Shake-The-Box with PeakCNN: Production ready ML for particle tracking at very high seeding densities?

Abstract

The Peak-CNN approach allows for peak detection in images even at high particle concentrations. Such a more reliable detetcion then leads to better performances of the downstream processing such as triangulation or IPR. As a supervised learning method Peak-CNN requires labeled training data. This talk showcases how such training data can be obtained in a simple and efffective manner for a testcase of the 2nd LPT/DA challenge