hugin-devs team mailing list archive

[tmodes <1097958@xxxxxxxxxxxxxxxxxx>]

This is difficult to debug. The assistant runs a lot of steps which affects the control points:
cpfind - linefind - celeste_standalone - cleancp
So from the results it's not clear in which part the bug is. So it would be better to run only one program at one time and compare then. So it's easier to locale the problematic part.

Some more information on some details:
1.) Cpfind is running first a feature detection and then the feature matching. The feature detection phase is deterministic. I fixed lately 2 bugs in this code which resulted in some randomness.
The feature detection phase is using the RANSAC algorithm to filter outliers. So in this step there is always some randomness involved and therefore you can not expect to get the same result in each run. But the number of affected cp is very small. Most of the cp (>99.5 % or even higher) is always the same. Only some minor cps are different. 

2.) All celeste variants (image tab, celeste_standalone, cpfind) are using the same code. There can be some differences in the parameters used to fine tune the algorithm (mainly radius and threshold). The image tab is using the settings in the preferences. All other can be tuned by command line parameters.
When running the assistant with "cpfind+celeste" and celeste_standalone, the celeste code is running two times (maybe with slightly different parameters). So currently it is not clear which one of the two programs has problems. So you need to test only one program at one time.
In the assistant cpfind is using the parameters given in the control points detector settings, celeste_standalone is using the parameters given on the misc tab.


** Changed in: hugin
       Status: New => Incomplete

-- 
You received this bug notification because you are a member of Hugin
Developers, which is subscribed to Hugin.
https://bugs.launchpad.net/bugs/1097958

Title:
  CPFind random in CP detection and not finding cloud points

Status in Hugin - Panorama Tools GUI:
  Incomplete

Bug description:
  A number of tests with CPFind, using Assistant for alignment, finds different control points each run.
  Same tests using APS-C finds same control points each time.
  See test results in attached pdf which also details the test conditions.

To manage notifications about this bug go to:
https://bugs.launchpad.net/hugin/+bug/1097958/+subscriptions