Those "in-between" ISOs...

[canoli]

Hey All -


I've read a number of conflicting opinions / theories about using in-between ISOs, the 320, 500, etc. - and I'm hoping someone here is able to share "just the facts ma'am." (!)


I'm curious if those 1/3 increment ISOs are good to use, or should be avoided. In various photo mags I see ISO 500 or 320 etc. but more often I see 100, 200, 800, etc.


Is it true that camera CPUs arrive at those 1/3 ISOs by interpolation? That would mean they're not "native" to the capture system, but does that make them (by definition) inferior?


I've used my 40D at various ISOs and don't see anything unexpected at the 1/3 ISOs...


I'd like to learn the "best practice" ways to shoot - is avoiding in-between ISOs part of it?


Thank you for any thoughts you can share.

[Maleko]

i dont tend to use many "in between" ISO's, I tend to either use 100, 200, 400, 800, but instead of jumping to 1600, I like to use 1000 for soem indoor shooting that doesnt quite need to be 1600, so i guess you coudl say I use 1/3 ISO's more so at higher ends.

[crystalshadow]

I don't use the 1/3 ISO's myself, but that's more out of habit than anything. I'd say you've answered yourself: "I've used my 40D at various ISOs and don't see anything unexpected at the 1/3 ISOs...". If it looks good to you, I say run with it!

[canoli]

Ha! I think you're right CS, I should just leave well-enough alone. Like Mal, indoors I use ISO 1000 frequently. I wonder if 1600 would be better though, since it's a "native" ISO.


I stick by the "use the lowest possible ISO" rule and it seems to serve me well. But I wonder if I'm missing something by avoiding those in-between ISOs. The only one I ever use is 1000, when 800 won't give me a fast enough shutter, but I don't necessarily need to double the sensitivity and go to 1600.

[Jarhead5811]

I'm sure Daniel Browning will be along shortly to explain things.

[Keith B]

It's not interpolation, it more like using levels. It is just like shooting something at 100 and the in RAW bumping up the Exposure by 1/3 takes it to 125 and 2/3 would take it to 160. I actually use ISO 160 on my camera when shooting portraits just to pay homage to Kodak VC 160 film. I know it isn't really making any difference but it is a mental thing. OCD if you will.

[Bill W]

This issue came up on a thread I started about my first baseball captures.....this is from Chuck Lee, click on the link;


Bill and Peety,
<blockquote>
<div>[img]../Themes/hawaii/images/icon-quote.gif[/img] Bill W:</div>
<div>Your
statement "walking ISO down" is fully understandable and shooting at
"tweener" ISO's is interesting...I'll search for DB's entry on this
subject.</div>
</blockquote>


The tweener ISOs on a 40D at 160, 320, 640, 1250 are better than the
default 100,200,400,800,1600. ISO 160 is as good as it gets on a 40D. Read Thisit's for the 30D but pertains to the 40D as well. This is not true for the 5D where the standards are best.


Regards


Bill

[canoli]

Thanks for the link Bill. The bottom graph is easy enough to understand, but the top one is a complete mystery to me.


for what it's worth, I've never seen any combing in my RAW files.


I worked my way through the links and found his 40D page, with the same "dark noise tests he did for the 30D. But I still can't make sense of it.


He refers to the 'tween ISOs as "upscaled fake ISOs" and "downscaled fake ISOs." The only conclusion I can draw (so far, without some help) is that the sensor multiplies or divides the native ISOs to arrive at the 'tween ISOs.


But he never says if that's a bad thing (though the word "fake" doesn't exactly inspire confidence in using them!).

[Daniel Browning]

The short answer is: don't use tweeners.

Read on for the long answer.

I'm curious if those 1/3 increment ISOs are good to use, or should be avoided.

It depends on how picky you are. If you want maximum dynamic range, you'll avoid at least half of them.

In various photo mags I see ISO 500 or 320 etc. but more often I see 100, 200, 800, etc.

Not surprising. Few people are aware of what the optimal settings are and the reason why they are optimal.

Is it true that camera CPUs arrive at those 1/3 ISOs by interpolation?

In some cameras, yes. This includes the 5D2, xxD (50D, 40D, 30D), xxxD (500D, 450D, 400D), and 1000D.

