The workshop includes a trajectory forecasting challenge focusing on both (i) human-human and (ii) human-space interactions. For a fair comparison among different models two different datasets are considered:
TrajNet++ focused on urban scenarios;
THÖR focused on human-robot interactions in indoor controlled environments.
Authors of the top one submission of both benchmarks will have the opportunity to showcase their work as spotlight talks during the workshop.
Human-Human challenge: Interactions among humans in crowded scenarios typically requires a huge amount of data to extract subtle interactions. Furthermore, in many scenes, space does not directly influence human dynamics which are manly affected by other humans. We propose a human-human sub-challenge to analyze human motions both in both indoor and outdoor environments.
Human-Space challenge: Scene topology, constraints and semantics represent a valuable source of additional information which can improve performance for trajectory forecasting tasks. The majority of studies mainly focused on human interactions without taking into account any other source of information. To this end, we propose a human-space sub-challenge which focuses on modeling human-space perception providing semantic annotations of proposed scenes.
TrajNet++ challenge (Round 3) is now open! Submissions deadline: August 24, 2020.