Idiot ISO and Numpty ISO

[Daniel Browning]

Every ISO setting on a digital camera can be classified using four different types. I recommend that you avoid Idiot and Numpty ISO settings whenever possible, and it's good to understand the trade-offs surrounding the use of Bargain ISO. The types are:

• Idiot ISO
• Numpty ISO
• Bargain ISO
• Smart ISO

Most ISO settings on digital cameras are made up of a mix of several types of ISO. For example, on my 5D2:

• ISO 50: Smart ISO
• ISO 125: Idiot ISO
• ISO 160: Bargain and Idiot ISO
• ISO 200: Bargain
• ISO 250+HTP: Bargain, Idiot, and Smart ISO
• [...]
• ISO 3200: Bargain and Numpty ISO
• ISO 4000: Bargain, Numpty and Idiot ISO
• ISO 5000+HTP: Bargain, Numpty, Idiot, and Smart ISO

The settings on other cameras may be made of different combinations (though tend to be largely similar for a given manufacturer). Here is a detailed example, showing how all four types combine to make ISO 5000 on the 5D2:

• +4 Bargain ISO to get up to 1600.
• +1 Numpty ISO to get from 1600 to 3200.
• -1/3 Idiot ISO to get from 3200 to 2500
• +1 Smart ISO to get from 2500 to 5000

The function of each of the types, briefly:

• Idiot ISO increases the brightness of the data in the raw file using digital gain (math inside the camera) and has no benefit whatsoever.
• Numpty ISO increases the brightness of the data in the raw file using analog gain and has no benefit whatsoever.
• Bargain ISO increases the brightness of the data in the raw file using analog gain, but has benefits/trade-offs.
• Smart ISO doesn't touch the raw file at all, it just changes metadata (similar to setting white balance).

Now a more detailed description of what each of the different types of ISO means.

Idiot ISO is when the brightness of the raw file is increased using math inside the camera. This is similar to using the "EC" slider in your raw converter, except that the changes are written permanently to the raw file. This causes the needless loss of data (clipped highlights or shadows), larger files, and increased quantization error (for non-integer factors).

There is never a reason or excuse for camera manufacturers to use such a braindamaged design when it is in fact *easier* to use Smart ISO than Idiot ISO. This type of ISO implementation has no redeemeing value whatsoever, hence the name.

The next two types of ISO are similar in that they both use an amplifier (analog gain) to increase the brightness of the data in the raw file. The only difference is this: if the resulting amplification has a better Signal-to-Noise ratio (SNR) than Idiot ISO, then I call it "Bargain ISO". If it has the same SNR, then I call it Numpty ISO.

Numpty ISO uses an amplifier (analog gain) to increase the brightness of the data in the raw file. It has only one possible benefit: reduction in quantization error. It has several negative effects, similar to Idiot ISO: loss of data (e.g. clipped highlights) and larger files.

Bargain ISO also uses an amplifier (analog gain) to increase the brightness of the data in the raw file. The only difference is that it results in less noise than using math to do the same thing (digital gain). The reason for this is that some of the noise comes from downstream electronics, such as the ADC (Analog-to-Digital Converter). By amplifying the signal before it gets to those noisy electronics, they are not able to contribute as much noise.

The disadvantages of Bargain ISO are similar to Idiot gain and Numpty gain: loss of data (e.g. clipped highlights) and larger files. But it has a significant advantage: less noise than other methods. This does not always make it the right choice, since it still clips one stop of highlights for every one stop increase in ISO, but at least it has a benefit.

Of course, it would be better if the manufacturers could fix the *real* problem (downstream electronics with high self-noise), but if they can't for whatever reason, this is at least a nice workaround. It allows the photographer to decide between noise and highlight headroom. With the introduction of the D7000, Sony seems to have mostly licked that problem, while Canon is still struggling with it, at least as of Summer 2011.

