In 3D classification, iteration 1 densities are scaled by normcorrection

[thurberk]

This is a template for reporting bugs. Please fill in as much information as you can.

Describe your problem

In 3D classification, if the input data.star file has NormCorrection values different from 1, the iteration 1 densities are seeming scaled incorrectly. The iteration 1 densities have a ~1/NormCorrection scaling, relative to the iteration 2, 3, etc. densities. This can affect the classification of the particles. Probably affects classification the most if you put in multiple different input reference densities. (If you input just one reference density, at iteration 1, the classes are probably similar enough that the mis-scaling doesn't matter as much.)
This density scaling issue can be avoided by setting all of the NormCorrection values to 1 in the input data.star file (but users will not necessarily know to do that).

I believe this is caused by:
line ~1023 in ml_optimiser.cpp in MlOptimiser::parseInitial
which is
mymodel.avg_norm_correction = 1.
This seems to set the average norm correction to 1, without checking whether there is already norm correction information in the data.star file.
Then, I think the iteration 1 densities are mis-scaled because the avg_norm_correction is set to 1, while the norm correction variables in the data.star file do not necessarily average to 1.

Environment:

• RELION version RELION-5.0-beta-3-commit-7d79f3
  but this issue is also present in Relion 4.

• Testing suggests that this issue doesn't depend on helical/non-helical processing, or --firstiter_cc

• Option --always_cc seems to not have this issue

Dataset:

Job options:

• Type of job: 3D classification

• Full command (see note.txt in the job directory):

which relion_refine_mpi --o Class3D/job822/run --i Class3D/job776/run_it001_data.star --ref model_classes_sc776.star --firstiter_cc --trust_ref_size --ini_high 8 --dont_combine_weights_via_disc --scratch_dir /lscratch/$SLURM_JOB_ID --pool 100 --pad 2 --ctf --iter 3 --tau2_fudge 1 --particle_diameter 400 --K 1 --flatten_solvent --zero_mask --oversampling 1 --healpix_order 4 --offset_range 8 --offset_step 4 --sym C1 --norm --scale --helix --helical_outer_diameter 110 --ignore_helical_symmetry --helical_keep_tilt_prior_fixed --sigma_tilt 5 --sigma_psi 5 --sigma_rot 0 --j 1 --gpu "" --pipeline_control Class3D/job822/
++++

Error message: