I am processing a dataset in cryosparc v4.4.1 and I noticed a strange pixel scaling issue when running heterogeneous classification. I generated some initial volumes using Ab-Initio from particles that were extracted with a box size of 160 pixels, apix 0.86. When running heterogeneous classification, I accidentally linked a set of particles that had been extracted at 256 pixels. The volumes that were output were strangely large – they have a box size of 128 and a voxel size of 1.72, which is scaled up by a factor of 1.6. It appears that when running heterogeneous classification, cryosparc only checks the apix of one source.
No - heterogeneous refinement by default uses a box size of 128px for speed. This is normal, and can be configured in the input parameters.
1.72 is 2x0.86, which is what one would expect if you extracted your initial particles in a 256px box, and then ran hetero with a 128px box (as is the default).
1 Like
As Oli says, this would be expected with the defaults for heterogeneous refinement.
One of the advantages of CryoSPARC is that it does things like this on-the-fly. 
Of course, it also turns out to be one of the disadvantages at times, because (as I discovered recently with some new users) they can get really confused when output is a different scale to the input.