Reference images for galaxy de-blending

Data Management
#1

We have just kicked off (today) a modest project within the Galaxies science collaboration to work on de-blending issues. If anyone would like to be included on the mailing list, please let me know.

The initial goal will be to create a reference sample of images (difficult cases) where truth is known. We’re starting with HST images and simulated images from the Illustris simulations, but if there are other sets of images (e.g. challenge data-sets for strong lensing), it would be good to know about them and potentially include them in the set of reference images.

We’re also hoping to make progress on defining/refining science metrics for helping to evaluate deblending/photometry algorithms. Examples might be what fraction of z=4 Lyman-break galaxies are lost because they are photometrically blended with a neighbor, or what fraction of galaxy-galaxy gravitational lens candidates cannot be identified. Ideas on metrics would be most welcome.

Beyond that, we hope to run the current LSST de-blender and experiment with alternatives. But we aren’t funded for any serious code development, so this would all be quick & dirty exploration.

#2

This sounds tremendously useful; building a collection of deblender test cases has long been our to-do list, and we’d love more metrics that can help us determine between good deblending and bad deblending without some sort of ad-hoc human inspection. Metrics that we can apply to real data (i.e. those we can measure without having ground truth) would be even more useful, but those would obviously be limited in what they can tell us.

Please do add me (jbosch@astro.princeton.edu) to the mailing list. I probably won’t be able to contribute any effort to the project, but I may be able to provide some help in understanding what our current deblender is doing.

#3

I am also very interested by this. I have the project to use the verification datasets (CFHT, HCS, DES) as test benches for deblending algorithms. Having reference images overlapping the verification datasets region of interest would be very useful.