Can anyone explain ISO invariance to me?

StewartR

StewartR

Suspended / Banned
Messages
11,513
Name
Stewart
Edit My Images
Yes
... Or at least provide a link to a good explanation online?

I've been trying to get my head around it, but most of the "explanations" I've found seem to have been written by people who don't understand it themselves.
 
ISO invariance basically means the camera itself has a very low noise floor and adds very little noise to the signal chain.
What this means is, because the data is "clean" it can be amplified in post (exposure recovery) or amplified in camera (high ISO) with equal quality results (not better results).

A non ISO invariant camera has a higher noise floor. Which means when the received signal is low (underexposure/shadows) the camera noise makes up a greater portion of the recorded data. This causes the results from pushing in post to be worse (noise/color) than what the camera would produce itself by using a higher ISO initially (to amplify the signal higher above the noise floor before the ADC)... some of this comes down to the software, OEM software can often do a better job here than 3rd party does. And it is likely to improve as software gets better.
 
Last edited:
ISO invariant cameras typically have a greater usable DR; because they need less signal/light in order for it to be readable above the lower noise floor. But (IMO) the main advantage of an ISO invariant sensor is the ability to selectively adjust exposures in post, rather than globally with ISO in camera.

Edit: the "greater DR" is only utilized when a lower ISO is used and portions of the sensor are fully saturated (min-max recorded). It does not exist if all portions of the sensor are underexposed. This is also true of all other sensor characteristics (i.e. color).
 
Last edited:
With an Iso invariant sensor one can shoot at virtually any shutter speed and aperture and simply adjust the exposure in post processing and see virtually no difference in quality.
I have used a lot of virtuals in that sentence because there are of course limits, but they are very wide.
Only a few cameras rely on amplifying the signal now, but many do remove excess noise as part of the firmware process.

One can think of invariance as being able to capture an extremely wide tonal range in a single exposure, from near total darkness to extreme brilliance.
Then choosing the segment of that range, needed to produce the best image, at a later stage.
This implies little to low noise at the shadow region and no clipping at the highlight end.

Such a scenario makes the concept of ISO meaningless.
 
It just occurred to me that it is possible for a camera to be very poor at applying ISO natively, and for software to be able to amplify the signal with equal quality... that would technically make that sensor also ISO invariant. But instead it is generally considered to just be a very poor performer.
 
sk66 said:
??? ISO is not merely an exif header setting (to be applied by software)... the amplification is real and it does change the data recorded. I am unaware of any camera where this is not the case.
Click to expand...

All sensors have a native ISO. they do not suddenly become more sensitive when you up the ISO. The firmware in the camera changes the way the captured data is handled when you change the ISO.setting.
The shutter speed and aperture deliver more of less light to the sensor.
The ISO setting does not change the sensitivity of the sensor.
 
Terrywoodenpic said:
All sensors have a native ISO. they do not suddenly become more sensitive when you up the ISO. The firmware in the camera changes the way the captured data is handled when you change the ISO.setting.
Click to expand...
Not quite, but the firmware can... that's the difference between analog ISO settings (amplified signal), and extended/digital ISO settings (firmware). But often you may not know where that switch occurs (with Nikons it used to be the Hi and Low settings, but now they don't necessarily tell you).

This chart is for the D850... the solid circles are pure analog amplification, the solid triangles indicate some additional firmware(digital) scaling, and the open symbols are pure firmware (digital).

Edit: but yes, ISO does not change the amount of light/data received, nor the sensitivity of the sensor... you are correct in that aspect.
 
Last edited:
SsSsSsSsSnake said:
So which cameras are iso invariant?
Click to expand...
Most of the current Nikons, the Canon 80D is supposedly close, Sony A7r series, several Fuji X models, K5... IDK of any up to date list.

ISO invariance is only one characteristic of a sensor. ISO invariance doesn't necessarily mean that the other characteristics are particularly good (i.e. DR/color accuracy/etc). But it is only of real use when they are...
 
Last edited:
The 80D has much better DR than it's big brother, the 6DmkII, but only up to 400 ISO. I don't really understand how all this works but I see this pushed a lot when I look at 6DII reviews.
 
