Performance degradation after optimization and alignment errors

[mrv2020]

Hi Alexey,

Some image sets are aligned incorrectly in version 2.2, but I only notice the misalignment after generating the DEM, so I use markers to manually realign.

So two things happen: if realignment optimization has already been applied, it takes an excessive amount of time to complete the manual alignment compared to a set that has not been optimized.

And another situation is that if a set has images without coordinates, they will almost always not be aligned with the others. I did a test in version 1.8.6 and the same dataset was aligned the first time and without any imperfections, with the DEM smooth and wrinkle-free.

Another test was activating the adaptive camera with a set that has more than one type of lens or camera model, the alignment always fails in version 2.2.1 requiring intervention with markers to correct even if they have precise coordinates, and again in 1.8.6 it passed perfectly the first time.

And when version 2.2.1 finishes the alignment, I notice that the global error is 0.23m and after optimization it reaches 0.048m. With filtering, we reach 0.02m. In version 1.8.6, the accuracy is 0.0186m the first time without needing to optimize after alignment.

[mrv2020]

Hi Alexey,

I ran a test on the current version 2.2. The scenarios are the facade of a small building. So, with the same set of cameras, all with precise coordinates and the check marked, it aligned perfectly with a global error of 0.016m. After that, I loaded the same cameras into a new project, disabled a random group of 30% and aligned the same set. The cameras with coordinates produced a global error of 0.017m, but the disabled group was completely wrong. I created markers to check the deviation and optimize, and found errors in the order of 700 error (pix).

The solution adopted was to misalign the group of wrong cameras, optimize the set and then manually align in small groups. Even without markers, they align perfectly, but it is a lot of rework and a long time to align 20 photos at a time, taking more than several minutes.

cameras without coordinates within a set with cameras with GPS have difficulty aligning in version 2.2, in version 1.86 it aligned perfectly, regardless of the number of images without coordinates with check or even without coordinates informed.

[Alexey Pasumansky]

Hello mrv2020,

Can you share the project with two chunks (or two projects) where the same images have been processed in 1.8 and 2.2 versions with the same parameters, but with the different results?