taskGNU Astronomy Utilities - Tasks: task #14980, Better way to grow true clumps...

 
 

You are not allowed to post comments on this tracker with your current authentication level.

task #14980: Better way to grow true clumps (full segmentation/deblending)

Submitter:  Mohammad Akhlaghi <makhlaghi>
Submitted:  Fri 06 Jul 2018 11:32:15 AM UTC
   
 
Should Start On:  Thu 05 Jul 2018 10:00:00 PM UTC Should be Finished on:  Thu 05 Jul 2018 10:00:00 PM UTC
Category:  Segment Priority:  5 - Normal
Item Group:  Enhancement Status:  Postponed
Privacy:  Public Assigned to:  None
Percent Complete:  0% Open/Closed:  Open
Effort:  0.00

Tue 30 Oct 2018 12:58:08 PM UTC, comment #3: 

This is also an interesting non-parametric (but machine-learning based: hard to configure) deblending method which we might be able to implement (after building a machine learning infra-structure within Gnuastro):

https://arxiv.org/abs/1810.10098

Since we know the number of clumps in each detection, we know the ultimate number of objects to deblend. In a large survey, we can also use this information to find detections with one or no clumps (which we may assume are not blended) and use them to define our training set.

Mohammad Akhlaghi <makhlaghi>
Group administrator
Sun 29 Jul 2018 09:46:12 PM UTC, comment #2: 

The attached slides PDF file was apparently not submitted in the previous comment. The slides are also available in this link:

http://akhlaghi.org/pdf/noisechisel-mtobjects.pdf

Mohammad Akhlaghi <makhlaghi>
Group administrator
Sun 29 Jul 2018 09:41:10 PM UTC, comment #1: 

As a complement to the main task description, this task is about how the true peaks are "grown", not about detection (separating astronomical objects from the background) or finding true peaks.

During the first annual SUNDIAL meeting, we done some tests comparing NoiseChisel+Segment to MTObjects, and the results are available in the attached slides. So at least on this dataset, and with these versions of the two softwares (most recent as of the meeting), NoiseChisel does better at detection of extended low surface brightness signal and also the completeness of the true peaks it finds.

As described in the main task definition (previous post in this task), the thing we should test in Segment to (hopefully) improve it is how true peaks are "grown" to cover the full detection. The current growth strategy is too simplistic and can be greatly improved.

Mohammad Akhlaghi <makhlaghi>
Group administrator
Fri 06 Jul 2018 11:32:15 AM UTC, original submission:  

In Segment, after true clumps are found, they are "grown" simply by sorted pixels (as you see in Figure 10 of the NoiseChisel paper).

This is only a very crude first approximation to identifying "objects" over the detection and has many problems. For example, an "object" associated with a small peak can easily cover a very large area of a large galaxy's diffuse underlying flux for example.

A better way to grow the true peaks/clumps over NoiseChisel's detected (non-background) pixels may be to use "Max-tree"s like what is used in MTObjects.

From the example images in that paper (for example Figures 7, 8, 12 and 13), you see that because of this higher-level structure, the smaller peaks over a large extended region are much more accurately separated from the diffuse flux of the background and they don't block it (like they do in segment currently, which is not good).

A simple visual inspection thus suggests that this is probably a much better way to grow the true clumps which is worth testing (and hopefully implementing) in Segment.

I will try to do this after I find some free time (my current project finishes), but if anyone is interested until then, you can start it already, and I would be happy to help ;-)...

Mohammad Akhlaghi <makhlaghi>
Group administrator

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

No files currently attached

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -email is unavailable- added by makhlaghi (Submitted the item)
  •  

    There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.

     

    Follow 5 latest changes.

    Date Changed by Updated Field Previous Value => Replaced by
    2018-09-19 makhlaghi SummaryMax-trees to grow true clumps Better way to grow true clumps (full segmentation/deblending)
    2018-07-19 makhlaghi Carbon-CopyRemoved -email is unavailable- -
    2018-07-19 makhlaghi Carbon-CopyRemoved -email is unavailable- -
    2018-07-06 makhlaghi Carbon-Copy- Added -email is unavailable-
        Carbon-Copy- Added -email is unavailable-

    Back to the top

    Powered by Savane 3.13-f8d8.
    Corresponding source code