Another, hard to tile photogrammetry based cobblestone texture. In that case, puzzle solving took me over an hour
I also confirmed that using full range spectrum (0-255) to store height information makes more sense as we have that part of memory allocated anyway. In result we get more useful data (better, more accurate and more responsive shape definition with less noise) during height control on final shader.
That scan is a good example of bad path planning. As surface was nondestructive I moved with my camera row by row backwards. I was very low on battery when I finished main capture, but I decided it would be worth to capture also side bricks to get more useful highpoly model with more opportunities to bake from. So I started chaotic shooting. As you can see, one side wasn't covered good enough to generate nice final mesh with two rows of side bricks. I got lost and wasn't sure what was and what wasn't captured anymore. Than sun went out :) and shooting was over.