Back to the roots of this thread, 6 years ago my 1st successful scans were walnuts. I decided to go a bit bigger this time and use 400mpx of pixel shift shots to see what comes out.
6144 total shots compiled into 384 pixel-shifted composites which resulted in 64(4 orbits of 16) focus stacked inputs for RC.
Resolution 23264x17448, that's roughly 400 Mpx - a sample of such is available here:
https://we.tl/t-L2OW7qTown. 310mln mesh resulted from quite quick RC calculation, which cost me just 2186 credits(~5$).
My takeaways from this: a lot of data, the whole processing of it takes so much longer than photogrammetry scan itself and post. I doubt pixel shift added all that much quality result in respect to regular 100Mpx of GFX100S - and there are lots of chromatic aberrations and/or pixelshift artifacts.
Even though I shot in a static place on a sturdy tripod fuji's pixel shift combiner software was complaining that it registered some movement, just a dozen of shots came without warning, but there was no much sharpness difference in them either.