I can believe there is a visible difference between large sensors and small sensors, both in terms of noise and resolution. But that difference won't necessarily be visible in a small image. So, when does it become visible?
When I did
this test , I found that the difference between a FF and 33x44mm sensor in good light was hard to see in 15-inch-high prints, but not difficult to see in 30-inch-high ones.
If I had done that test with a GFX 100x, it might have been easier to see the improvement in smaller sizes. If there was an opportunity for aliasing, that would have also increased the probability of seeing differences at 15 inches high.
Interesting.
Coincidentally, I use A2 as a benchmark paper size. With a small border, that works out similar to your 15" high print.
I used to have a D800, but was transitioning my APS-C cameras to Fuji (16 MP, then later 24). I made two A4 prints of 50% crops from both cameras. I could see a small but clear difference between the D800 and 16 MP Fujis when I looked closely - ie from 12" - but not from a normal viewing distance for A2 - ie 24-30".
Note, as far as possible I equalised sharpness using the best techniques I could, and used C1 rather than ACR, which doesn't mess up Xtrans files so much.
Of course, as noise levels increase, the Fuji's tailed off a little earlier and had less DR, but nothing dramatic at low ISO.
Then I bought a 24 MP Xpro2, and did the same experiment. The small increase in resolution, but without an AA filter, leveled the gap. It was almost impossible to see any meaningful difference that I could not attribute to other factors, even viewing closely.
I repeated the test using downloaded images from DPR using Bayer array cameras to eliminate Fuji's Xtrans sensor from the equation, and the results were even more similar.
Since perceptual differences with larger sensors narrow progressively if we keep the image size constant, I would have expected the GFX and A7Rii to look almost identical, based on this experience.
However, one could roughly predict visibility of detail in prints from CSF and other factors. I am particularly interested in what people refer to as 'tonal gradation'. This is a particularly tricky one because it's not defined, but widely touted as a benefit of MF to the extent that its worth investigating.
I imagined that 'good tonal gradations' could be tested using a continuous gradient, so I created several in photoshop and photographed them. I didn't print them, just looked at results on the screen, but if I used ETTR to stop the highlights from clipping, I could not see any difference between FF and APSC, even when I used a tone curve to match them almost exactly.
So, by inference I would be less likely to see an improvement with a larger sensor at the same image size.
So, what am I missing here? It's clearly less to do with resolution, and more a combination of colour gamut, SNR, DR and bit-depth, in which case my screen could well be the limiting factor, even though it's a pretty good one.
