ISO vs noise levels

[Kalunga Lima]

Having just acquired a 5D Mkii, I was looking for info on the merits of manually setting the ISO vs Automatic when I came upon this ISO vs Noise Level graph.

Canon 5D Mark II Noise Levels vs. ISO - Canon Photography Group

It would seem that the 5D respond best when set to it's "naive" ISO settings (100, 160, 320, 640, 1250) than to intermediary settings which it apparently must extrapolate.

Strangely enough, according to the graph, the 5D Mkii shows less noise at ISO 1250 than at ISO 125, after which things get exponentially nosier from ISO 1600 onwards.

In the mean time my question remains unanswered: when shooting video, what is the merit of setting the ISO manually vs the automatic settings.

Ironically, this digital camera is forcing us to think "celluloid" all over again.

cheers to all

[Evan Donn]

Well, issues of noise aside, the primary advantage to setting ISO manually vs. auto is that it prevents the camera from changing exposure mid shot.

In terms of noise, yes you can minimize the noise by using the native ISOs, so that's one reason to do it. Another reason is that on auto the camera will often use a higher ISO than necessary in dark shots - it's trying to make the whole scene brighter when you might be fine with the overall look being dark. Setting it manually lets you decide what portion of the image is most important in terms of exposure.

None of this, other than the native ISO noise issue, is any different than any other pro video camera. Auto anything is generally to be avoided if you care about getting the best possible shot and maintaining shot-to-shot consistency.

[Ilya Mamonov]

Hey Kalunga thanks for the link. Those are some interesting results. From that test it looks like it's a good idea to keep your camera set at ISO 640. By the way I think you are mistaken about "native" ISOs (the most popular ones are 50, 100, 200, 400, 800, 1600, 3200). And yes, stay away from Auto mode.

[Daniel Browning]

Originally Posted by Ilya Mamonov

From that test it looks like it's a good idea to keep your camera set at ISO 640.

That would result in more noise than ISO 400, 320, 200, 160, and 100.

[Daniel Browning]

Originally Posted by Kalunga Lima

I came upon this ISO vs Noise Level graph.

There are just so many things wrong with the "testing" for the graph to have any useful information. I don't even know where to start. For one, analyzing test frames with the lens cap on is fine if you know how to correlate that with actual performance. Clearly the tester who made that graph does not. Furthermore, the bend in the curve occurs at exactly the arbitrary tonal level selected by the tester. Other levels would have bent the curve in other places, which obviously removes all meaning from the graph. Even for a still photographer, it has no correlation to anything useful whatsoever, but that goes several times over for video use.

Originally Posted by Kalunga Lima

It would seem that the 5D respond best when set to it's "naive" ISO settings (100, 160, 320, 640, 1250) than to intermediary settings which it apparently must extrapolate.

Actually, it is the 100, 200, 400, 800, 1600, and 3200 settings which are not interpolated. 160, 320, 640, and 1250 are the result of digital manipulation.

Originally Posted by Kalunga Lima

Strangely enough, according to the graph, the 5D Mkii shows less noise at ISO 1250 than at ISO 125, after which things get exponentially nosier from ISO 1600 onwards.

For a fixed exposure and brightness, it is true that 1250 has less read noise than 125. And so does ISO 1600. But the test was not based on equal brightness, which was not the condition under which the test was made, so it only shows the correlation by accident.

In any case, fixed exposure/brightness only applies to raw photographers. For video use, exposure has to float with ISO to retain fixed brightness, and in that regime, ISO 1250 has far more noise than ISO 125. Anyone with a camera knows that already.

[Ilya Mamonov]

Originally Posted by Daniel Browning

That would result in more noise than ISO 400, 320, 200, 160, and 100.

Maybe I did not understand the chart correctly. Could you explain please

[Daniel Browning]

Originally Posted by Ilya Mamonov

Maybe I did not understand the chart correctly. Could you explain please

You did understand the chart correctly, but the chart is wrong.

[Kalunga Lima]

Thanks Daniel for the insight.

So much for "point and shoot" :)

[Ilya Mamonov]

Originally Posted by Daniel Browning

You did understand the chart correctly, but the chart is wrong.

But why do you think the chart is wrong? Just interesting to know what you are basing your responses.

