The purpose of this repo is to support automated QA of DECaLs imaging used to select DESI spectroscopic targets. The three main components are described below.
Process the community pipeline output for all ~120K candidate DECaLS exposures to produce (inverse-variance weighted) downsampled thumbnails.
More details here.
Allow rapid visual inspection of about 5% of the thumbnail images to collect expert labels for the main classes of bad exposures.
Try it out here.
Train various algorithms using the expert labeled data to automate the identification of bad exposures from the full dataset.
More details here.