Canon cameras and in-between ISO question

[Axel Hildebrandt]

When the 1D3 came out, I remember that Emil Martinec said that Canon implements the in-between ISO value through software and Nikon through hardware. If memory serves right, he said that between ISO 200 and 800 it would be better to use 'real' ISO values if low noise is the priority. If this is the case, can we expect that this is true for the Mark IV as well?

Thanks!

[Sabyasachi Patra]

I would also be interested to hear his views. I don't use the intermediate ISOs in the 1D Mark II. Although trying all the ISOs including intermediate ones in the Mark IV. Look forward to hearing from Emil.

Cheers,
Sabyasachi

[Roger Clark]

The differences are so small in using the in-between ISOs that it would take careful measurements to see the differences is real world images. Where it will matter a little is very low light imaging, e.g. astrophotos, or very high ISO imaging, above about 3200. And even then it will only be a slight factor in the deepest shadows where noise would appear slightly higher.
A larger concern is at the high end, the 1/3 stop ISO above the factors of 2 ISOs (100, 200, 400, 800, 1600) is multiplied up, so if you saturate, the top 2/3 stop will be saturated, a little sooner one would expect.

In general, I feel one has enough control on exposure you shouldn't need to split ISO hairs.

Emil does have a very detailed web page on this. ---Hmmm I can't find it on his web page--probably tied up in the string theory.;)

Roger

[Axel Hildebrandt]

Thanks, Roger! In your experience, would you say that as far as noise is concerned it makes sense to use ISO 640 instead of ISO 800?

[Emil Martinec]

I gave a detailed explanation to this question some time ago on NSN:

http://www.naturescapes.net/phpBB3/v...p?f=2&t=139621

The observations for Canon 1 series cameras are based on observations

http://theory.uchicago.edu/~ejm/pix/...ml#read_vs_iso

see figure 15.

I have not done measurements on the 1D4, so I don't know if it implements intermediate ISO's as the 1D2/2n and 1D3 do (via a secondary hardware amplifier) or the way that the xxD, 7D and 5D2 do (via software multiplication; BTW, I speculated incorrectly in the NSN post -- my understanding is that the 5D2 does intermediate ISO's in the manner of the xxD cameras, as does the 7D).

Bottom line: On the 1D3 and 1D2/2n, the intermediate ISOs are so poorly implemented that there is no noise advantage in the RAW data relative to underexposure at the next lowest "main" ISO (eg ISO 400 instead of ISO 500 or 640) and pushing the exposure compensation in the RAW converter. The benefit is an extra 1/3 or 2/3 stop of highlight headroom.

On the xxD cameras, 7D and 5D2, the intermediate ISOs are implemented via a digital multiplication that pulls or pushes 1/3 stop from the nearest (rather than next lowest) "main" ISO, eg 320 and 500 are both derived from hardware amplification at ISO 400. So eg 320 is simply ISO 400 overexposed by 1/3 stop (so you have 1/3 stop less headroom relative to metering), and 500 is 400 underexposed 1/3 stop and pushed in the camera. So the pushed ones throw away 1/3 stop of highlights in the camera, and I'd rather have the choice of keeping them or not at the postprocessing stage; for the pulled ones, I'd rather have my meter telling me the truth and make a conscious decision to do the overexposure. Thus I have the custom function set that turns off intermediate ISO's on both my 1D3 and 40D.

[Roger Clark]

Thanks, Roger! In your experience, would you say that as far as noise is concerned it makes sense to use ISO 640 instead of ISO 800?

ISO 640 is one of those in-between ISOs and would be ISO 800 moved down 1/3 stop, so no real benefit. I just use the factors of 2 ISOs: 100, 200, 400, 800, 1600, etc.

Roger

[Axel Hildebrandt]

Thanks, Emil! It would be good to know if the intermediate ISO values in the 1D4 are implemented as in the 1D3.

[Axel Hildebrandt]

ISO 640 is one of those in-between ISOs and would be ISO 800 moved down 1/3 stop, so no real benefit. I just use the factors of 2 ISOs: 100, 200, 400, 800, 1600, etc.

Roger

Thanks, Roger. I photographed in very low light the other day and got the impression that ISO 2000 looked less noisy than 2500 and 3200 but have to experiment more.

[Emil Martinec]

Thanks, Roger! In your experience, would you say that as far as noise is concerned it makes sense to use ISO 640 instead of ISO 800?

That's a question of where your exposure lies. For a fixed exposure (Tv and Av) ISO 800 will have less noise (mostly in shadows) than ISO 640 (which is as noisy as 400 underexposed on 1 series prior to the 1D4, and perhaps also the latter if and when it is tested for read noise). So the real question is whether you are free to open up the exposure to the point where ISO 800 would lose you highlights that you want to keep. If not, use 800; if so, then use a lower ISO.

For a fixed exposure, you are always getting the same number of photons; what changes with ISO is that the number of photons that corresponds to saturation of the raw data drops in proportion to ISO, and as a secondary effect the read noise goes down relative to a fixed signal as the ISO increases. This fact is about the only reason there needs to be and ISO setting in the camera, rather than just a metadata tag saying what the metering was, so that the converter could apply an appropriate signal boost before doing the conversion. Since the read noise goes down with ISO in terms of photon equivalents, it pays to boost the ISO in the camera for weak photon signals (low light).

[Emil Martinec]

Thanks, Roger. I photographed in very low light the other day and got the impression that ISO 2000 looked less noisy than 2500 and 3200 but have to experiment more.

What camera model? As I mentioned above, the decision is different for 1 series vs the rest.

[Axel Hildebrandt]

What camera model? As I mentioned above, the decision is different for 1 series vs the rest.

I'm using a 1D Mark IV now.

[Axel Hildebrandt]

That's a question of where your exposure lies. For a fixed exposure (Tv and Av) ISO 800 will have less noise (mostly in shadows) than ISO 640 (which is as noisy as 400 underexposed on 1 series prior to the 1D4, and perhaps also the latter if and when it is tested for read noise). So the real question is whether you are free to open up the exposure to the point where ISO 800 would lose you highlights that you want to keep. If not, use 800; if so, then use a lower ISO.

For a fixed exposure, you are always getting the same number of photons; what changes with ISO is that the number of photons that corresponds to saturation of the raw data drops in proportion to ISO, and as a secondary effect the read noise goes down relative to a fixed signal as the ISO increases. This fact is about the only reason there needs to be and ISO setting in the camera, rather than just a metadata tag saying what the metering was, so that the converter could apply an appropriate signal boost before doing the conversion. Since the read noise goes down with ISO in terms of photon equivalents, it pays to boost the ISO in the camera for weak photon signals (low light).

For birds in flight I use AV or manual and raising the ISO to 800 is beneficial because of shutter speed anyway. Thanks again!