Learning-Deep-Learning

RefinedMPL: Refined Monocular PseudoLiDAR for 3D Object Detection in Autonomous Driving

March 2020

tl;dr: Sparsify pseudo-lidar points for monocular 3d object detection.

Overall impression

The paper is based on the work of Pseudo-lidar. The main contribution seems to be the faster processing time, and the performance gain is not huge.

Both the unsupervised and supervised method identify foreground regions using 2D image, then perform a distance stratified sampler to downsample the point cloud.

Key ideas

Technical details

Notes