Add placeholder to reject obvious problematic objects before association

[rearmstr]

Previous surveys have shown that there is significant improvement in the association performance if a basic algorithm is used before association to reject obvious artifacts.

This issue is only to put in a placeholder in the code that can be fully defined later.

[johannct]

@rearmstr can you add some references?

[wmwv]

@rearmstr Do you mean

1. Just simply using the available flags in the diaSrc table? We can certainly get significant improvement our current results by just doing this.
2. More actively using the already-calculated properties such as dipole scores to screen for good detections?
3. More complicated independent/additional pixel-level analysis, including potential ML trained real/bogus classifiers.

[rearmstr]

We discussed this somewhat on the last desc-dia call. The suggestion was from Rick Kessler and his experience in DES (I'm not sure how to @ Rick into this conversation). They found that a pixel level ML trained classifier before association made a big difference. His suggestion was that we build the option into the framework now since it needs pixels, so that it is already in place. I'm sure we can do better by using the information already available in the catalog, but my impression was that this was a step beyond that.

[wmwv]

Thanks for the clarification.

[RobertLuptonTheGood]

I'm a bit confused here. The AP team is planning to use some true/bogus classifier as part of the processing, and these days it's likely to be pixel-based (e.g. a CNN). Is this another layer of DESC processing beyond this?

[rearmstr]

I think the question is whether we can use the structure the AP pipeline within the DRP difference imaging pipeline. We currently do not use the AP association algorithm because we have different needs and requirements. Because of this, we need to explicitly add the classifier into the existing association algorithm wherever it comes from. If we can use whatever the AP team is planning that would be great.