Open stscijgbot-jp opened 2 months ago
Comment by Brett Graham on JIRA:
A modified abs_deriv
that does not have the edge effects (returns all 5s for the above example) is at #8635
Comment by Brett Graham on JIRA:
Here is a regression test run with the above linked fixed abs_deriv
:
https://plwishmaster.stsci.edu:8081/blue/organizations/jenkins/RT%2FJWST-Developers-Pull-Requests/detail/JWST-Developers-Pull-Requests/1595/tests
showing 62 test failures. A more thorough investigation of these differences will be needed but many have different OUTLIER
dq bits as expected.
Issue JP-3683 was created on JIRA by Brett Graham:
The
abs_deriv
function used in outlier detection assumes 0 values for off-edge pixels which leads to edge effects.https://github.com/spacetelescope/jwst/blob/033b81756c2cc04c0a38930d1c88b3c931586bd7/jwst/outlier_detection/outlier_detection.py#L517
For example providing the following as input:
I would expect that the output should be 5 everywhere in the result (since for example the absolute difference between for example arr[4, 4] and arr[3, 4] is 5