smf85 Loc: Freeport, IL

There’s been a lot of talk about the phenomenon of ISO invariance posited to exist in certain cameras. ISO invariance implies that there is one and only one read voltage and amplification gain setting that is used to generate value for a specific pixel. That’s not the case in every data sheet for every sensor I could find. The sample was biased towards Sony but a smattering of others. Since I could actually build such a sensor I’m not going to globally state that it isn’t there and I’m not claiming my review was comprehensive.

The data sheets all talk about a large amount of programmable gain in the analogue amplification stage (there’s no reason for on sensor digital amplification as such). In the case of the IMX411ALR/AQR it talks about +36db gain available in the CDS/PGA and selectable A/D conversion bit depths. It’s Sony’s flagship prosumer 151MP sensor (matches the specs for Phase1 camera model). It’s not ISO invariant as delivered. More interestingly from what I’ve read it uses a combination of different read sensitivities (pixel well voltage) and amplification. That’s interesting. If it changes the read sensitivity its changing the absolute light sensitivity of the sensor itself. As always, the higher the voltage the more dark current and read noise. Not as much information is available for the IMX309 series used in Nikon’s D850 and Z7 but there seem to be at least two read sensitivities used plus analog amplification. So what does all this mean - its going to depend on the camera. It’s easy to take a ISO variant sensor and use it in an ISO invariant fashion or semi-invariant (Nikon D850/Z7?). Before I end - its possible to take a ISO variant setup and normalize the output data so that it seems ISO invariant.

E.L.. Shapiro Loc: Ottawa, Ontario Canada

The practical theory, as I understand it is to shoot everything at a fixed low ISO, not a variable setting and the underexposed images will be normalized in post-processing. The cameras with this technology are said to have low noise using this methodology, and only a little shadow detail will be lost.

The engineering is, unfortunately, way above my paygrade, however, I would be interested in the practical advantage s this system in terms of results, aesthetics and image quality of the final product. In my own work, often rendition of shad details important and anything that could diminish that could be problematic.

smf85 Loc: Freeport, IL

If certain conditions are in effect, meaning the camera was engineered in a specific way, then the practical theory holds. Examining real hardware I'm very doubtful the camera is engineered that way. The fun, fascinating, and frustrating part of digital imaging is that there isn't one way to do anything. Even what to normalize into the 'raw' image (which is anything but raw). Basically, its what the mfr. say's they've done. Only way to know. Working backwards from images taken at various ISO values with a 14bit A/D sensor probably won't reveal anything. Ultimately is becomes a question of the best way to map the image from the lens on the sensor to a digital description of that image. From an engineering standpoint you have a sensor capable of 'x' bit depth, a AD converter of 'y' bit depth and an output field of 'z' size. If x=y=z then you have a true invariant device as the images can be directly mapped. If x<y=z then you need to vary x to pick up the most image information - select the most significant bits. Thats a variant device. Its my current understanding that the situation with nearly all sensors the pattern is x<y=z.

Rongnongno Loc: FL

I posted this long ago when I received my D500 (invariant sensor test)

Ysarex Loc: St. Louis

ISO invariance is an unintended consequence. None of the camera manufacturers advertise it as a feature. The OP is correct in noting that from an engineering perspective it's not a "thing" and modern sensors aren't really ISO invariant. In fact three of my newest cameras seem to have gone the other way in that they all contain sensors with dual impedance read channels and yet they're also the most ISO invariant cameras I've ever had.

ISO invariance, from as you note a practical perspective given that most of us own cameras because we take photographs with them, is a by-product of the fact that our camera manufacturer's engineers have managed to basically engineer the read noise out of the systems. The better they do that the more ISO invariant our cameras become whether that's an intended consequence or not.

Noise in our photos comes from different sources but the two main sources are shot noise and what used to be read noise. Shot noise is in the signal itself and is the noise most of us see when we complain about noise. Read noise is added by the electronics -- it's in the sensor and supporting electronics and it's quickly becoming a non-issue from our "photographic" perspective. The gain applied to the analog signal that comes with increased ISO suppresses read noise and so using the camera's higher ISO settings was important to get best results. Notice how I had to write than sentence in the past tense.

