Have to say, after a brief gander, what he's talking about is that essentially, there is not ISO in digital. There is the base reading put out by the sensor, which is the same in every instance, and then there is baking done by the rest of the camera to create the effect of an ISO change for jpeg shooting. Why is it necessary to understand this... because for us film shooters, the only way to make a film faster was to increase the size of the grain. So, faster film was by definition, lower res. Larger grains but fewer of them.
Digital is different, you have the same number of pixel sites whether high or low ISO, the base exposure at the sensor level is taken at one value. All the camera does is change the mid point in a sort of internal "levels" way. That is relevant to jpeg export, because jpeg throws out the unused date that was made extraneous buy shifting the mid point.
And the shadow detail from the second. That is where the DR comes into play. In my experience, if you shoot at ISO 400 and achieve that look, in my experience, you will not be able to get the richness the original image has in the sky. The sky has to be properly exposed.
To prove ISO invariance he would have had to try and normalize both a 400 ISO image and a 100 ISO image and shown they were the same. He claims the simulated 400 ISO image is demonstrating ISO invariance, but to do that he should have used a 400 ISO image and tried to get the same result out of them.
Quote: No cameras are perfectly ISO invariant starting from base ISO; however, some are quite close, and many cameras become ISO invariant starting at a high enough ISO value (once the camera begins to “simulate” ISO values).
In his opening line he defines th parameters. Understanding those restrictions, when you understand you are only talking about ISO invariance when discussing what he calls, "Back end noise", the concept is pretty limited in use and application.
However, I often shoot base ISO at 2 under , 1under, even , 1 over and 2 stops over, in a sunset sequence. And from that perspective this is all nonsense. I can tell you without fear of contradiction, one of those exposures will give you the best image, although sometimes the exposure you need is half way between the two and 2 of the 5 will give you almost identical images.
So in my expereince, ISO invariance is irrelevant. What there is, on a given sensor there is a point at which the mid point on the curve is positioned at an optimal spot to produce the best image. But where that point is, is highly subjective, to the point where I've given up anticipating it. And that mid point is changed, not by the ISO but on how long the exposure time is and how much light is allowed to reach the sensor relative to the scene you are trying to capture.
So to mee the whole question of ISO invariance is subject to the absolute need to get the best exposure for your subject. How does ISO invariance help with that? Why are we talking about it. Where is the practical application. Especially since the changes are subjective. If you have a scene with 20 EV and you only can capture 15 of it, how long you expose lets you control whether you throw out the top five EV, the bottom 5 EV, or some combination of the two extremes. That is an aesthetic decision that has nothing to do with ISO invariance. And to me that to me is the issue. I'm not sure why ISO invariance is important, but to me, it's a non-issue. It doesn't address how to get a good image.
And much of what is explained in the linked article is irrelevant, because it's either over simplified, or irrelevant to the point., and because as admitted by the author, their are no ISO invariant cameras.
All you need to know is, that once your camera starts using digital enhancement, your post processor can do the same with raw files. But that's not true at all with in any circumstance with jpeg files. And it's not true where for aesthetic reasons you want more or less light than the camera can capture and use your exposure time to control the actual number of photons hitting each pixel site.
So, apart from the observable obvious, digital enhancement usually sucks, which can be explained other ways than a term as nebulous as "ISO invariance" I'm not seeing any useful application for this knowledge. It's like someone invented a topic for no apparent reason.
I find it interesting that you would imply if I did the work, I would think what you think. Well, I've done the work , many times on many images, and i don't think what you think. And the link you posted is neither peer reviewed nor technically relevant in all of it's information. I'd compare it to a guy quoting a bunch of accepted knowledge and trying to sneak a new concept into it, without doing the work he needed to do to prove his point. Especially since he says in his opening, it doesn't exist and provides no examples to show why he says that, even though if included, that would be the most valuable part of the article.
I'm a great fan of science, pop science like this, not so much.
My objection is not that it's not a thing, my objection is that, it's a worthless thing.
Relevant to the the notion of ISO invariance when you use 400 ISO you use a 4 time less light hitting the sensor than you would at 100 ISO. IN a simple system that will cost you two stops of dynamic range. Claiming ISO invariance is to claim that this does not effect not he final image, noise etc.
Nonsense.
Very interesting topic though. As long as you remember, Low ISO means more light hits the sensor, less ISO means less light hits the sensor, and that you can have too much or too little light hitting the sensor for the image you are trying to create, so you understand the limitations of Invariance, it could be useful for some people, I'm just having trouble thinking of who it might be and for what application.
The fundamentals, expose properly, use the lowest possible ISO, use high ISO to shorten shutter speed, not increase IQ, and as far as i can tell, Invariance makes little difference to what you need to think about.