Point cloud gradual selection threshold

[Yoann Courtois]

Hello !

We have recently compare our process using old 1.5 release and new 1.7 release.

Using the same dataset, we were surprised of some major differences of number of points selected with gradual selection, using of course same thresholds.

Our questions are :

• Has anything changed in calculation method of gradual selections ?
• We have found that Reprojection error threshold come from pixel error divided by key point size. Right ? We have different results between 1.5 and 1.7 while mean key point size is closely the same.
• What about Reprojection uncertainty and Projection accuracy ? The first come from ratio between distance between images and distance between image and tie point. But it seems we miss some coefficients. For the second one, we know it comes from ratio between both distances between image and tie points, but something else seems also to be taken into account.
• Is there any possibility to get or calculate the mean key point size, or all individual the key point size ?

Regards

[Alexey Pasumansky]

Hello Yoann,

Do you apply gradual selection to the same project (processing in 1.5) or the source data is only the same, whether the processing has been performed independently in 1.5 and 1.7 version?

Description for the gradual selection criteria is given on pages 111-112 of Manual:
https://www.agisoft.com/pdf/metashape-pro_1_7_en.pdf

[Yoann Courtois]

Indeed, at first, we apply gradual selection on process independently done on both version.

We tried to do it with the two version on the same process, and gradual selection works exactly the same. By the way, thank you to have share criteria formula on manuel ! Very useful !


So now, our conclusion is that images alignment looks less robust is last release as more points were filtered using same gradual selection thresholds.

Regards