Smart ISO simply writes the desired change to metadata instead of the raw data itself. Then, the digital gain can be reversed or done differently in post production (such as with a nonlinear, highlight-preserving curve). This is the ideal type of ISO setting and how all of them should work. But until everyone else catches up with Sony's low-ISO read noise levels, Bargain ISO will still be needed.

Idiot and numpty ISO, however, are always braindamaged. They never have any useful purpose. If you shoot raw, then I recommend that you avoid them as much as possible. Choosing to use Bargain ISO or not is a little more difficult, since it's a trade off between a lot of clipped highlights and a slight reduction in noise. If I'm shooting a scene that needs a lot of highlight headroom, I'll often set ISO 800 or even 400 and increase brightness in post rather than use the Bargain ISO 1600, which clips 2 stops of highlight headroom compared to ISO 400 for a fixed exposure.

While we're on the topic, in non-raw cameras (e.g. JPEG-only digicams and most video cameras), there is another important distiction in the types of ISO possible:

• Linear ISO: clips highlights.
• Non-linear: compresses highlights without clipping.

Also, any time a lens reporting to be faster than f/2.8 is used, Canon applies a hidden Idiot ISO. This is Canon's way of falsifying (er, "compensating for") the sensor's poor angle of response.

Keep in mind that this thread is about the ISO *setting* on the camera (which is more properly called "gain" in my view). Other uses of the term ISO are more similar to the meaning of Exposure Index (a better use of the word ISO, IMHO), and are only indirectly related to this thread.

[strobe monkey]

I have set my camera to use only "real ISO's".

I've read somewhere that "fake" ISO's also introduce noise, hence my reason above.

[tonylong]

That's an interesting read Daniel. And, your names are pretty entertaining!

[uOpt]

This is sarcasm, right?

[Drozz119]

Interesting... I see a lot of people recommend ISO 160 over 100. How would you rate the top ISO's to use?

[Zivnuska]

Originally Posted by Daniel Browning

"Idiot and numpty ISO, however, are always brain damaged."

"There is never a reason or excuse for camera manufacturers to use such a brain damaged design"

Please consider the possibility that your use of intemperate adjectives will be an impediment to your argument.

[banquetbear]

Originally Posted by Daniel Browning

Every ISO setting on a digital camera can be classified using four different types. I recommend that you avoid Idiot and Numpty ISO settings whenever possible, and it's good to understand the trade-offs surrounding the use of Bargain ISO. The types are:

• Idiot ISO
• Numpty ISO
• Bargain ISO
• Smart ISO

Most ISO settings on digital cameras are made up of a mix of several types of ISO. For example, on my 5D2:

• ISO 50: Smart ISO
• ISO 125: Idiot ISO
• ISO 160: Bargain and Idiot ISO
• ISO 200: Bargain
• ISO 250+HTP: Bargain, Idiot, and Smart ISO
• [...]
• ISO 3200: Bargain and Numpty ISO
• ISO 4000: Bargain, Numpty and Idiot ISO
• ISO 5000+HTP: Bargain, Numpty, Idiot, and Smart ISO

The settings on other cameras may be made of different combinations (though tend to be largely similar for a given manufacturer). Here is a detailed example, showing how all four types combine to make ISO 5000 on the 5D2:

• +4 Bargain ISO to get up to 1600.
• +1 Numpty ISO to get from 1600 to 3200.
• -1/3 Idiot ISO to get from 3200 to 2500
• +1 Smart ISO to get from 2500 to 5000

The function of each of the types, briefly:

• Idiot ISO increases the brightness of the data in the raw file using digital gain (math inside the camera) and has no benefit whatsoever.
• Numpty ISO increases the brightness of the data in the raw file using analog gain and has no benefit whatsoever.
• Bargain ISO increases the brightness of the data in the raw file using analog gain, but has benefits/trade-offs.
• Smart ISO doesn't touch the raw file at all, it just changes metadata (similar to setting white balance).

Now a more detailed description of what each of the different types of ISO means.