Try this Stewart (based on my undertsnading which may be flawed)......

On a non-invariant sensor, increasing the ISO setting increases the sensor's analogue gain giving increased sensitivity but with a corresponding increase in noise.

An invariant sensor does not change the gain at the analogue stage but merely amplifies (multiplies) the recorded signal level after it has been converted to a digital value.....as would the brightness slider in PP

Bob
 
Canon Bob said:
An invariant sensor does not change the gain at the analogue stage but merely amplifies (multiplies) the recorded signal level after it has been converted to a digital value.....as would the brightness slider in PP
Click to expand...
No, analog gain is still applied for all "native" (analog) ISO levels. See the link in post #8 for the D850, one of the most ISO invariant cameras currently available...
 
Last edited:
I didn't think this was so confusing... let me try to simplify.

ISO invariance is a measure of the noise/noise floor of the camera itself. Primarily at native/base ISO (min saturation/signal strength) where there is minimal amplification being applied. That's all...

When this is very low, the signal/recorded data is very "clean" and software can manipulate it with fewer errors (ISO invariant). However, when the noise levels are higher the recorded data is more "contaminated" and software does a poorer job of manipulating it as a result. And if the majority of the noise is added after the photosite readout, then amplifying the signal farther will result in a greater SNR in the results (occurs after readout/prior to the ADC). I.e. the camera applies ISO (analog gain) better/non-invariant.

However, since the determination of ISO invariance is a comparison between how well the camera applies ISO and the results w/ external software, it is inherently also a measure of how good the software is at manipulating the data.
 
So ISO invariance means, for example, that you can shoot at ISO 400, increase exposure by 3 stops in post and this would give the same result as shooting at ISO 3200 in terms of noise? Would you get better or worse DR and colour rendition if adjusting in post? On the D850 the DR at 400 is approx 13ev, yet at 3200 it's around 10.2ev. My brain tells me that if I shoot at 400 ISO and then increase by 3 stops in post I'll still have 13ev rather than 10.2ev, but I'm sure it's not that simple?
 
snerkler said:
So ISO invariance means, for example, that you can shoot at ISO 400, increase exposure by 3 stops in post and this would give the same result as shooting at ISO 3200 in terms of noise?
Click to expand...
Yes.
snerkler said:
On the D850 the DR at 400 is approx 13ev, yet at 3200 it's around 10.2ev.
Click to expand...
These measures are based on a "proper exposure" of the same test scene. And ISO 3200 results in less DR because the sensor receives less total light (less saturation at the high end, min is always the same). So the answer to your question depends on the scene and how the ISO invariance is used... if you use ISO 400 you will have recorded 13stops DR *if* at least some photosites reach saturation. But if you use ISO 400 and significantly underexpose everything, then you will have something less than 13stops of DR.

This is the best article I could find... still a bit more convoluted than it needs to be.
 
sk66 said:
Yes.

These measures are based on a "proper exposure" of the same test scene. And ISO 3200 results in less DR because the sensor receives less total light (less saturation at the high end, min is always the same). So the answer to your question depends on the scene and how the ISO invariance is used... if you use ISO 400 you will have recorded 13stops DR *if* at least some photosites reach saturation. But if you use ISO 400 and significantly underexpose everything, then you will have something less than 13stops of DR.

This is the best article I could find... still a bit more convoluted than it needs to be.
Click to expand...
Thanks, that makes sense (y)
 
snerkler said:
Thanks, that makes sense (y)
Click to expand...
I should clarify...
If the highlights in a scene requires ISO 3200 combined with the Ap/SS for proper exposure (rt side of histogram @ saturation) and that equates to 10stops of DR and 10bit color, then that is the most you will get. Regardless of the camera, you are best off using ISO 3200. You will get "equivalent" results with an ISO invariant camera, and "better" results with an ISO variant camera.

