Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-7503

Discuss deblender requirements with stakeholders

    Details

    • Type: Story
    • Status: Done
    • Resolution: Done
    • Fix Version/s: None
    • Component/s: meas_deblender
    • Labels:
      None

      Description

      Chat with stakeholders in the deblender (at least RHL, Jim Bosch) to get a better feeling for their vision of how it should be developed in the future. Understand the architectural requirements for multi-band deblending.

        Attachments

          Issue Links

            Activity

            Hide
            swinbank John Swinbank added a comment -

            We should make sure that information uncovered in these discussions is captured somewhere. Are there any notes? Perhaps it'll all be clear once DM-7504 & DM-7505 are done.

            Show
            swinbank John Swinbank added a comment - We should make sure that information uncovered in these discussions is captured somewhere. Are there any notes? Perhaps it'll all be clear once DM-7504 & DM-7505 are done.
            Hide
            rearmstr Bob Armstrong added a comment -

            Here is what we discussed.

            In the short term, we need to get the removal of degenerate templates, to avoid shredding galaxies, working as it was in SDSS. This also needs to be able to work during the multiBand processing such that we can consistently remove peaks from all filters.

            We decided to hold off on doing a building a full multicolor deblender, because of the complexity and because it has some overlap with what others have been working on. However, as we design new features of the deblender we need to make it easy to add color information later on. We also need to be able to work with multiBand which will force us to handle things between different filters.

            We discussed two extensions to the existing algorithm. First, we talked about adding the option to build the galaxy templates from analytic models. This may help some of the problems we have seen in the centers of clusters and would allow us to propagate deblending information from one band to another or from the coadds to the single visits. The second extension was to do a hierarchical deblend where you try to model the brightest objects first and then remove them one by one and proceed to the fainter ones. Erin Sheldon is doing something like this in DES. He originally tried to do simultaneous fitting of multiple objects, but found that did not work very well. We talked about inviting him down for a day to discuss his experience.

            To implement these new extensions we will probably need to refactor the current deblender. We decided to hold off on this question until we implement the degenerate template fix.

            Show
            rearmstr Bob Armstrong added a comment - Here is what we discussed. In the short term, we need to get the removal of degenerate templates, to avoid shredding galaxies, working as it was in SDSS. This also needs to be able to work during the multiBand processing such that we can consistently remove peaks from all filters. We decided to hold off on doing a building a full multicolor deblender, because of the complexity and because it has some overlap with what others have been working on. However, as we design new features of the deblender we need to make it easy to add color information later on. We also need to be able to work with multiBand which will force us to handle things between different filters. We discussed two extensions to the existing algorithm. First, we talked about adding the option to build the galaxy templates from analytic models. This may help some of the problems we have seen in the centers of clusters and would allow us to propagate deblending information from one band to another or from the coadds to the single visits. The second extension was to do a hierarchical deblend where you try to model the brightest objects first and then remove them one by one and proceed to the fainter ones. Erin Sheldon is doing something like this in DES. He originally tried to do simultaneous fitting of multiple objects, but found that did not work very well. We talked about inviting him down for a day to discuss his experience. To implement these new extensions we will probably need to refactor the current deblender. We decided to hold off on this question until we implement the degenerate template fix.
            Hide
            swinbank John Swinbank added a comment -

            Thanks Bob!

            Is there anything beyond the above list that you think should be included in DM-7504? If not, maybe we can close that as done as well?

            As we've currently scoped it, in order to complete the epic, we should at least come up with a design for the refactored deblender (that's DM-7505). That can certainly wait until the shredding issue has been resolved, though.

            Show
            swinbank John Swinbank added a comment - Thanks Bob! Is there anything beyond the above list that you think should be included in DM-7504 ? If not, maybe we can close that as done as well? As we've currently scoped it, in order to complete the epic, we should at least come up with a design for the refactored deblender (that's DM-7505 ). That can certainly wait until the shredding issue has been resolved, though.

              People

              • Assignee:
                rearmstr Bob Armstrong
                Reporter:
                swinbank John Swinbank
                Watchers:
                Bob Armstrong, Fred Moolekamp, John Swinbank
              • Votes:
                0 Vote for this issue
                Watchers:
                3 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved:

                  Summary Panel