In this paper we present a data-driven approach for fitting behaviors to simulated pedestrian crowds. Our method annotates agent trajectories, generated by any crowd simulator, with action-tags. The aggregate effect of ...