TrajNet is a large-scale trajectory-based activity benchmark that provides a unified evaluation system for testing methods. The challenge involves predicting 3,161 human trajectories, observing 8 consecutive ground truth values (3.2 seconds) for each trajectory in world plane coordinates, i.e., t-7, t-6, …, t, and predicting the next 12 trajectories (4.8 seconds), i.e., t+1, …, t+12.

TrajNet is a superset of multiple datasets, including various trajectory-based activity prediction datasets, such as BIWI Hotel, Crowds UCY, MOT PETS, and Stanford Drone Dataset (SDD).