Uploaded image for project: 'Data Management'
  1. Data Management
  2. DM-11720

Investigate feasibility to use colors to estimate source boundary boxes

    Details

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

      Description

      For the deblender to be optimal in the long term survey, which includes possibly deblending entire images as a single blend, we need to be able to define a bounding box for each source that is a small fraction of the total image size (see the issue created by Peter Melchior
      at https://github.com/fred3m/deblender/issues/14).

      Issue DM-11718 has been created to develop an effective algorithm to do this, but will not likely be added to the stack until after testing has been run on a large patch of real HSC images.

      This ticket is created to track work that has already been done to test whether or not such an algorithm might be possible, and resulted in this ipython notebook, which shows that such a procedure is likely to exist.

      It also shows the possibility of performing multi-band object detection in the future, although discussion with Peter Melchior indicate that this might be difficult (and/or time consuming) in practice.

        Attachments

          Issue Links

            Activity

            Hide
            fred3m Fred Moolekamp added a comment -

            Hi Peter, I created this ticket to reflect the work I did the other day, showing that using the peak SED to estimate the size of objects might be a viable solution. I also opened another ticket to cover the actual implementation of this, so would you mind just checking off that I created the notebook listed in the description?

            Show
            fred3m Fred Moolekamp added a comment - Hi Peter, I created this ticket to reflect the work I did the other day, showing that using the peak SED to estimate the size of objects might be a viable solution. I also opened another ticket to cover the actual implementation of this, so would you mind just checking off that I created the notebook listed in the description?
            Hide
            pmelchior Peter Melchior added a comment -

            This approach holds promise. Here we still require initial peak detection to provide a reference pixel for the color differences. That allows us to determine a reasonable bounding box per object. It's not clear how it'll work with multiple objects having quite similar colors.

            In general: How to ideally couple color approaches with detection remains to be seen (but isn't part of this ticket)

            Show
            pmelchior Peter Melchior added a comment - This approach holds promise. Here we still require initial peak detection to provide a reference pixel for the color differences. That allows us to determine a reasonable bounding box per object. It's not clear how it'll work with multiple objects having quite similar colors. In general: How to ideally couple color approaches with detection remains to be seen (but isn't part of this ticket)

              People

              • Assignee:
                fred3m Fred Moolekamp
                Reporter:
                fred3m Fred Moolekamp
                Reviewers:
                Peter Melchior
                Watchers:
                Fred Moolekamp, Peter Melchior
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved:

                  Summary Panel