I found very interesting repositories in Cramer lab's GitHub (but they have not been updated since last year).
micromachines "Performance and Quality Comparison of Movie Alignment Software for Cryogenic Electron Microscopy"
Great study. I am glad they confirmed multi-threading scalability of RELION's motion correction. Did they use Intel compilers and MKLFFT? This makes a big difference for RELION's motion correction.
"Relion MotionCor implements Bayesian polishing": this is not correct. RELION's motion correction (reference free and patch based) and Bayesian Polishing (reference and particle based) are different modules and use different algorithms.
RELION's motion correction is meant only for an initial stage of processing. Because Polish will give better results anyway, the initial motion correction is not optimised for quality. The dampening of high resolution shells is a deliberate design choice. A slower but higher quality interpolation was intentionally avoided.
They should also compared CC against the ground truth for synthetic data. A SNR is a better metric than a power spectrum (e.g. one can reversibly dampen certain resolution shells without losing SNR).
BBC "Luton Airport flights suspended after large car park fire"
BBC "Kenyan publisher recalls book after uproar over Prophet Muhammad image"