So as photographers who use cameras whether it's deliberate engineering or not is moot. It's real. Here's a quick example and to do this I haven't grabbed one of my newest most ISO invariant cameras but rather my little Canon 1" sensor G7 compact that costs $400.00; it illustrates just how common and effective this has become. I put the camera on a tripod and metered the scene for ISO 6400 and took the photo with the camera set to ISO 6400 -- 1st photo below. Then I changed the ISO to 400 and took the photo again at the same exposure -- 2nd photo below. I saved raw files and processed both the same with the exception of the exposure variance as the ISO 400 photo is technically 4 stops underexposed. I turned all luminance noise filtering off. (Point worth noting: If I used modern noise filtering on these photos there'd be nothing here to see.)

The third photo below is a 100% side by side of the noise in the shadows between the first two photos. You're seeing the shot noise in each and it's the same -- it should be as it's in the signal and the same exposure is the same signal. There isn't enough read noise in that camera to spit at. With the same exposure the sensor recorded the same data. Now if I had tried that with a decade older camera like a Canon 5dIII or 60D the ISO 400 shot would be horrible by comparison with a lot of the shadow detail swamped in read noise. Nothing wrong with a 5dIII it's a great camera the point simply would be that if you need ISO 6400 you better set it and use it. With my little G7 I can process the ISO 400 shot to have the same noise and shadow detail as the ISO 6400 shot.

Given that the JPEG from the ISO 400 shot is worthless and not a very good indication of what you've captured and your exposure this can all seem like useless trivia real fast. Why for heaven's sake do that? Just use the appropriate ISO right? There is one other difference that can occasionally matter. With my G7 set to ISO 6400 I'm capturing at best a dynamic range of 4 stops. With it set to ISO 400 I'm capturing a DR of about 8 stops. That can be meaningful in some situations. So look at the 4th photo below which is again a 100% side by side comparison. Four stops of DR in the ISO 6400 shot wasn't enough to prevent highlight clipping in that lampshade. There's no data there in the raw file -- effectively it's ISO clipped off. But 8 stops of DR in the ISO 400 shot was plenty to capture that lampshade highlight. It's still an esoteric application but it can matter and in our newer cameras it's becoming increasingly real and available intended or not.

(Download)


(Download)


(Download)


(Download)

selmslie Loc: Fernandina Beach, FL, USA

The problem with that viewpoint is that, if there is not enough light falling in the subject, all of the data will get recorded as low raw values. You end up using only the lowest bits in the binary range when the camera's A/D converter is done.

The reason to increase the gain (by using a higher ISO) is to use more bits. You end up amplifying both the signal and the noise and the S/N ratio does not materially change but at least you don't introduce banding in the ranges where you actually have something to record.

The bottom line is that, to get an acceptable image, you need to increase the ISO in the camera or by moving the Exposure slider in your raw editor on the computer. Neither of these will change the S/N ratio or the visibility of noise. The only thing that is going to reduce the noise is an actual increase in exposure.

So for all practical purposes, ISO invariance is a theoretical exercise that does not actually have a practical impact on anyone's photography. You already have the camera, it takes pictures, you can reduce the noise by increasing the exposure. The question of whether or not the camera is invariant is immaterial.

Ysarex Loc: St. Louis

ISO invariance doesn't help with noise due to underexposure as you note but it does provide one practical advantage. Raising ISO via analog gain reduces sensor DR. Taking advantage of ISO invariance allows a photographer to avoid DR loss due to ISO clipping and so retain more recorded data. That certainly has had a practical, positive impact on my photography as I've taken advantage of it in the past.

selmslie Loc: Fernandina Beach, FL, USA

I understand highlight clipping in the raw file which is easily shown with RawDigger. It can be caused by too much exposure for the chosen ISO or an ISO that's too high for the chosen exposure.

I also know what it looks like when an otherwise healthy raw capture is converted to an image and the Exposure slider moved too far to the right.

But since I apparently don't have a camera that is not ISO invariant I can't see what "ISO clipping" means.

I tried to Google the term and ended up in an area that seems to have nothing to do with photography. It was difficult to determine what was being discussed.

As the OP pointed out, it's really not relevant whether the sensor is invariant. A sensor without a camera has no context. What matters is whether the camera makes it seem invariant.

A failure to perform in an invariant manner is not the fault of the sensor. It's simply because the camera manufacturer did not implement it correctly.

In ISO Invariance: What it is, and which cameras are ISO-less we see a simple definition of ISO invariance, "The short answer is that ISO invariance means that a camera will produce the exact same image quality by staying at ISO (or whatever the base ISO is on the camera) and dramatically underexposing the photo and then brightening it up again in Lightroom, as if you had shot the camera at the proper ISO in the first place." That's an oversimplification and we might take issues with "the exact same image quality" but it seems to be what we are discussing.

zenagain Loc: Pueblo CO

After reading this, i feel like im just a

"point and shoot monkey" to you all.

