search

alihaydaroglu / suite3d

Fast, accurate, volumetric cell detection. Developed for Light Beads Microscopy, usable for other volumetric 2P. In development
6 stars 0 forks source link

some recordings have bright blood vessels in correlation map #19

Closed alihaydaroglu closed 1 year ago

alihaydaroglu commented 1 year ago

image image this only happens occasionally, but it's annoying when it does

alihaydaroglu commented 1 year ago

raising the normalization (sdmov) to an exponent seems to address the issue? I think this happens because of aggressive normalization, so dark but noisy areas become too bright

Title in images below are the "degree" of normalization (e.g. exponential we raise sdmov to). 0 is no normalization, 1.0 is standard normalization, 0.8 is with sdmov**0.8

I will add this as a parameter and test on problematic datasets. This one is from /mnt/md0/runs/s3d-AH007_2022-03-28_2-4-5

image image image

alihaydaroglu commented 1 year ago

sdnorm_exp being set lower fixes it for most cases. For smaller blood vessels, sometimes the gaussian neuropil filter creates artifacts on either side of the vessel. In this case I'd recommend setting params['npil_filt_type'] = unif and doing a sweep of neuropil filter sizes.