The resolution estimation bug discussed some time ago (Resolution in FSC), where cryoSPARC estimates the resolution as the highest resolution where the FSC crosses 0.143, rather than the lowest, is no longer present in Homogeneous refinement jobs, but it still persists in local refinement (I haven’t checked non uniform). See attached screenshots - one with local refinement, the other with homogeneous.
Would be good to have a fix, and also to use the corrected FSC during intermediate iterations of local refinement - presently, if small masks are used, local refinement has a tendency to spiral towards unrealistically high resolutions. Using the corrected FSC to set res limits for alignment at intermediate iterations might help with this.
Hi @olibclarke, thanks for reporting this in detail. It’s certainly on the radar and we’re working on finding a way to avoid these degenerate cases (ideally without reducing the ability of refinements/local refinements to converge to high resolutions in other cases).
I observe this phenomena duirng homogenous refinement. The issue is that because it takes the highest resolution FSC cross and not the lowest ,the dynamic mask is applied too early in refinement. It turns out not to be a problem for this particular case but coud be for other samples.
This issue is still present in 2.14.2
Yes would be great to see a fix for this @apunjani, is there any timeline or any fix on the horizon?
It would also be nice to change the y axis to a negative starting value to observe the expected ringing of the FSC although i can make my own plot from the txt file after calculating resolution (Å) = box size x pixel size / wave number…