Forum

Author Topic: Added a handful of pics and grouped, re-align taking 10x longer?! (1.1.6)  (Read 3234 times)

andyroo

  • Sr. Member
  • ****
  • Posts: 406
    • View Profile
I have an aerial project in PS 1.1.6 build 2038. Originally had 2400 images from two cameras (Ricoh GR and Canon D10) where I aligned the images in ~10 hours then entered GCPs. Because alignment failed above a narrow canyon I pulled in another few hundred pics of the canyon reach from an earlier flight and ran the align task again. It's now been about 12 hours and PS tells me that I have 165 hours left.

In both cases I used generic pair preselection, high quality setting. The only other thing I did was group the pictures (so I could easily disable the few hundred pics from an earlier flight before making the dense cloud and mesh). WHat would cause the alignment task to take 10x longer just because I added GCPs, a few hundred images, and grouped the pics?

Total memory is 192 GB, memory in use is 32.8 GB. All CPUs are maxed. I'll leave it running...

andyroo

  • Sr. Member
  • ****
  • Posts: 406
    • View Profile
I removed GCPs and ungrouped and the realign was rapid. Not going to bother figuring out which was the issue unless Agisoft expresses some interest in this bug report though.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14003
    • View Profile
Hello Andy,

PhotoScan uses markers as velid matching points during Align Photos stage. In case there are any problems with the projections, the processing time will be much longer and the result will be affected.
Best regards,
Alexey Pasumansky,
Agisoft LLC

andyroo

  • Sr. Member
  • ****
  • Posts: 406
    • View Profile
I didn't notice any problem with the projections with the markers I added, but I am guessing if my hand-placement isn't within the error tolerances in PS preferences that could make things more difficult for alignment solutions.

Would you recommend standard practice in this case be to export markers, realign without, then import? that solution was effective for me. I suppose I could try relaxing tolerances on GCPs in the future, but I am hesitant to do that since I get low residual error with current methods.