Once RC reports over the top projection errors via the red exclamation triangles, are these simply ignored on subsequent alignments based on the max projection error setting? If so, and if one is confident of carefully placed CPs on strong candidate features, then on subsequent alignments, assuming the model is taking shape with more pictures and a decreasing mean projection error, would the projection error reported on these over the top CPs come down with the youngest component active? If so, I’d be tempted to keep these around, since at some point they might contribute once they come under the threshold of the max projection error. If these need to be deleted because they poison the well, also useful to know. Thanks.
Hi Benjy,
if there are huge repro errors, that means that RC has placed those images already somehwere. If those errors don’t go away after the 2nd alignment in the project, my experience is they won’t do that anymore at a later stage. There are cases, where not even Ground Control Points can convince RC to rethink its camera positions. In those cases,the only thing that helps is to delete all and start from scratch but including those CPs, which then might make RC to put them in the right place. What I haven’t tried is how it would work out if one would do this sort of surgery on a smaller image selection and then re-import the correct model part as a component…
hello,
in my experience, you do not always need to delete all - sometimes when you delete one, it takes away more red exlamations, if not even all the rest
Ah but how to know which one (to delete?)
I’m seeing CPs all anew, see this post. I’m going to make a feature request, which I’m tempted to say belongs in the Bug section. It never made any sense to me that projection errors in otherwise solid CPs climb higher and higher. My present take is that RC benefits from CPs in cases where the human eye outperforms RC’s ability to detect good tie points, but that once these are placed, if bad tie points exist in neighboring images, either within two components that won’t merge or in a case where a loop is trying to close but can’t because misshapen geometry got off somewhere, as structured by the tie points, then no amount of increasing weight of the CPs has any effect. The present fix is to go after the bad tie points. I also don’t see how deleting red exclamation point CPs has any bearing. As Lubenko states, these values can change to the positive by changing or deleting a CP in a particular image, or moving it, but that’s useless as I see it, moving what you know is a good CP to make bad tie points happy. That problem will only bite you later with separations in the model. Read my post for a more in depth explanation.
Hi Benjy,
you are absolutely right - deleting CPs does not change anything. The errors merely represent the difference from what RC calculated in the current alignment to what you defined in the 2D images…
I’m sorry for being unclear. What I meant is to delete COMPONENTS ! This ist so far ingrained in my workflow now that I seem to have assumed it is clear what I am talking about…
Oh and to answer Tom: I usually keep the first one because it is “virgin” and in my workflow always with exif grouping, which means more reliable results at the cost of some percent in reprojection errors. But you can experiment a bit, just make sure you save the project before deleting anything…