I am trying to use some older drone imager for mapping. This wasn’t taken with the best practices or Pix4D stitching in mind, but it is not possible to refly the survey. I need to maximize the quality of the stitch with this imagery as best I can.
I have noticed that I can run initial processing, then, without clicking anything or changing anything about the inputs, I can run initial processing again and get a totally different output. I have attached four quality reports; all of these are from the exact same inputs. As you can see, the GCP error ranges from 0.001 to 21 meters. The Computed Image/GCPs/Manual Tie Points Position look extremely different across the different runs.
I have noticed that while re running initial processing can result in really different maps, just hitting rematch and reoptimize does not change the outcome signficantly. I have three questions:
-
what is the difference between rematch and reoptimize and re running initial processing?
-
I am assuming that these differences are resulting from some sort of random seed process at the beginning of initial processing. What is that process? Where can I get info about the non-deterministic modeling in Pix4D?
-
Can I trust a "good’ run if I’m getting a map that looks correct and for which the error calculation is relatively low?
Thanks!
20210129 ANM Stitch_report1.pdf (140.2 KB)
20210129 ANM Stitch_report.pdf (726.6 KB)
20210129 ANM Stitch_report3.pdf (159.2 KB)
20210129 ANM Stitch_report2.pdf (319.3 KB)