Details
-
Type:
RFC
-
Status: Adopted
-
Resolution: Unresolved
-
Component/s: DM
-
Labels:None
Description
DMS-REQ-0274 requires that the image cutouts provided in alert packets contain WCS, PSF, variance, and mask information. Providing the cutouts as serializations of afw.Exposure would require alert packet consumers to use the stack to read the cutouts, which is not desirable for portability.
We propose instead that alert packets contain a serialization of the image cutouts converted to astropy.nddata.CCDData objects. This class currently provides a data structure with numpy.ndarray-like data, uncertainty, mask, and flag planes; metadata, and WCS. Users would be able to read and interpret alert cutouts using astropy only.
The DMSR-required PSF is not presently included in CCDData, so we would need to develop a suitable PSF representation and attempt to upstream it to astropy if this RFC is approved.
Attachments
Issue Links
- is triggered by
-
- Done
- is triggering
-
- To Do
-
- To Do
-
- Done
-
- Done
Activity
I'm fine with this as a pragmatic approach. The important point is for us to be upfront about exactly what we are saying.
Note that although CCDData includes API support for flags, it does not appear to come with serialization support for that data - this is something we've recently run into in SPHEREx and are starting to work on.
I didn't see any mention of this here or on DM-24325's ticket or PR. Was this issue encountered in the course of doing the work?
Gregory Dubois-Felsmann I don't think we recognized that at the time, no.
Thanks Tim Jenness, this is helpful.
What I would like to propose is that our interface be "whatever FITS CCDData writes"--that is, we commit to a FITS serialization, but rather than define our FITS format on a blank page we follow along with CCDData as it updates. (Again, at minimum we'd have to get them to add PSF support.) I suspect you'll say that's a bad idea due to version instabilities etc. I'd just prefer we not define our own thing that requires our own new implementation if we can avoid it. I'd welcome your thoughts on how to approach this.
I think the PSF we include should be a simple one appropriate only for the cutout in question (which as you say is small)--and similarly the WCS can and should be a simple local WCS.