Idiot ISO is when the brightness of the raw file is increased using math inside the camera. This is similar to using the "EC" slider in your raw converter, except that the changes are written permanently to the raw file. This causes the needless loss of data (clipped highlights or shadows), larger files, and increased quantization error (for non-integer factors).

There is never a reason or excuse for camera manufacturers to use such a braindamaged design when it is in fact *easier* to use Smart ISO than Idiot ISO. This type of ISO implementation has no redeemeing value whatsoever, hence the name.

The next two types of ISO are similar in that they both use an amplifier (analog gain) to increase the brightness of the data in the raw file. The only difference is this: if the resulting amplification has a better Signal-to-Noise ratio (SNR) than Idiot ISO, then I call it "Bargain ISO". If it has the same SNR, then I call it Numpty ISO.

Numpty ISO uses an amplifier (analog gain) to increase the brightness of the data in the raw file. It has only one possible benefit: reduction in quantization error. It has several negative effects, similar to Idiot ISO: loss of data (e.g. clipped highlights) and larger files.

Bargain ISO also uses an amplifier (analog gain) to increase the brightness of the data in the raw file. The only difference is that it results in less noise than using math to do the same thing (digital gain). The reason for this is that some of the noise comes from downstream electronics, such as the ADC (Analog-to-Digital Converter). By amplifying the signal before it gets to those noisy electronics, they are not able to contribute as much noise.

The disadvantages of Bargain ISO are similar to Idiot gain and Numpty gain: loss of data (e.g. clipped highlights) and larger files. But it has a significant advantage: less noise than other methods. This does not always make it the right choice, since it still clips one stop of highlights for every one stop increase in ISO, but at least it has a benefit.

Of course, it would be better if the manufacturers could fix the *real* problem (downstream electronics with high self-noise), but if they can't for whatever reason, this is at least a nice workaround. It allows the photographer to decide between noise and highlight headroom. With the introduction of the D7000, Sony seems to have mostly licked that problem, while Canon is still struggling with it, at least as of Summer 2011.

Smart ISO simply writes the desired change to metadata instead of the raw data itself. Then, the digital gain can be reversed or done differently in post production (such as with a nonlinear, highlight-preserving curve). This is the ideal type of ISO setting and how all of them should work. But until everyone else catches up with Sony's low-ISO read noise levels, Bargain ISO will still be needed.

Idiot and numpty ISO, however, are always braindamaged. They never have any useful purpose. If you shoot raw, then I recommend that you avoid them as much as possible. Choosing to use Bargain ISO or not is a little more difficult, since it's a trade off between a lot of clipped highlights and a slight reduction in noise. If I'm shooting a scene that needs a lot of highlight headroom, I'll often set ISO 800 or even 400 and increase brightness in post rather than use the Bargain ISO 1600, which clips 2 stops of highlight headroom compared to ISO 400 for a fixed exposure.

While we're on the topic, in non-raw cameras (e.g. JPEG-only digicams and most video cameras), there is another important distiction in the types of ISO possible:

• Linear ISO: clips highlights.
• Non-linear: compresses highlights without clipping.

Also, any time a lens reporting to be faster than f/2.8 is used, Canon applies a hidden Idiot ISO. This is Canon's way of falsifying (er, "compensating for") the sensor's poor angle of response.

Keep in mind that this thread is about the ISO *setting* on the camera (which is more properly called "gain" in my view). Other uses of the term ISO are more similar to the meaning of Exposure Index (a better use of the word ISO, IMHO), and are only indirectly related to this thread.

...while interesting from a technical point of view: how does a photographer apply this in the real world?

I've just googled the phrases Idiot and Numpty ISO, and I've found a whole lot of posts by you over the internet, so to be honest it is really hard to do any further research on what you have posted. Do you have any cites? Any references for anyone interested in finding out more?

How did you determine the Numpty settings of the 5D? How does one find out the numpty settings on any camera? What real world difference does it make to the average photographer if they use ISO 50 to ISO 160 in image quality?

