings at the following link:

http://photography-on-the.net/forum/showthread.php?t=1081982

Core ISO levels up to ISO 1600 (which, in the case of the 5D III, might now be ISO 12800) are a base ISO setting (initial amplification) with a push or pull of 1/3rd stop of analog gain. Beyond the maximum core ISO setting, things tend to get a bit nastier, and you might have initial amplification, analog gain, digital gain, and metadata settings that tell the RAW processor to boost even more in post.

Personally, one of the things I was really hoping Canon would improve on with the 5D III and 1D X was their approach to ISO. Its annoying to alternatively lose a third stop of DR for "non-native" ISO settings. It would be much nicer to simply have electronic amplification right off the pixel directly to the appropriate ISO. Thats what Sony Exmor sensors do, and they seem to do a pretty good job of it until higher ISO settings. In the grand scheme of things, though, it doesn't really matter a wit to IQ...having shot Canon cameras for a number of years now, lower-ISO noise is largely indistinguishable across settings when you properly expose. It may only manifest as a minor problem if you need multiple low-light shots from a sequence that involved differing lower-ISO settings to be very consistent...in which case you might notice slight differences in noise characteristics between frames. On an individual shot basis were no consistency is necessary, the only real drawback is that 1/3rd stop DR loss for non-native (100, 200, 400, etc.) settings.

160 is not 200 underexposed 1/3 stop and then given an analog boost, that doesn't even make sense.

And the 160,320, etc. ones do NOT lose any DR compared to 200,400, etc.

Again all they are are the 1/3 ISO above ISO (200,400, etc.) exposed 1/3 longer than the metering indicates and then digitally brought back down. So you get the expected 1/3 stop better SNR than 200 because it was exposed for 1/3 stop longer (NOT underexposed 1/3 stop, it gets over-exposed 1/3 stop) and you get 1/3 stop better shadows and the 1/3 stop clipped highlights.

Sorry, the extra analog gain is only applied to settings above 1600 (or possibly higher with the 5D III). It is a digital correction, not an analog correction (apologies for the miss), that would either underexpose (as in the case of ISO 160, down from 200) or overexpose (as in the case of ISO 250, up from 200). BOTH third-stop cases, 160 and 250, DO indeed lose a third stop of dynamic range. Whether you actually clip highlights and gain shadow SNR would entirely depend on the photographer...generally speaking, any half-witted photographer can use the in-camera histogram to avoid an overexposure, which would mitigate any SNR gain on the shadow end. Thus...loss in DR, not a gain in shadow performance.

Actually, one of the DR comparison charts that pitted the D800 against the 5D III clearly indicated the slight drop in DR in the 1/3rd stop ISO settings between ISO 100 and 400 on the 5D III. I'll see if I can find the link you posted, I forget what thread it was to. I am not exactly sure how that particular chart is derived, it seemed to be offset from DXO numbers by a few stops, but it definitely demonstrated the loss in DR on Canon cameras at non-native low-ISO settings.

 

[LetTheRightLensIn]

All Canon cameras most definitely have NATIVE ISO 100, 200, 400, 800, etc. stops. The reason noise is cleaner on the -1/3rd stop ISO settings is because Canon uses a Base/Push/Pull approach to ISO. As an example, ISO 320 is actually ISO 400 with a -1/3rd stop "pull" that underexposes the ISO 400 shot by one third of a stop, then "recovers" the underexposure with a bit of extra analog gain. That "pushes down" noise at ISO 320, which is why it looks cleaner...but, its at the cost of an additional third-stop of dynamic range.

You can find a complete analysis of Canon's method of achieving the various ISO settings at the following link:

http://photography-on-the.net/forum/showthread.php?t=1081982

Core ISO levels up to ISO 1600 (which, in the case of the 5D III, might now be ISO 12800) are a base ISO setting (initial amplification) with a push or pull of 1/3rd stop of analog gain. Beyond the maximum core ISO setting, things tend to get a bit nastier, and you might have initial amplification, analog gain, digital gain, and metadata settings that tell the RAW processor to boost even more in post.

Personally, one of the things I was really hoping Canon would improve on with the 5D III and 1D X was their approach to ISO. Its annoying to alternatively lose a third stop of DR for "non-native" ISO settings. It would be much nicer to simply have electronic amplification right off the pixel directly to the appropriate ISO. Thats what Sony Exmor sensors do, and they seem to do a pretty good job of it until higher ISO settings. In the grand scheme of things, though, it doesn't really matter a wit to IQ...having shot Canon cameras for a number of years now, lower-ISO noise is largely indistinguishable across settings when you properly expose. It may only manifest as a minor problem if you need multiple low-light shots from a sequence that involved differing lower-ISO settings to be very consistent...in which case you might notice slight differences in noise characteristics between frames. On an individual shot basis were no consistency is necessary, the only real drawback is that 1/3rd stop DR loss for non-native (100, 200, 400, etc.) settings.

160 is not 200 underexposed 1/3 stop and then given an analog boost, that doesn't even make sense.

And the 160,320, etc. ones do NOT lose any DR compared to 200,400, etc.

Again all they are are the 1/3 ISO above ISO (200,400, etc.) exposed 1/3 longer than the metering indicates and then digitally brought back down. So you get the expected 1/3 stop better SNR than 200 because it was exposed for 1/3 stop longer (NOT underexposed 1/3 stop, it gets over-exposed 1/3 stop) and you get 1/3 stop better shadows and the 1/3 stop clipped highlights.