In other cameras, the tweener ISO settings are achieved through a separate analog amplifier. That includes the 5D1, xD (1D, 1D2, 1D3), and xDs (1Ds, 1Ds2, and 1Ds3).

That would mean they're not "native" to the capture system,

Correct.

but does that make them (by definition) inferior?

Yes. Digital manipulation of the raw file is always inferior because it always increases quantization error and sometimes also clips highlights by 1/3 stop. It could have been implemented as metadata ISO and would have avoided posterization completely while still offering the same benefit to the user.

If the cameras were engineered with the correct bit depth (e.g. 12 bits for the 5D2 instead of 14), then the quantization error from tweener ISO settings might actually show up in the extreme shadows as posterization. In this case the two mistakes cancel each other out (14 bits is enough to hide posterization).

I've used my 40D at various ISOs and don't see anything unexpected at the 1/3 ISOs...

The difference is not visible to most people when they use typical tone curves and default raw conversion parameters. That's because most defaults are tuned for very little dynamic range: just 5-7 stops between highlights and shadows. If you shoot scenes that have a little more contrast, such as 9 to 11 stops, the difference is more noticable. That is because the difference is in read noise dominated tones.

I'd like to learn the "best practice" ways to shoot - is avoiding in-between ISOs part of it?

On some of cameras with analog tweeners (1D3, 1Ds3), some of the settings have slightly less read noise, but generally the loss in highlight headroom is not worth it. In all other situations the tweeners are either worse or the same.

I should mention that tweeners can sometimes help broken raw converters. Canon's DPP, for example, doesn't use the correct white point for many of their own cameras. They set it too low, ignoring plenty of good data in the highlights completely. In that case, the -1/3 ISO settings (160, 320, 640, etc.) will move the actual white point down to the same spot where DPP "thinks" the white point is, so you get back that 1/3 stop of highlights. Of course, other raw converters don't have this bug, so they see the highlights either way. I don't suggest forming your shooting strategy around bugs in raw converters. Better to just live with the 1/3 loss (which is very minor) or switch to a better converter.

The +1/3 digital tweeners (ISO 125, 250, 500, etc.) are the ones that have negative effects. Compared to ISO 100 with a fixed exposure, ISO 125 clips 1/3 stop of highlights with no improvement in read noise. Compared to ISO 200 with fixed exposure, ISO 125 has 1/3 stop *more* highlight headroom, but it also has nearly double the read noise. That means ISO 200 actually has more dynamic range and less shadow noise than ISO 125.

That's not to say they that tweeners should never be used. For JPEG shooters (not raw), it's important to get the image brightness just so, and sometimes ISO is easier to change than exposure. Video, too, benefits from tweeners for the same reason.

...instead of jumping to 1600, I like to use 1000 for soem indoor shooting that doesnt quite need to be 1600, so i guess you coudl say I use 1/3 ISO's more so at higher ends.

I would suggest ISO 1250 instead, because it has noticably less noise than ISO 1000 for the same exposure.

in RAW bumping up the Exposure by 1/3 takes it to 125 and 2/3 would take it to 160.

Yep. That's exactly how 125/250/500 work in the camera, too. The other tweeners (160, 320, 640) are slightly different: they start with the next-highest ISO and subtract 1/3.

The tweener ISOs on a 40D at 160, 320, 640, 1250 are better than the default 100,200,400,800,1600.

I wouldn't put it that way. They yeild the same result as shooting ISO 200 +1/3 EC. You lose 1/3 stop highlights, but gain less noise. They're really the same, not better. (In a fixed exposure, ISO 160 has lower read noise in absolute ADU, but the SNR is the same, so it's not better that way.)

ISO 160 is as good as it gets on a 40D.

ISO 100 has 1/3 stop more dynamic range and 1/3 stop higher SNR in the shadows than ISO 160.

To attain maximum dynamic range, tweeners should be avoided until they are implemented the correct way, as metadata.

[Keith B]

Ah. I didn't think interpolation was the correct term for this. I guess it makes sense. I was just thinking how it relates to resizing images.


So if I understand correctly, ISO 160 on my 5D mkII isn't necessarily a bad thing as 125 would be? But 200 would be better and then bump my lights down a third. Right?