ISO & Dynamic range - what is going on?

There's a parallel with Sigma-Foveon cameras in their low-res raw capture. On-sensor, pixels are binned 2x2 connecting the outputs of four pixels together in parallel. The effect of that is, for a given exposure, four times the current goes into a four times bigger capacitor resulting in the same voltage output from the now big pixel which is the average of the previous four pixels.

P.S. Shot noise is also reduced.

Read noise, in the context of imaging and electronics, refers to the noise introduced during the process of reading out the signal from an image sensor, like a CCD or CMOS sensor. It's essentially random fluctuations in the electronic signals that represent the image data as it's being converted from analog to digital.

Why the different lenses and focal lengths, for the same lens mount? The right-side image has color fringes (red and blue) in the specular lights on the cornea, and the left-side image does not. Obviously the lenses and/or different processing are playing roles here; not just different pixel densities. Also, you lose the guarantee of equal exposure when you use different optics, especially at low f-numbers like 2.8, where the actual transmission varies wildly with different lenses. Also, at f/2.8, most zoom lenses have much more aberration than diffraction, and so total blur will be different. The focal lengths are different; did you adjust distance, accordingly, for the doll's eye to be the same size in mm on the sensors?

Again, Nearest Neighbor upsampling gives predictable results with the same scene and optics; the larger the pixels, the higher the maximum contrast between neighboring pixelated pixels. If you use a sharper lens with the bigger pixels, then you're going to have even more high-contrast between the pixelation neighbors (horizontal and vertical, only). This is what the primitive levels of our brains look for to feel like we have successfully focused on an object, which can fool us into mistaking this acuity for actual image information.

So, there are just too many wild cards here for your test to have any meaning.

The 2 tamron lens are both equally as sharp as each other i just framed the doll to the correct composition. lens breathing is most likely playing a part in the different focal lengths, its only 5mm 😏, as a guy over in the sony forums has also noticed his a7iv images appeal more over his a7r5, at least he being honest with him self. i just shot identical images from 2 tripod set ups at different iso readings to see when noise started to decreases detail . it starts at iso 320.

It's the key to understanding ISO-invariant (ISOless) sensors.

Modern sensors have been designed to have very low read noise. This means that they do not benefit much (if at all) from raising ISO because there is very little read noise to suppress. Raising ISO (usually, not all cameras' ISO knobs do the same thing) reduces dynamic range because the amplification boost drives highlights into clipping. If it brings no useful read noise suppression, then it is simply a dynamic range destruction knob, and using it is pointless. You will get the same read noise leaving the ISO knob at base ISO, deliberately underexposing and brightening in post. The increase in shadow noise from lifting the shadows will be minimal because the read noise is so low to start with. By leaving the ISO knob alone, you hold on to your highlight dynamic range. This is the ISOless strategy.

In saying this, I don't actually shoot ISOlessly, even though for best image quality I should. The reason why is a personal foible. When I use the ISOless strategy I have found that I regularly let shutter speeds fall too low and lose sharpness to camera shake. Switching to auto ISO and letting the camera vary ISO helps avoid falling into the too low shutter speed trap, even though I sacrifice dynamic range as a result. I have decided that losing dynamic range is less bad than suffering camera shake. But that is a foible, anyone who can reliably remember to keep a good eye on their shutter speeds and adjust settings accordingly would get a dynamic range advantage from following the ISOless strategy (assuming they have an ISO-invariant camera) without suffering more read noise.

p.s.

I'm not an engineer and this is a gross simplification of what read noise is. The details are, frankly, beyond me, and my description lumps several sources of noise into one simplified bucket, but for the purposes of this level of debate, treating it this way gets across the gist.

p.p.s.

If you shoot with a dual gain sensor and want to follow the ISOless strategy, you'll want to modify it slightly into a Two-ISO strategy. Shoot at base ISO for shots where there is plenty of light then switch to the higher gain ISO when you know you will be underexposing a large amount.

p.p.s.

Personally, I'm too stupid and get too excitable under real field conditions to keep all this complexity straight in my head, so good old school auto-ISO works for me, even though I know I am sacrificing dynamic range. Better a shot that 'comes out' than a tuned shot that is broken, as I have figured out from hard experience.

For anyone interested in the technical details which are not all what Alan said:

chatgpt.com/share/6876c085-0750-8013-a408-c6d5451619af

thanks for the link.

heres another Ai overview

Read noise in an image sensor occurs after the photons are converted to electrons and before the analog-to-digital conversion (ADC). Essentially, it's the noise introduced during the process of reading out the pixel values and converting them into a digital signal. This noise is a combination of various factors, including the sensor's pixel and amplifier noise, as well as the ADC's noise.

Sorry, Alan, I will not be taking the time to provide you with "proper sources" for all ten conclusions (bullet points) in the link. I did read the conclusions and they looked OK to me.