DB-Bench: Benchmarking Deblenders for LSST DESC Using the Blending ToolKit

Abstract

Blending will be a major source of systematic uncertainty in downstream science analyses of LSST data. We benchmark the performance of several deblenders, leveraging the Blending ToolKit (BTK) to perform rigorous, end-to-end testing. This benchmark incorporates key deblending algorithms, including SourceExtractor, SCARLET, and DeepDISC, with the goal of comparing their effectiveness in handling blended galaxy images from LSST/Rubin simulations. A key focus is characterizing algorithm performance…