Per-particle scale in 3D Variability

Hi All,

I noticed that the default for Per-particle scale in 3D Variability is optimal. Should the default be none? Thanks.

Hi @donghuachen,

The per-particle multiplicative scales are treated as following for the various options:

  • none means all particles are fixed to have scale of 1.0 – this is useful for small particles or low SNR, where scale estimation is more challenging.
  • input means to read and use the scale factors that are input to the job – these likely have come from another refinement, and may be previously fit (e.g. if the “Minimize over per-particle scale” parameter is on).
  • optimal means that scales are optimized on the fly during processing, based on the 3D Variability model that’s described in the corresponding publication. Typically this should give the best results, because the scales are inferred using the same generative model as the variability modes and latent coordinates. However, it may be the case that using none gives better results in situations as described above.