PIE: A Large-Scale Dataset and Models for Pedestrian Intention Estimation and Trajectory Prediction

June 2020

tl;dr: Pedestrian intension estimation dataset and achieves ~80% of accuracy and ~90% F1 score.

Overall impression

Intention and trajectory prediction are different. Trajectory prediction are only effective when the pedestrians have started crossing or about to do so, so basically the algorithm react to an action instead of anticipating it.

Intention helps trajectory prediction. This is similar conclusion as in IntentNet.

The key question is “does this pedestrian want to cross the street?”

Key ideas

Technical details