The defaults for dynamic masking (in non uniform refinement for example, though it also applies elsewhere with slightly different values) are to expand the local mask by 6Å, and then set a 8 Å soft edge.
Would it not be better to have these values specified in pixels? Because let’s say one is working with data binned to 8Å per pixel (an extreme case), then this would mean that the masks used would be effectively hard-edged binary masks, which is not desirable.
(The reason I ask is that in local refinement the defaults correspond to a 3Å soft edge, which is problematic if one has a pixel size of e.g. 2Å and just uses the default values)
Hi @olibclarke, thanks for the input!
We thought about this a bit - we’re leaning towards leaving the units as is, because otherwise the defaults would have to be different for each dataset, plus many users are not certain of their pixel size during a refinement (due to on-the-fly f-cropping etc) and may be at a loss if these params need to be adjusted every time. However, your note about the defaults in local refinement is probably right - we can change those. What do you think are reasonable default values (in A) for the masking in local refinement?
Hi Ali - that’s fair enough - but if the set value in Å works out to be less than 2 pixels, perhaps cryosparc could revert to using a 3 pixel soft edge, to avoid inadvertent use of a hard-edged binary mask for refinement?
In my limited experience with local refinement, ~a 9Å soft edge has been fine for most things. I would note that we are still seeing some odd results with non uniform refinement in local refinement, even after the fix - greatly overestimated resolution values in some cases where it appears resolution is being affected by artefacts near the mask edges.