Thanks for the answers fellas.
@mraw I have tried doing this but not much luck to fix these kind of issues, I would almost argue the results are slightly worse after doing this. What I did was:
- Gradual select (reprojection error) 10% of the total amount of tie points, delete
- Optimize cameras with the default settings
- Repeat until I have reprojection error of less than 0.4
I also tried doing similar steps with projection accuracy (down to 3) and reconstruction uncertainty (down to 10).
I'm I missing something or is there a different approach to doing this I'm unaware of?
@olihar What info would you need exactly? It's a scan of a medium scaled broken trunk, so a lot of awkward angles. Images are of good quality (Sony A7R4 w/cross polarized flash), albeit there is quite a bit of DOF in them due to the awkward shapes of the trunk. There we're two passes done on it as I was verifying the data on location and the first pass showed some poorly covered areas. The second pass is what is causing the alignment problems, it was about 20 min after the first pass. The object had not been moved or touched, nothing really changed hence I'm a bit confused why the second pass caused this.