Multi-Class 3D Object Detection with Single-Class Supervision

TL;DR

This paper introduces a novel single-class supervision setting for training multi-class 3D object detectors, reducing labeling costs while maintaining performance, demonstrated through experiments on the Waymo dataset.

Contribution

It defines the single-class supervision setting and adapts various algorithms to effectively train multi-class 3D detectors under this setting.

Findings

SCS can approach full supervision performance

Proper algorithm selection is crucial for SCS effectiveness

Significant labeling cost reduction achieved

Abstract

While multi-class 3D detectors are needed in many robotics applications, training them with fully labeled datasets can be expensive in labeling cost. An alternative approach is to have targeted single-class labels on disjoint data samples. In this paper, we are interested in training a multi-class 3D object detection model, while using these single-class labeled data. We begin by detailing the unique stance of our "Single-Class Supervision" (SCS) setting with respect to related concepts such as partial supervision and semi supervision. Then, based on the case study of training the multi-class version of Range Sparse Net (RSN), we adapt a spectrum of algorithms -- from supervised learning to pseudo-labeling -- to fully exploit the properties of our SCS setting, and perform extensive ablation studies to identify the most effective algorithm and practice. Empirical experiments on the Waymo Open Dataset show that proper training under SCS can approach or match full supervision training while saving labeling costs.