[uOpt]

It doesn't.

[Daniel Browning]

Originally Posted by tonylong

That's an interesting read Daniel. And, your names are pretty entertaining!

Thanks.

Originally Posted by uOpt

This is sarcasm, right?

No.

Originally Posted by Drozz119

Interesting... I see a lot of people recommend ISO 160 over 100. How would you rate the top ISO's to use?

For Canon, how I would put it is this: ETTR then ITTR (ISO to the right), avoid the +1/3 ISO (125, 250, 500, etc.), but the -1/3 ISO (160, 320, etc.) are fine and don't go over 1600 (on most Canons) -- use negative EC instead.

Originally Posted by Zivnuska

Please consider the possibility that your use of intemperate adjectives will be an impediment to your argument.

Thanks for the feedback, I'll definitely take that under consideration.

Originally Posted by banquetbear

...while interesting from a technical point of view: how does a photographer apply this in the real world?

Two primary ways: first, avoid the negative effects of idiot+numpty ISO. Second, be aware of the noise/headroom trade-off of the Bargain ISO. (That is, be aware of times when it's better to shoot ISO 800 and push to 3200, rather than shoot 3200 and lose 2 stops of headroom.)

Originally Posted by banquetbear

I've just googled the phrases Idiot and Numpty ISO, and I've found a whole lot of posts by you over the internet, so to be honest it is really hard to do any further research on what you have posted. Do you have any cites? Any references for anyone interested in finding out more?

Sure. I think the best one is this:

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

And you can get a lot of useful data about your particular model (probably) here:

http://www.sensorgen.info/

Originally Posted by banquetbear

How does one find out the numpty settings on any camera?

In my testing I used IRIS to compare the SNR (std. dev.) of a fixed exposure at varying ISO settings, as well as a visual (subjective) assessment of pattern noise to determine that all ISOs up to and including 1600 (on my 5D2) were analog and useful (reduced downtream noise contribution), and hence "Bargain ISO", while 3200 (for me) did not, so 3200 is numpty. An easier program to do that comparison is Rawnalyze.

Originally Posted by banquetbear

What real world difference does it make to the average photographer if they use ISO 50 to ISO 160 in image quality?

Sorry, I don't understand the question. In any case, no one can really tell you how important something is, because we all have different standards. For some, a small difference is easily noticeable, while others require a much larger difference before they can see it.

For example, do you ever shoot ISO 6400 (which uses a lot of idiot ISO ) and get clipped highlights? If so, you could use 1600 instead, push to 6400 in post, and get the exact same noise but 2 stops less highlight clipping.

BTW, I'll be out of town for a big video shoot this weekend, so I may not make it back to the thread until Sunday.

[radekg]

So you're saying that it's better to take a picture with lower ISO setting and use exposure compensation in software (eg Lightroom) instead? But isn't it that without the analog gain you lose some data?
Photography is just my hobby, I come from a sound recording background. In this respect it has some similarities, as we convert an analog signal to digital. In sound recording you do have to use analog gain in order to make the signal level appropriate for the analog-digital converter. The digital dynamic range is always finite (eg 16bit - 90dB). Digital (mathematical) gain is always worse than good analog gain (if using good, low-noise components). The thing is that there is always some minimal signal level for the ADC and if it's lower than that it won't be converted. Data will be lost. Analog gain amplifies existing signal, digital gain has to make up the data that doesn't exist (using some alogrithms to predict what "should" be there). Of course with analog gain we also amplify the background noise and you have to decide what's better - having the data + noise or no noise and no data.

[Daniel Browning]

Originally Posted by radekg

So you're saying that it's better to take a picture with lower ISO setting and use exposure compensation in software (eg Lightroom) instead?

When the increased headroom is worth more than the noise benefit, yes.

Originally Posted by radekg

But isn't it that without the analog gain you lose some data?

In all but the numpty ones, yes, you lose some data. How much depends on the camera.

Originally Posted by radekg

Photography is just my hobby, I come from a sound recording background.

