Decide which products to include. Current contenders: segmentation maps, object classification from survey (star / galaxy, galaxy type, ...), parametric parameters (flux, size, elliptiicity, PA, ...)
Decide on a format for such data.
Select one survey (DESI?) and build out pipeline to ingest such data into AstroPile
Resources needed
Any subset of the items below:
pen and paper
enthusiasm
python experience
astropy to handle fits files, probably
huggingface datasets
internet connection
Detailed description
There are tons of ML products that could come out of AstroPile, and many of them would be trying to reproduce classical analyses but with improvements on precision or speed. Providing a curated interface with classical data products alongside AstroPile would make for easy benchmarking, and also provide basic "ground truth" to train on in some cases. Further, having object classifications could make for some specialized applications where one wishes to train on a particular object type (star/galaxy), or size (to fit in a particular stamp size).
DESI may be an easy testbed for this kind of setup since it has all such data products over a large area of sky with a uniform processing pipeline. The products are described/located on this site: https://www.legacysurvey.org/dr10/files/
Design and implement scheme for including survey reduced data products
Many of these surveys produce data products (ie segmentation maps), this could be valuable information depending on the project.
Contacts: @ConnorStoneAstro Participants: @ConnorStoneAstro + anyone interested
Goals and deliverable
Resources needed
Any subset of the items below:
Detailed description
There are tons of ML products that could come out of AstroPile, and many of them would be trying to reproduce classical analyses but with improvements on precision or speed. Providing a curated interface with classical data products alongside AstroPile would make for easy benchmarking, and also provide basic "ground truth" to train on in some cases. Further, having object classifications could make for some specialized applications where one wishes to train on a particular object type (star/galaxy), or size (to fit in a particular stamp size).
DESI may be an easy testbed for this kind of setup since it has all such data products over a large area of sky with a uniform processing pipeline. The products are described/located on this site: https://www.legacysurvey.org/dr10/files/