Details

    • Type: Story
    • Status: Done
    • Priority: Major
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_base
    • Labels:
      None
    • Templates:
    • Story Points:
      8
    • Epic Link:
    • Sprint:
      DRP F16-1, DRP F16-2, DRP F16-3
    • Team:
      Data Release Production

      Description

      In DM-4882, we observed a number of centroids measured while running the lsst_dm_stack_demo routines fall outside their associated Footprints. This was seen with both the NaiveCentroid and the SdssCentroid centroiders.

      For the purposes of DM-4882 we quieted the warnings arising from this, but we should investigate why this is happening and, if necessary, weed out small Footprints entirely.

        Issue Links

          Activity

          Hide
          jbosch Jim Bosch added a comment -

          Would turning on detection.doTempLocalBackground help make these cases disappear?

          Probably, but I think that really ought to be a separate issue.

          Show
          jbosch Jim Bosch added a comment - Would turning on detection.doTempLocalBackground help make these cases disappear? Probably, but I think that really ought to be a separate issue.
          Hide
          pgee Perry Gee added a comment -

          Jim Bosch One thing we didn't settle was whether there should be a default distance from footprint peak test. The code you reviewed does not have this test in place for any of the algorithms. Only the check for inside footprint test is automatically performed. Is that what you had in mind?

          Show
          pgee Perry Gee added a comment - Jim Bosch One thing we didn't settle was whether there should be a default distance from footprint peak test. The code you reviewed does not have this test in place for any of the algorithms. Only the check for inside footprint test is automatically performed. Is that what you had in mind?
          Hide
          jbosch Jim Bosch added a comment -

          Good question. Let's not do it just yet. I imagine a 3-pixel distance would be safe and helpful for the small, spurious objects we've focused on so far, but I'm a bit worried that this would spuriously flag some large objects with shallow profiles.

          Show
          jbosch Jim Bosch added a comment - Good question. Let's not do it just yet. I imagine a 3-pixel distance would be safe and helpful for the small, spurious objects we've focused on so far, but I'm a bit worried that this would spuriously flag some large objects with shallow profiles.
          Hide
          pgee Perry Gee added a comment -

          With GaussianCentroid, out of 4200 sources in 5 bands, the inside check eliminates 687. The maxDistToPeak < 3 eliminates 1031, and contains almost all of the failures in the first test.

          The SdssCentroid and NaiveCentroid checks eliminate less than 100 with either test.

          Show
          pgee Perry Gee added a comment - With GaussianCentroid, out of 4200 sources in 5 bands, the inside check eliminates 687. The maxDistToPeak < 3 eliminates 1031, and contains almost all of the failures in the first test. The SdssCentroid and NaiveCentroid checks eliminate less than 100 with either test.
          Hide
          jbosch Jim Bosch added a comment -

          Could you take a look at the sources eliminated by the check SdssCentroid centroid only, and just by-eye inspect the 10 with the highest SNR? If those all look like junk or cases where the centroid really did fail, I think we're good to merge.

          Show
          jbosch Jim Bosch added a comment - Could you take a look at the sources eliminated by the check SdssCentroid centroid only, and just by-eye inspect the 10 with the highest SNR? If those all look like junk or cases where the centroid really did fail, I think we're good to merge.

            People

            • Assignee:
              pgee Perry Gee
              Reporter:
              swinbank John Swinbank
              Reviewers:
              Jim Bosch
              Watchers:
              Jim Bosch, John Swinbank, Michael Wood-Vasey, Nate Lust, Paul Price, Perry Gee
            • Votes:
              0 Vote for this issue
              Watchers:
              6 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Development

                  Agile