Hello,
I am currently performing non-uniform refinements (with minimize over per-particle scale) on a dataset after reference-based motion corrections. And at some instances, the job completes after the 5th iteration, yielding a ~7 A map.
Oddly, a clone of the same job ran at a different time proceed to completion with 11 iterations, yielding a ~ 2.6 A map.
I inspected the 2D classes from these two jobs and they look very similar without any clear outlier and both generates masks that are very similar.
This issue seems quite persistent with minimize over per-particle scales after reference-based motion corrections and global and local CTF corr. Similar job ran without this option resulted in a high resolution map of ~2.7A. Interestingly, as you can see for each job, they all ended with a different resolution between 7.9A to 8.7A
Couple things to add, if I remove the static mask from the parent NU-refinement, the current NU-refinement can proceed to 3A.
furthermore if I perform homogeneous reconstruction from the particle stack after CTF corr, we obtain a 3 A map that the following NU-refinement with the same condition (applied static map and minimize over per-particle scales) went fine with resolution of 2.5 A.