Using a lower ISO with an invariant camera will only make image view/review more difficult and require more extensive editing. And while it will probably not decrease the total DR recorded, it may cause more of the scene to drop to/below minimum readable, and that can also reduce the color accuracy. ISO invariance is not a free ride... image quality is still determined by the total amount of light recorded/readable (Ap/SS). The only real advantage is the ability to expose to retain highlights and selectively recover the underexposure of the remainder of the scene which is not below minimum (which can be amazingly low/dark on the best invariant cameras).
 
Last edited:
I think it's helpful to remember that actual 'exposure' to light (total photon capture) only involves shutter speed and lens aperture. ISO makes no difference to that - it just amplifies the signal and adjusts brightness to the right level.

Film has a fixed sensitivity, indicated by the ISO rating, and it cannot be changed. Digital also has a fixed sensitivity - the native or base ISO - but the effective sensitivity can be changed by adjusting the gain applied (amplification).

Traditionally, applying gain to the analogue signal at sensor level produces much better results than doing it in post-processing software after the analogue-to-digital conversion has taken place in-camera (as Steven explains). With a truly 'ISO-invariant' or 'ISO-less' sensor, it would make no difference whether the gain was applied in-camera with ISO adjustment, or afterwards with post-processing software - the quality of the result (basically in terms of shadow detail and noise) would be the same. No camera is currently 100% ISO-invariant, though some do seem to get pretty close.

There are several practical applications of this, and they can be very beneficial, but it does not mean that shutter speed and aperture settings no longer matter. They do, and ultimately the quality of the result is always directly related to total photon capture.
 
Cagey75 said:
The 80D has much better DR than it's big brother, the 6DmkII, but only up to 400 ISO. I don't really understand how all this works but I see this pushed a lot when I look at 6DII reviews.
Click to expand...


I'll just google it :D
 
That’s true in a lot of things like food exercise or most personal stuff,nobody can know yourself like you .

But just incase,this might help


The more sophisticated design of the 80D's sensor means it adds less noise to its images than its big brother. This means that, at low ISO settings, the 80D will produce more flexible Raw files, that make it easier to represent the detail in high-contrast scenes, before you hit the noise floor. If you've become used to exploiting the 80D's pretty impressive dynamic range, it may be a bit of a shock to find you end up with more prominent noise if you try to manipulate an image shot in high-DR circumstances, such as sunsets or backlit subjects.

That said, we're aware that a great many people primarily shoot JPEG. Since the differences in performance between the two cameras' sensors tends to occur in very dark tones within the image, so may well be either too dark to perceive or clipped entirely to black if you're only looking at JPEG images. Even engaging Auto Lighting Optimizer or Highlight Tone Priority - the camera's two DR compression modes that risk pulling noise into the image - isn't a problem (though it's interesting you can't use the two in conjunction). However, you don't get the noise improvement at low ISO you might reasonably expect from the move to full-frame.
 
SsSsSsSsSnake said:
That’s true in a lot of things like food exercise or most personal stuff,nobody can know yourself like you .

But just incase,this might help


The more sophisticated design of the 80D's sensor means it adds less noise to its images than its big brother. This means that, at low ISO settings, the 80D will produce more flexible Raw files, that make it easier to represent the detail in high-contrast scenes, before you hit the noise floor. If you've become used to exploiting the 80D's pretty impressive dynamic range, it may be a bit of a shock to find you end up with more prominent noise if you try to manipulate an image shot in high-DR circumstances, such as sunsets or backlit subjects.

That said, we're aware that a great many people primarily shoot JPEG. Since the differences in performance between the two cameras' sensors tends to occur in very dark tones within the image, so may well be either too dark to perceive or clipped entirely to black if you're only looking at JPEG images. Even engaging Auto Lighting Optimizer or Highlight Tone Priority - the camera's two DR compression modes that risk pulling noise into the image - isn't a problem (though it's interesting you can't use the two in conjunction). However, you don't get the noise improvement at low ISO you might reasonably expect from the move to full-frame.
Click to expand...


Now that, makes sense. I was looking at the 80D, and watched a bunch of reviews where it was compared to the 6DII, and this stood out. As I tend to always under expose a little [if I'm in any semi-auto mode I set my exp comp to -2/3 or so] - it's a habit that came from a couple cameras I had that tended to over expose no matter the settings. My thinking was a camera like the 80D would be ideal for my style, I don't like to go high on ISO, I tend to UE, never blow anything out, and pull up in post, so should benefit from it's impressive DR.
 
