alisterburt
commented
2 years ago I'm keen but would definitely prefer that we have a way to do this from the e/o frames, the results were pretty significantly better in the cryoCARE paper :) what's this better way of incorporating?
First step into incorporating denoising.
Most denoisers require 2 copies of the same tomogram reconstructed from slightly different data. This PR generates a split_tilt_series.star file which can be used to generate 2 copies of the same tomogram. I've called this tilt-series splitting.
Some recommend doing this by generating two images during motion correction: one taking all the even frame numbers from the raw movie and the other taking all the odd ones. Then processing both movies to generate the same tomogram, albeit from different movie frames. I don't like this approach because: 1. to do it in rln you have to go all the way back to the movies and unless we track motioncor settings in star files this is a faff, and more importantly, 2. I know quite a lot of people who motion correct on the fly and these people would not be able to use this approach.
Instead, I've taken the odd and even tilts from the tilt stack file to generate tomogram halves (Warp recommends doing it this way I think), i.e. one half will be tilt stack images 1 3 5 7 etc and the other 2 4 6 8 etc. This works quite well.
Run via: relion_tomo_train_denoiser --tilt-series-star-file AlignTiltSeries/job068/aligned_tilt_series.star --output-directory test
Once this job is finished, use Reconstruct Tomograms to generate the tomograms. Then use the output of reconstruct tomograms as the input for a wrapper for Topaz and/or CryoCARE (in the process of writing).
We should probably have a quick chat about the best way to incorporate this into the rln GUI / the general workflow! This version currently works well in the workflow but I have a better way of incorporating it which needs some alterations to the rln GUI.