Select 2D using "templates_excluded" and "particles_excluded" from previous selection gives incorrect class assignments (bug?)

issue_recorded

(Olibclarke) #1

Hi,

Sometimes, when one orientation of a particle is rare or low contrast (e.g. top views vs side views), I would like to process it separately from the bulk, by selecting these 2D classes and then performing subclassifications on e.g. side and top views separately.

I thought a convenient way to do this (and to make sure that the same particle doesn’t get assigned to more than one group, and that no good 2D classes are missed), would be to perform an initial select2D job, where for example side view classes are selected, then take the templates_excluded and particles_excluded outputs from this job and use them to start a second Select2D job (where top view classes will be selected).

However, when I do this the second Select2D job has class assignments that don’t make any sense - the thumbnails corresponding to previously highly occupied classes (that weren’t selected in the first round) have 0 particles, and those that were previously minor junk classes with few particles now have many particles assigned. So I suspect that somehow cryosparc is displaying the wrong thumbnail for each class in this case. Possible bug?

Cheers
Oli


(Ali Punjani) #2

Hi @olibclarke could you let us know which version you found this to occur in?
Definitely a possible bug.
If you run multiple select2D jobs with the same inputs and select different classes in each, does that at least allow you to continue using this workflow for now?


(Olibclarke) #3

Hi Ali, I noticed this in 2.9 - I haven’t checked if the issue remains in 2.11, but I assume it does. And yes, the workflow you describe works - it is just sometimes hard to remember which classes one selected in the other select2D. It would be useful if the classes were numbered for this purpose (and also for tracking the progression of a particular class over the course of Class2D).

Cheers
Oli