HoppyUK said:
I think it's helpful to remember that actual 'exposure' to light (total photon capture) only involves shutter speed and lens aperture. ISO makes no difference to that - it just amplifies the signal and adjusts brightness to the right level.

Film has a fixed sensitivity, indicated by the ISO rating, and it cannot be changed. Digital also has a fixed sensitivity - the native or base ISO - but the effective sensitivity can be changed by adjusting the gain applied (amplification).

Traditionally, applying gain to the analogue signal at sensor level produces much better results than doing it in post-processing software after the analogue-to-digital conversion has taken place in-camera (as Steven explains). With a truly 'ISO-invariant' or 'ISO-less' sensor, it would make no difference whether the gain was applied in-camera with ISO adjustment, or afterwards with post-processing software - the quality of the result (basically in terms of shadow detail and noise) would be the same. No camera is currently 100% ISO-invariant, though some do seem to get pretty close.

There are several practical applications of this, and they can be very beneficial, but it does not mean that shutter speed and aperture settings no longer matter. They do, and ultimately the quality of the result is always directly related to total photon capture.
Click to expand...

A bit more on the practical applications of ISO-invariance.

In these debates, it's quite common for some to argue that with an ISO-invariant sensor, exposure settings don't matter - just shoot at base ISO with whatever shutter speed and aperture you want for creative purposes and adjust brightness in post-processing - based on the fact that photon capture will be the same whether whether ISO is adjusted or not. That may be true in very rare instances where base ISO also happens to be right for normal exposure setting purposes, and there's certainly more leeway, but given that best exposure always depends on maximum possible photon capture regardless, the only way of optimising that is to adjust exposure settings in the normal way so you know where you are. With that done, an ISO-invariant sensor may allow you to make further beneficial adjustments.

ISO-invariance changes the popular method of optimising digital exposure known at ETTR - Expose To The Right (of the histogram). This basically means over-exposing to put more detail (photons) into the shadows, then darkening the image back down again in post-processing but with shadow detail retained. It works well but there is a downside because over-exposure will blow highlights so a careful judgement must be made to make sure only unimportant highlights are effected - basically, it's a compromise.

With ISO-invariance, that technique can be changed to ETTL (Expose To The Left) which ensures that all highlights are retained and shadows can be lifted in post with little or no penalty. In some situations with very high dynamic range where you might otherwise need multi-exposure HDR technique, it can make a big difference, eg most back-lit situations where you want detail in the very bright background but also good detail on the shaded foreground. Unlike HDR, ISO-invariance also works with moving subjects as everything is captured in a single shot and since it often means you can use a faster shutter speed, that's another bonus.

A practical downside of ETTL and deliberate under-exposure is that the LCD image goes dark, sometimes very dark to the point of useless, and the histogram will also reflect that.

Exploiting the full potential of ISO-invariance only works when shooting Raw. It's not much use with JPEGs where tonal values are pretty much locked with very little scope for adjustment in post.
 
HoppyUK said:
...... but given that best exposure always depends on maximum possible photon capture regardless.........,
Click to expand...

About time someone said that in this thread!!

Perfect ISO Invariance doesn't exists, and neither does smooth linear analog sensor gain, but careful use of either/both can help, but getting the exposure right in the first place will bring the most benefits.
 
Cagey75 said:
The 80D has much better DR than it's big brother, the 6DmkII, but only up to 400 ISO. I don't really understand how all this works but I see this pushed a lot when I look at 6DII reviews.
Click to expand...
What you probably don't see being pushed is that the original 6D is reputedly around 'half a stop' lower in noise than the 6D Mk 2 at higher ISOs, which is another reason I've still got my 6D.
 
Mr Perceptive said:
About time someone said that in this thread!!

Perfect ISO Invariance doesn't exists, and neither does smooth linear analog sensor gain, but careful use of either/both can help, but getting the exposure right in the first place will bring the most benefits.
Click to expand...

how do you define " getting exposure right"
I would tend to define it as capturing "all the needed and significant tones in an image".
This is quite different to what a camera or exposure meter normally does unaided.
nor does it necessarily produce a pleasing result directly out of the camera.
 