But really after reading all the posts and going to the site selmslie posted and reading more i do now understand and have learned something today.
Thanks for cramping up my brain.

Thank you for all the monkeyed down explanations. Not something i see myself running out and using but very interesting stuff.

I think i am changing my username to
"Pointnshootmonkey"

BebuLamar

I think you're right.

Ysarex Loc: St. Louis

And if the camera does seem to be or behave in an ISO invariant manner it provides a practical advantage that a non-ISO invariant camera does not. You said there's no pratical impact: "So for all practical purposes, ISO invariance is a theoretical exercise that does not actually have a practical impact on anyone's photography." That's wrong. A camera that behaves as if it were ISO invariant offers an advantage in that it allows the photographer to retain more sensor DR without paying a read noise penalty that would otherwise result from failure to raise the ISO with a camera that does not behave as ISO invariant. You just linked an article that says exactly that -- the same thing I said:

"Another reason why you'd want to shoot an image at base ISO and brighten later is when you are concerned about preserving highlight detail. Suppose you're in Time Square in New York and you want to take a shot of the colorful lights above the streets at night. If you shoot at a high ISO to properly expose, there isn't enough dynamic range to properly preserve highlights and still expose the shadows. However, if you do this after the fact, you could easily brighten the image without blowing out highlight detail in the bright signs."

You can see that in operation in the example I posted above.

selmslie Loc: Fernandina Beach, FL, USA

Not wrong.

What you are describing is the basis for exposing to the right (ETTR) and watching for blinkies so as to not blow the raw highlights.

Suppose you're in Time sic Square in New York and you want to take a shot of the colorful lights above the streets at night. If you shoot at a high ISO to properly expose, there isn't enough dynamic range to properly preserve highlights and still expose the shadows.

In this case you don't want to blow out the colorful lights so you expose to the right so as to not blow one or more of the raw channels. You can do this at any ISO setting. That has nothing to do with ISO invariance.

But we need to use a lower ISO in a case like this is to leave ourselves enough dynamic range (DR) so that the shadows can be recovered. Shooting at base ISO will maximize the DR. ETTR is normally accomplished by increasing the exposure. That has nothing to do with ISO invariance either.

Remember that ETTR was developed 18 years ago to help 12-bit cameras that were definitely not ISO invariant. It was a workaround to address both weaknesses.

But the Times Square example points to the flaw in much of what is written here about ETTR. Since ETTR is usually accomplished by reducing the shutter speed to increase exposure, it does not do much good unless you are also at a low or base ISO.

ETTR provides these benefits regardless of whether or not the camera is ISO invariant. It's just that newer 14-bit cameras do all of this better.

Ysarex Loc: St. Louis

No. That's just your usual smoke blowing sophistry. ETTR is limited in the same way by the ISO invariant/non-ISO invariant behavior of cameras. A practical advantage gained from a camera that behaves as if it were ISO invariant is that you can push ETTR beyond the limits that a non-ISO invariant camera can manage without paying a hefty read noise penalty.

cactuspic Loc: Dallas, TX

Is there and advantage to raising the ISO in a camera that is ISO invarient, if we have a very dimly scene that does not have a limited dynamic range...just different colors information of similar luminosity poorly lit (say red graffitti on a green wall), with a limit of how long I can handhold the camera. If I raise my ISO 7 stops from base, from 100 to 12800 would there be a difference in the image? In other words, would I be able to register an image at higher ISO that would otherwise be lost in the black at lower ISO?

selmslie Loc: Fernandina Beach, FL, USA

That still has nothing to do with ISO invariance.

It's no coincidence that cameras that we consider to be ISO invariant also happen to have a wider DR at base ISO. It's because they are newer. There has been more R&D.

The companies making those cameras are now better able to interpret the information captured by the sensor in a linear fashion.

Among the goals that the manufacturers set for themselves are more precise shutters and a linear response to changes in ISO which is ISO invariance.

My X100T (made in 2018) is ISO quite invariant within a limited ISO range (200 to 2500), even though its DR does not match any of my other cameras.



But its mechanical leaf shutter is imprecise compared to its electronic shutter which is linear up to to 1/4000s.