Learning-Deep-Learning

MPV-Nets: Monocular Plan View Networks for Autonomous Driving

September 2020

tl;dr: Project 3D object detection into BEV map to train a better driving agent.

Overall impression

Monocular 3D object detection in a way similar to Deep3DBox. Then the 3D object detection results are rendered into a BEV (Plan view). Having access to this plan view reduces collisions by half.

Key ideas

Technical details

Notes