I've been doing some more testing on the D-90 and 10.5 fisheye combination and my results are slightly different from what the Camera Settings page suggests. (Assuming the D-80 figures remain the same for the newer camera, which I believe they do.) Instead of 48 - 86, I'm showing best results with 47 - 83. Is this enough of a difference to matter or should I just shut up and go take some pictures?
Interestingly, when I tried the "Sawtooth" test, as recommended by John H. and others, the result was slightly different from what the "aim the camera straight down and center the target in the viewfinder" method gave me. Is that because of a slight de-centering of the viewfinder in a camera that's not set up for 100% view?
Are you using the quick-release device? I believe that would add 10mm to your lateral number, which would make your figures with the NN3:
46 - 81. This would certainly be close to my:
47 - 83, though even further from the numbers on the Camera Settings page.
So it looks like our numbers are pretty close, maybe so close that the difference doesn't matter. That's what I'm still curious about.
I will mention one other note. When I tested my "B" setting the first time, I lined up near and distant subjects while aiming the camera to put them at the very edges of the viewfinder. Then John's tutorial reminded me about the odd behavior of the Entrance Pupil in fisheye lenses, so I retested and swung the lens only 60 degrees, the same amount I swing for my six shot horizontal series. The results were slightly different, though I can't remember in what way. And, again, the change might have been caused more by sample variation or my sloppiness.
And for what it's worth, my technique lately has been to shoot six up at 30 degrees, six down at 30 degrees, one shot straight down (even though it's redundant and the tripod's still in the shot, it sometimes helps with stitching later), then a nadir off the tripod. No zenith.
A small update on my sequence... I've reduced the number of shots and the results are as good, if not better, than before. Instead of two rows of 6 shots each, one at 30 degrees tilted up and one at the same angle down, I'm just shooting one row straight across. And now I'm adding a zenith, or two of them 120 degrees apart if there's a lot of blank sky. One nadir on tripod and one off.
Another update if you're interested. With my new NN-3 settings and sequence I'm very happy with the results. In PT Gui, control points for the first six shots around generally all fall below .5. Usually the zenith and nadir are that close, too. This is definitely better than my old system. It's nice to shoot with greater confidence.
I use Heavy+lens shift in the "optimizer" - advanced setting.
Did you already realize that in the top header under control points you find "Delete Worst Control points", and "Generate Control Points for selected images", which will generate more CP's once you have set some manually if needed for nadir and zenith?
Yes, all those things help, and I usually set them, though I'm always glad to get a reminder or confirmation of what I'm doing! I think the only way I work differently is that I prefer to just grab the most distant control points in the Control Point Table and delete them manually. If you arrange the Table so that the biggest distance values show at the top of the list they're easy to find.
Is there a way to separate the two Control Point tabs so that I can see them at the same time? That would save me a few clicks each time I delete control points and re-optimize.