Optimization with few bad points

[Michail S.]

 Hi! I often stitch panoramas of 10-100 microchip photos with autogenerated control points. Due to regular structure on microchips often there are quite a few (0.1-1%, sometimes even 5%) incorrect control points. It is clear that optimizing control point detector is first step to resolving the issue, but it is hard to prevent it completely.

In the past Hugin somehow was able to optimize it correctly, leaving these bad points with large error. It was easy to delete these points based by error, and do final clean pass. Now it looks like it tries to minimize total error, and breaks alignment to reduce error on all bad points too. I.e. 1% of bad points can easily break alignment.

Does anyone know if optimization algorithm can be adjusted to behave differently? Are there different optimizers and did optimizer drastically changed over last few years?

Best regards,

Mikhail

[T. Modes]

Hi,

byfl...@gmail.com schrieb am Sonntag, 24. März 2024 um 15:56:48 UTC+1:

 Due to regular structure on microchips often there are quite a few (0.1-1%, sometimes even 5%) incorrect control points. It is clear that optimizing control point detector is first step to resolving the issue, but it is hard to prevent it completely.

Does anyone know if optimization algorithm can be adjusted to behave differently? Are there different optimizers and did optimizer drastically changed over last few years?

There are no different optimizer in Hugin/panotools and it did not change drastically in the last years.

For this use case there is cpclean which should delete most of the wrong control points automatically.

Thomas