[--iter=<i>] | Number of iterations. 10 is almost always sufficient. For less noisy data 3-4 may be enough. |
[--ninitcls=<n>] | Number of classes to generate (though some may be discarded). Typically < #ptcl/10 |
[--finalsep=<n2>] | In the final iteration, each class will be subdivided into n2 subclasses, producing n*n2 classes |
[--nbasis=<b>] | Number of basis vectors to use in SVD. Default is usually fine. |
[--proc=<p>] | Number of processors to use (EMAN standard classification routine) |
[--ctfcw=<sffile>] | Perform full CTF amplitude correction when making final class-averages. Only used in the final iteration. |
[--minptcl=<m>] | Minimum number of particles in a class to keep. |
[--aliref=<ref img>] | If specified, the first image will be used as an alignment reference for the first class-average to help 'anchor' the set |
refine2d.py --iter=8 --ninitcls=50 --proc=4 particles.hed
This program will generate very high quality 2-D class averages based on a SVD/MSA iterative classification scheme. It is not typically involved in the 3-D reconstruction process other than perhaps to generate an initial model, due to resolution and noise bias issues, but for 2-D analysis, it performs extrememly well.
This program produces a large number of intermediate/output files. It is suggested that you run it in an empty directory. The main output files are : iter.final.*
Many 'tricky' things can be done with the results of refine2d.py. Check the FAQ on the EMAN wiki for tips on how to use this for dynamics analysis, particle separation, etc.