Granularity Beyond Hardware: Super-Resolution for Enhanced Particle Reconstruction in Calorimeters

The spatial resolution of calorimeters is a crucial parameter in particle detector design which is often constrained by cost and construction complexity. We propose machine learning based super-resolution as a software technique to increase effective calorimeter granularity, enhancing a detector’s performance with zero changes to hardware. Upsampling is performed with a transformer-based continuous normalizing flow conditioned on low-granularity calorimeter data. We showcase the impact of our approach on particle reconstruction using a generic particle flow algorithm based on machine learning. Our results demonstrate that super-resolution can be readily applied at current and future collider experiments.