Forum

Author Topic: Gradual Selection Order?  (Read 5952 times)

theFinn01

  • Newbie
  • *
  • Posts: 12
    • View Profile
Gradual Selection Order?
« on: June 18, 2024, 05:00:31 PM »
I've seen 2 different opinions on the order of the various Gradual Selection processes and was curious if anyone had any thoughts or reasoning between them.

One method is:
1. Reconstruction Uncertainty
2. Reprojection Error
3. Projection Accuracy

and the other flips the last two steps:
1. Reconstruction Uncertainty
2. Projection Accuracy
3. Reprojection Error

James

  • Hero Member
  • *****
  • Posts: 769
    • View Profile
Re: Gradual Selection Order?
« Reply #1 on: June 20, 2024, 12:49:29 PM »
If you do all three and only optimise at the end, then it makes no difference what order you do them.

Try duplicating a chunk and doing it one way in one, and the other way in the other, and see if you can see any difference in number of points or error values, or anything.

It would only make a difference if you optimised after each gradual select/delete, but i don't know why you would do that.

theFinn01

  • Newbie
  • *
  • Posts: 12
    • View Profile
Re: Gradual Selection Order?
« Reply #2 on: June 21, 2024, 12:46:27 AM »
Thanks James - both these guys optimize after each selection/deletion, so monkey see, monkey do!

https://www.youtube.com/watch?v=je79gV8HsZI

https://www.youtube.com/watch?v=zIa_SNz3IuA

andyroo

  • Sr. Member
  • ****
  • Posts: 460
    • View Profile
Re: Gradual Selection Order?
« Reply #3 on: July 20, 2024, 12:17:36 AM »
It's been a while since I dove into this, but last time I did (I think around 2.0.0), I convinced myself that optimizing between gradual selection filtering of Reconstruction Uncertainty ("RU") and Projection Accuracy ("PA") doesn't affect the results (number of points selected/deleted), so I do both of those before optimizing once (typically using a standard value unless I'm doing something with a different camera system or environmental conditions). I optimize after each iteration of gradual selection for Reprojection Error ("RE") because the optimization makes new points appear for the same RE threshold. For normal collections with our primary camera system I find that one RU/PA and two RE optimizations, enabling "fit additional corrections" on the second RE optimization, are adequate. I try to minimize the number of optimizations to reduce the risk of over-fitting on camera parameters, especially after enabling "fit additional corrections" - primarily because the paper that method is based on hurts my brain.

CheeseAndJamSandwich

  • Full Member
  • ***
  • Posts: 208
    • View Profile
    • Sketchfab Models
Re: Gradual Selection Order?
« Reply #4 on: May 12, 2025, 03:01:09 AM »
If there is a recommended order which we should process, could the items in the dropdown be reordered to reflect it?  Such that we could just step through them.
When starting out with MS, it's seemed a bit mad that we choose them out of order, and that they're not even in alphabetical order! lol.

I still don't understand fully what they each do, or what values to aim for (other than the lovely USGS guide!)
We could really, really do with some tool-tips, with some descriptions, general guidelines in them!
My 'little' scan of our dive site, 'Manta Point'.  Mantas & divers photoshopped in for scale!
https://postimg.cc/K1sXypzs
Sketchfab Models:
https://sketchfab.com/cheeseandjamsandwich/models