Mr Perceptive said:
About time someone said that in this thread!!

Perfect ISO Invariance doesn't exists, and neither does smooth linear analog sensor gain, but careful use of either/both can help, but getting the exposure right in the first place will bring the most benefits.
Click to expand...
One of the many benefits of mirrorless technology with WYSIWYG EVF's :D
 
HoppyUK said:
A bit more on the practical applications of ISO-invariance.

In these debates, it's quite common for some to argue that with an ISO-invariant sensor, exposure settings don't matter - just shoot at base ISO with whatever shutter speed and aperture you want for creative purposes and adjust brightness in post-processing - based on the fact that photon capture will be the same whether whether ISO is adjusted or not. That may be true in very rare instances where base ISO also happens to be right for normal exposure setting purposes, and there's certainly more leeway, but given that best exposure always depends on maximum possible photon capture regardless, the only way of optimising that is to adjust exposure settings in the normal way so you know where you are. With that done, an ISO-invariant sensor may allow you to make further beneficial adjustments.

ISO-invariance changes the popular method of optimising digital exposure known at ETTR - Expose To The Right (of the histogram). This basically means over-exposing to put more detail (photons) into the shadows, then darkening the image back down again in post-processing but with shadow detail retained. It works well but there is a downside because over-exposure will blow highlights so a careful judgement must be made to make sure only unimportant highlights are effected - basically, it's a compromise.

With ISO-invariance, that technique can be changed to ETTL (Expose To The Left) which ensures that all highlights are retained and shadows can be lifted in post with little or no penalty. In some situations with very high dynamic range where you might otherwise need multi-exposure HDR technique, it can make a big difference, eg most back-lit situations where you want detail in the very bright background but also good detail on the shaded foreground. Unlike HDR, ISO-invariance also works with moving subjects as everything is captured in a single shot and since it often means you can use a faster shutter speed, that's another bonus.

A practical downside of ETTL and deliberate under-exposure is that the LCD image goes dark, sometimes very dark to the point of useless, and the histogram will also reflect that.

Exploiting the full potential of ISO-invariance only works when shooting Raw. It's not much use with JPEGs where tonal values are pretty much locked with very little scope for adjustment in post.
Click to expand...

I only ever shoot RAW and pretty much ettl slightly, it works even though I'm constantly told to ettr, when I shoot using constant view, one of the many benefits of mirrorless, I see what I will get and I don't push it far enough to need to pump it tonnes in post. With a DSLR ovf I'd have to chimp a bit more I guess?
 
Mr Badger said:
What you probably don't see being pushed is that the original 6D is reputedly around 'half a stop' lower in noise than the 6D Mk 2 at higher ISOs, which is another reason I've still got my 6D.
Click to expand...

I have seen that too, and like the look of that camera for the price, pity it doesn't have even the standard (these days) tilt LCD that I use a lot for macro
 
Terrywoodenpic said:
how do you define " getting exposure right"
I would tend to define it as capturing "all the needed and significant tones in an image".
This is quite different to what a camera or exposure meter normally does unaided.
nor does it necessarily produce a pleasing result directly out of the camera.
Click to expand...

I'd agree with your definition, taking into account the limitations of the camera sensor (especially with regard to highlights, something my current Fuji's don't recover very well!!)
 
Terrywoodenpic said:
how do you define " getting exposure right"
I would tend to define it as capturing "all the needed and significant tones in an image".
This is quite different to what a camera or exposure meter normally does unaided.
nor does it necessarily produce a pleasing result directly out of the camera.
Click to expand...

'Correct' exposure, in theory, is when mid-tones in the subject exactly match mid-tones recorded by the sensor and are output as mid-tones without adjustment in post-processing. This is what the meter tries to do and will deliver good looking JPEGs straight out of the camera. But 'best' or 'optimum' exposure may be very different and can be anything you want it to be with a good understanding of how it all works. And all cameras vary with what's possible and how far you can push things. ISO-invariant cameras are new and only a handful really fit the bill (mainly Sony and Nikon).

