search

gerlichlab / looptrace

Fork (from EMBL Gitlab) of the looptrace project: https://git.embl.de/grp-ellenberg/looptrace
MIT License
2 stars 1 forks source link

Consider padding with 'linear_ramp' rather than 'edge' values #139

Closed vreuter closed 9 months ago

vreuter commented 10 months ago

When padding an ROI / spot image to account for near-boundary, when the natural ROI size doesn't conform to what's prescribed, consider using a descending linear ramp (mode='linear_ramp' as argument to the mode parameter in numpy.pad https://numpy.org/doc/1.26/reference/generated/numpy.pad.html). Consider also controlling end_values parameter to something other than 0, e.g. to the min on the other edge. This could more appropriately represent the signal decay we'd expect to see.

@TLSteinacker what do you think?

vreuter commented 10 months ago

Counterpoint: spots are expected to be rather punctate, particularly in the case of individual FISH spots rather than the regional ones, so perhaps a steeper decay than linear would be desired.

vreuter commented 10 months ago

Related to #133 #136 #135

vreuter commented 10 months ago

Per discussion with @TLSteinacker , will make method configurable but default to 'edge'.