Automasking

[biochem-fan]

There are repeated requests for automasking.

Things to consider:

• How to do this robustly?
• Is this really necessary? If this is for resolution estimation, we can use a very permissive mask (e.g. sphere) and correct FSC as in cisTEM.
• Is this really useful? It might apparently automate processing but users have to check the density and mask anyway. Otherwise they might miss an unexpected weak densities below the masking threshold.

[xeniorn]

1) the way cryosparc does it seems to make the people on our site very happy

2) at the very least, makes it faster to do things at first. While there are cases where one could miss a weak local density due to auto mask cutting it off, most people most of the time, especially people with less experience, would inadvertently do it anyhow manually. Likewise, many structures don't have a feature of interest that is lost this way. An experienced user can always choose not to use an auto mask, but a provided one.

3) arbitrary defaults are already used in Refine3D in initial lowpass filter (60), symmetry (C1), choice of zero-masking (true), solvent flattening (false), angular sampling, and most importantly the regularization parameter T (2), and the mask diameter (200 A). Each of these can affect the result to a different extent, but do have provided defaults. A sensible, "works in more than 75 % non-tricky cases" auto mask would be aligned with the norms used in relion regarding other inputs.

Progressive tightening of the mask during refinement could also be a useful feature, saving at least computational time, but possibly also leading to better refinements (even before manual interventions and fiddling with masks). Can be done both with the automask or the provided mask (e.g. user provides a tight mask, which is in first iterations extended ~10 px, then progressively tightened to the original size as the refinement proceeds).