The Legacy Survey of Space and Time holds the promise of revolutionizing astrophysics with
unprecedented volumes of data, but this necessitates the development of methods for
automated data processing. Foundation models like the Segment Anything Model (SAM, Kirillov+2023)[1] from MetaAI offer a promising solution. However, astrophysical images are inherently out of sample for a model trained on every-day world images, so we tested SAM’s performance on astrophysical extended sources in space based galaxy images (GALEX) and ground based galaxy and light echo images (ATLAS, DECAM) testing the potential of its
application to data from the LSST.
rodiat_rcw2025.pptx.pdf (1.8 MB)