Sorry, the extra analog gain is only applied to settings above 1600 (or possibly higher with the 5D III). It is a digital correction, not an analog correction (apologies for the miss), that would either underexpose (as in the case of ISO 160, down from 200) or overexpose (as in the case of ISO 250, up from 200). BOTH third-stop cases, 160 and 250, DO indeed lose a third stop of dynamic range. Whether you actually clip highlights and gain shadow SNR would entirely depend on the photographer...generally speaking, any half-witted photographer can use the in-camera histogram to avoid an overexposure, which would mitigate any SNR gain on the shadow end. Thus...loss in DR, not a gain in shadow performance.

Actually, one of the DR comparison charts that pitted the D800 against the 5D III clearly indicated the slight drop in DR in the 1/3rd stop ISO settings between ISO 100 and 400 on the 5D III. I'll see if I can find the link you posted, I forget what thread it was to. I am not exactly sure how that particular chart is derived, it seemed to be offset from DXO numbers by a few stops, but it definitely demonstrated the loss in DR on Canon cameras at non-native low-ISO settings.

What you say doesn't match what countless people other than what the one Nikon guy linked to above discovered.

Nobody else has found the 160,320 to lose 1/3 third stop of DR compared to 200,400, etc.

Measure black frame read noise at ISO160 and 200 and then measure saturation point, in the end, you end up with the same DR, the only difference is the metering system is biased in favor of shadows if you use ISO160 and biased in favor of saving highlights if you use ISO200.

 

[Janco]

I literally don't get NOTHING of this discussion, simply because maybe I have a misunderstanding how different ISO's in digital cameras are achieved. I always thought ISO settings in digital are only made by different multiplication factors of the amount of sensor information (signal+noise) readout, so for example ISO200 would be achieved simply by boosting/amplifying the number of photons that hit any example pixel by the factor 2 (double) in comparison to the value it would have for ISO100 and ISO320 would be amplified 3.2x the value it would have at ISO100. With this theory there would not be a reason to take the picture at ISO400 but underexpose it by a third or pulling pushing something? OR is this because of FULL stops at aperture and speed? I guess I answered the question already, but I'm not sure if all is that simple. Or hmm, I guess it is?
Please enlighten me! Thanks....

 

[Jettatore]

I would say in general this simply doesn't matter for real world use. I actually like the discussion that is going on above, but it really doesn't matter for an end user who's goal is to take nice clean images. The best way to reduce noise in an image before editing is to expose the image properly and not try to use the camera as a nightvision scope. Properly exposed shots that are adequately lit, using higher ISO's to buy smaller aperature and faster shutter speeds do not suffer the same noise issues that improperly exposed shots making up for dramatically bad lighting conditions do.

Example: ISO 1600, f/8.0, 1/250 (using higher ISO to stop action and incread DOF) :: ISO 1600, f/2.8, 1/25 + a steady hand (using high ISO to see in the dark)

When I shoot in severe low light, I try to capture exactly what my eyes can see in that situation and work my compositions around that philosophy. A small touch of RAW noise removal and post editing yields phenomenal results with this type of approach. The nice thing on top of all this is that as the cameras keep getting better and better, you actually can use the camera's as if they were night vision goggles and get away with it if you're clever.

 

[esi32]

I literally don't get NOTHING of this discussion, simply because maybe I have a misunderstanding how different ISO's in digital cameras are achieved. I always thought ISO settings in digital are only made by different multiplication factors of the amount of sensor information (signal+noise) readout, so for example ISO200 would be achieved simply by boosting/amplifying the number of photons that hit any example pixel by the factor 2 (double) in comparison to the value it would have for ISO100 and ISO320 would be amplified 3.2x the value it would have at ISO100. With this theory there would not be a reason to take the picture at ISO400 but underexpose it by a third or pulling pushing something? OR is this because of FULL stops at aperture and speed? I guess I answered the question already, but I'm not sure if all is that simple. Or hmm, I guess it is?
Please enlighten me! Thanks....

Broad strokes: Photons hit's photo site, some get converted into an electron. The electrons accumulate during the exposure and create a vary small voltage. That voltage is amplified by one or more variable gain amps then fed into the Analog-digital converter (ADC). The number that comes out of the ADC goes to the RAW file.

Now fun part is when you get to doing the amplification. There are several ways to go about doing it, the following 3 have actually been used to my knowledge in cameras.
1) a single amp that has 1/3rd stop steps (AFAIK Nikon does this)
2) a single amp that has 1 stop steps and push and pull via software and adjusting the meter (AFAIK Canon's non EOS-1 bodies do this)
3) 2-stages of amps that handle full and fractional stops in stages (AFAIK, Canon's EOS-1 bodies, at least I know for sure the Mk3s do it this way).

For case 1, lower ISOs are always better (less noise more DR) than higher ones.

For case 2, the +1/3rd stop ISOs are worse for noise and DR than the base ISOs. The +2/3rd stop ISOs are better for noise but not DR than the base ISO, and at low ISOs often better for noise than the +1 ISO. Once you hit unity gain it doesn't matter, things just go down hill anyway and you want to use the lowest ISO you can.

For case 3, below unity gain, IME all +1/3 ISOs are worse than the base ISO, and +2/3rds ISOs are worse than the base and base +1 stop ISOs. Once you hit unity gain ISOs get progressively worse as you'd expect.

Expanded ISOs (L, H1, H2) are merely mathematical manipulations of the highest (or lowest) real ISO setting the camera can do. IMO, these are the "when having a picture is more important than having a good picture" settings, at least H1 and H2.

Unity gain, is the point where 1 collected photon == 1 RAW value step.

 

[Janco]

Thank you esi32! I could imagine you're either some kind of camera engineer or maybe a teacher with excellent camera knowledge (assuming you're right with your post ;D )

 

[psolberg]

the noise at low ISOs is very disturbing on the 5DmkIII
http://diglloyd.com/ (subscription required, I can't post links but you can read his "reader comments")