[Daniel Browning]

Originally Posted by Ilya Mamonov

But why do you think the chart is wrong? Just interesting to know what you are basing your responses.

I touched on some of the reasons in post #5 above. The test conditions are with the lens cap on, raw stills, converted in ACR, with arbitrary settings. This results in a chart of completely arbitrary results. If he had measured the raw data directly, instead of with ACR, and correlated the bias frame measurements with the conversion gain for each ISO setting, the he could have created a chart with useful information. There's really no useful information in this chart at all for video shooters, though it's misleading for still photographers as well.

[Bill Binder]

Originally Posted by Daniel Browning

There are just so many things wrong with the "testing" for the graph to have any useful information. I don't even know where to start. For one, analyzing test frames with the lens cap on is fine if you know how to correlate that with actual performance. Clearly the tester who made that graph does not. Furthermore, the bend in the curve occurs at exactly the arbitrary tonal level selected by the tester. Other levels would have bent the curve in other places, which obviously removes all meaning from the graph. Even for a still photographer, it has no correlation to anything useful whatsoever, but that goes several times over for video use.



Actually, it is the 100, 200, 400, 800, 1600, and 3200 settings which are not interpolated. 160, 320, 640, and 1250 are the result of digital manipulation.



For a fixed exposure and brightness, it is true that 1250 has less read noise than 125. And so does ISO 1600. But the test was not based on equal brightness, which was not the condition under which the test was made, so it only shows the correlation by accident.

In any case, fixed exposure/brightness only applies to raw photographers. For video use, exposure has to float with ISO to retain fixed brightness, and in that regime, ISO 1250 has far more noise than ISO 125. Anyone with a camera knows that already.

Finally someone rebutting that ridiculous theory. It's incredible how that crap has permeated the Net to the point it's become "true." LOL. People, this is not hard, the native ISO's are 100, 200, 400, etc. Canon makes that very clear, and these "tests" are as good as useless.

[Olof Ekbergh]

Here are my purely personal non scientific observations, after using the 5DmkII for a bit more than a year.

Stills and video look best at ASA 100.

Stills and video start to show grain/noise at ASA 400.

Stills and video are acceptable up to and including ASA 1600.

So the way I use the camera is always shoot at 100 if you can. It is fine to go to 400, 800 and 1600 though it will add some grain/noise.

Only if you really have to go higher than 1600.

Like I said this is from looking at and working with video and stills professionally for about a year. And it is only my personal esthetic opinion, and it is how I use the camera.

In low light I try to use a fast lens f1.4 prime or f2.8 zoom, rather than up the iso past 1600.

[Khoi Pham]

Originally Posted by Daniel Browning

I touched on some of the reasons in post #5 above. The test conditions are with the lens cap on, raw stills, converted in ACR, with arbitrary settings. This results in a chart of completely arbitrary results. If he had measured the raw data directly, instead of with ACR, and correlated the bias frame measurements with the conversion gain for each ISO setting, the he could have created a chart with useful information. There's really no useful information in this chart at all for video shooters, though it's misleading for still photographers as well.

I get almost same result as the chart shooting real live video, all you have to do is hook up your hdmi out to your HD edit monitor, point your camera into a dark area and change the ISO settings, noise are definetly less at 160, 320, 640, 1250, the higher the ISO the more noise, but if you need ISO100, 160 has less noise... for some of us that works mostly in low light and have no control of lighting, noise is what we worry about most, not clipping the highlights cuz there ain't no highlight to be clip.

[Daniel Browning]

Originally Posted by Khoi Pham

noise are definetly less at 160, 320, 640, 1250,

That's true, but the trade off is 1/3 stop clipped highlights. For example, when you set the camera to ISO 160, what it actually does behind the scenes is use ISO 200 (metered for 160), then reduce brightness by 1/3 stop. If the picture profile settings are set for enough contrast that the 1/3 stop isn't used anyway, then the loss of highlights wont be noticed. It's the exact same thing as ISO 50, except ISO 50 clips a full stop of highlights (and reduces noise by a full stop).

[Khoi Pham]

Yeah but I'm shooting full manual and don't rely on any in camera meter so whatever it does behind the scene doesn't matter as long as I get less noise in my video.