With the current (DES, KIDs, HSC) and future (LSST, Euclid, WFIRST) generation of large scale deep surveys, we will have to deal with an increasing number of crowded fields in which a high number of objects are blended. This poses a challenge both in terms of photometry (recovering the flux of individual objects) as well as for shape measurements (one of the main science goals) and galaxy morphology.
We organize this 1-day workshop to gather astronomers interested in exploring the use of deep learning methods to tackle image deblending. The scope is to discuss the creation of a data/science challenge on the topic to foster the emergence of new algorithms. We will need to define a score (i.e. what does "deblending" means computationally) and a strategy to gather labeled datasets or produce them.
Just before and after lunch, the participants will be able to give presentations on their work on this topic (or closely related). The talks should aim at preparing the afternoon discussion by focusing on one or both topics (metric & dataset).
Then we will brainstorm on the two main topics related to the challenge :
The event will be hosted at the François Arago Center (FACe).
There will be a possibility to join remotely the workshop.