Repository navigation
Confirm ShapePipe bad-pixel handling in cutouts (ignore vs noise-fill) #835
Description
Activity
@aguinot says:
I am not sure there is an argument to run maximask at all. The main things we were interested in from maximasks are easy to flag at the single exposure level. Like: saturation, bleeding, bad columns, etc. The only potential thing are cosmic rays. Stephen is masking them but I don't know how reliable his masking is. From visual inspection it is mostly ok.
Both: zero weight and noise fill. On
develop,prepare_ngmix_weights(ngmix.py:1612–1668) masks every pixel with a nonzero instrument flag or zero exposure weight: its weight goes to 0 and, under the production default (BLEND_HANDLING = noisefill), its value is replaced by noise at the per-pixel background RMS before metacal. Epochs more than 1/3 flagged are dropped. ngmix's metacal fills nothing itself, so this fill is what keeps defects out of the sheared images; #811 doesn't change it.So MaxiMask on exposures is redundant for bad columns, saturation and bleeding. It could only add cosmic rays and faint trails the flags miss.
One gap: under
uberseg(#814), flagged pixels get zero weight but are not filled, so raw defect values pass through metacal; noted on #814 to fix before that A/B.— Claude (Opus) on behalf of Cail
Confirm whether ShapePipe ignores or noise-fills masked bad pixels within a postage-stamp cutout. Needed to judge whether running Maximask on individual exposures would be redundant. Action item: Cail / Martin.
Context: Paris meeting notes