Ah, this will be a piece of cake for you. Let's say you're trying for peaks at around -20db to give you a good amount of headroom. Idiot gain is like a +12db digital gain applied in the audio ADC. No audio manufacturer in their right mind would do that because it reduces your 20db of headroom down to 8db.

Bargain gain is when your microphone and pre-amp have -80db noise, but the ADC has -40db self-noise. If you record with peaks at -20, you only have from -20 to -40 before you hit the noise floor. Ideally it would be better to just replace the ADC with a better one so you can get the full -80db dynamic range, but in the mean time, you can turn up the pre-amp analog gain to improve the SNR so that peaks are -6. The cost of course is less headroom, so if it goes louder than you expect then you're going to hit -0 and clip.

Same thing with photography. On most Canons, by the time you get to ISO 800, the ADC is no longer contributing very much noise to the image. Most of the noise is now "room sound" (upstream noises such as photon shot noise and read noise). You can up the gain another stop (ISO 1600), but it will only reduce the contribution from the ADC slightly (but noticeably). For me, in many cases, I'd rather keep the headroom. By the time you get to ISO 3200, the benefit was invisible (to me).

[radekg]

Ah, so it's ok to use the so-called "bargain ISO" (analog gain) but it's not worth using any other "types" of ISO which use digital gain. Thus, we should get a better quality picture on ISO200 (bargain) than ISO160 (bargain+idiot). However, is it noticeable? Do you see any differences between pictures taken at iso160 vs iso200?
You say that you often set iso400 instead of iso1600. That's all very well if you've got enough light but when I use iso1600 there usually isn't a lot of light. So let's say I'm doing 2 shots, one: 1/50s, f/2.8, iso1600, properly exposed, and the second: 1/50s, f/2.8, iso400, heavily underexposed. I can't go longer than 1/50 on iso400 because I'm not using a tripod. You're saying that the underexposed shot + EC in post-process (lightroom) will yield better results than the iso1600 properly exposed shot?
I don't usually use high isos because I want to but because I have to as there's not enough light...
I'll have some time tommorow then I'll do some comparison shots

[Mil]

Originally Posted by radekg

Ah, so it's ok to use the so-called "bargain ISO" (analog gain) but it's not worth using any other "types" of ISO which use digital gain. Thus, we should get a better quality picture on ISO200 (bargain) than ISO160 (bargain+idiot). However, is it noticeable? Do you see any differences between pictures taken at iso160 vs iso200?
You say that you often set iso400 instead of iso1600. That's all very well if you've got enough light but when I use iso1600 there usually isn't a lot of light. So let's say I'm doing 2 shots, one: 1/50s, f/2.8, iso1600, properly exposed, and the second: 1/50s, f/2.8, iso400, heavily underexposed. I can't go longer than 1/50 on iso400 because I'm not using a tripod. You're saying that the underexposed shot + EC in post-process (lightroom) will yield better results than the iso1600 properly exposed shot?
I don't usually use high isos because I want to but because I have to as there's not enough light...
I'll have some time tommorow then I'll do some comparison shots

Not to mentioned moving subject...

[TeamSpeed]

Originally Posted by uOpt

It doesn't.

Daniel is explaining more accurately and more interestingly the way the ISO "performance" or ISO output is obtained with Canon cameras than what you had tried to state in some of your other posts about how you think ISO works.

Daniel (and another gentleman that has passed away some time ago) are very involved in raw analysis, as is John Sheehy, and a couple of others, so their input is very valuable. Please don't knock what you don't understand.

It is always good to understand the tech we hold in our hands and how to best set it up for success, especially if all we are going to do later is pixel peep and compare notes with others.

[S.Horton]

Interesting. I would bet if I tested this in the real world with real subjects I might be hard pressed to notice the difference. But if I used a bad setting for an actual shoot I would never know what I was missing except I would clip more.

Why do the manufacturers create the 'idiot' settings at all? Surely they know this.