Cagey75 said:
I only ever shoot RAW and pretty much ettl slightly, it works even though I'm constantly told to ettr, when I shoot using constant view, one of the many benefits of mirrorless, I see what I will get and I don't push it far enough to need to pump it tonnes in post. With a DSLR ovf I'd have to chimp a bit more I guess?
Click to expand...

But mirrorless is not inherently WYSIWYG when you start to play about with ETTR and ETTL. What you're seeing through the EVF or rear LCD is a JPEG generated in-camera, so it can't show what you'll get after adjusting the Raw file in post-processing. Cameras vary in how they adjust image brightness for viewing with exposure simulation etc and the histogram may or may not be helpful.

It's an interesting dilemma. No current camera is designed to be operated using these techniques, ISO-invariant or not, but given that it's likely to become more popular and even commonplace in the future, camera manufacturers might want to address it, at least with enthusiast-oriented models.
 
Last edited:
HoppyUK said:
But mirrorless is not inherently WYSIWYG when you start to play about with ETTR and ETTL. What you're seeing through the EVF or rear LCD is a JPEG generated in-camera, so it can't show what you'll get after adjusting the Raw file in post-processing. Cameras vary in how they adjust image brightness for viewing with exposure simulation etc and the histogram will certainly be unhelpful.

It's an interesting dilemma. No current camera is designed to be operated using these techniques, ISO-invariant or not, but given that it's likely to become more popular and even commonplace in the future, camera manufacturers might want to address it, at least with enthusiast-oriented models.
Click to expand...

It's still an aid though, if you can see details in the highlights on cam, you at least know it's there to be brought back, just a quick initial adjustment in LR to get where you were on camera to start
 
Last edited:
I've found with the 2 Canon camera's I've got (50d and 80d) that a ⅓ stop to the right works best for me, well most of the time (but that could just be me).
 
HoppyUK said:
'Correct' exposure, in theory, is when mid-tones in the subject exactly match mid-tones recorded by the sensor and are output as mid-tones without adjustment in post-processing. This is what the meter tries to do and will deliver good looking JPEGs straight out of the camera. But 'best' or 'optimum' exposure may be very different and can be anything you want it to be with a good understanding of how it all works. And all cameras vary with what's possible and how far you can push things. ISO-invariant cameras are new and only a handful really fit the bill (mainly Sony and Nikon).

.
Click to expand...
.
That was certainly true for incident light meters. which peg the tone extremely accurately.
However digital cameras, and raw files in particular, can often give a better starting point and final result when exposed either to the left or right of this point to avoid clipping.
In camera Jpegs, though usually near enough, give less opportunity to take advantage of the entire tonal range captured by the sensor, especially in contrasty conditions.
Most photographers are well aware of this, and take such things into account when setting the exposure.

Most digital cameras are for all intents and purposes invariant over three or four stops. That is certainly true in practice for most Fuji X cameras.
 
Last edited:
Cagey75 said:
The only way around is to quick adjust to as near as possible to your on cam settings upon import to LR maybe?
Click to expand...

There are ways around it - basically the fact that you've adjusted exposure away from what the camera thinks it should be and will either show an image that's too bright on the LCD (ETTR) or too dark (ETTL).

I've just got used to this with DSLRs and ETTR (I don't have any ISO-invariant cameras) and know where I am with the LCD image, assisted by the histogram and especially blinkies. I've used a couple of ISO-invariant cameras though and it's pretty amazing what can be achieved, but in realty what you're doing is maximising dynamic range. This is great, no doubt about that IMHO, but there are limits to what's either necessary or desirable. After all, we still want our shadows to be dark and if you push it too far, it just looks naff and unrealistic - like so much HDR does ;)

So, in practise, if you don't go mad and keep everything within a couple of stops of where the camera thinks it should be, you've still gained a big slug of extra exposure to play with, and I find the LCD is still perfectly usable for things like sharpness, focus, composition etc. I then use blinkies to tell me where the highlights threshold is and work around that.
 
You must log in or register to reply here.
Back
Top