Rethinking Voxelization and Classification for 3D Object Detection

TL;DR

This paper introduces a fast dynamic voxelizer and a lightweight classification sub-head to enhance the speed and accuracy of 3D object detection from LiDAR data, enabling real-time performance without sacrificing reliability.

Contribution

It proposes a novel fast voxelization method and a lightweight classification head to improve inference speed and precision in 3D detection models.

Findings

Improved inference speed with the dynamic voxelizer.

Enhanced detection precision with the lightweight classification head.

Code implementation is publicly available.

Abstract

The main challenge in 3D object detection from LiDAR point clouds is achieving real-time performance without affecting the reliability of the network. In other words, the detecting network must be confident enough about its predictions. In this paper, we present a solution to improve network inference speed and precision at the same time by implementing a fast dynamic voxelizer that works on fast pillar-based models in the same way a voxelizer works on slow voxel-based models. In addition, we propose a lightweight detection sub-head model for classifying predicted objects and filter out false detected objects that significantly improves model precision in a negligible time and computing cost. The developed code is publicly available at: https://github.com/YoushaaMurhij/RVCDet.