Are Dense Labels Always Necessary for 3D Object Detection from Point Cloud?

TL;DR

This paper introduces SS3D++, a framework that reduces annotation costs for 3D object detection by using sparse labels and progressively generating fully-annotated scenes, achieving competitive results with less supervision.

Contribution

The paper proposes a novel sparse annotation strategy and an iterative learning scheme to generate confident fully-annotated scenes, significantly reducing annotation costs while maintaining high detection performance.

Findings

Achieves comparable performance to fully-supervised methods with 5x less annotation cost on KITTI.

Attains 90% of state-of-the-art performance on Waymo with 15x less annotation cost.

Unlabeled scenes further improve detection accuracy.

Abstract

Current state-of-the-art (SOTA) 3D object detection methods often require a large amount of 3D bounding box annotations for training. However, collecting such large-scale densely-supervised datasets is notoriously costly. To reduce the cumbersome data annotation process, we propose a novel sparsely-annotated framework, in which we just annotate one 3D object per scene. Such a sparse annotation strategy could significantly reduce the heavy annotation burden, while inexact and incomplete sparse supervision may severely deteriorate the detection performance. To address this issue, we develop the SS3D++ method that alternatively improves 3D detector training and confident fully-annotated scene generation in a unified learning scheme. Using sparse annotations as seeds, we progressively generate confident fully-annotated scenes based on designing a missing-annotated instance mining module and reliable background mining module. Our proposed method produces competitive results when compared with SOTA weakly-supervised methods using the same or even more annotation costs. Besides, compared with SOTA fully-supervised methods, we achieve on-par or even better performance on the KITTI dataset with about 5x less annotation cost, and 90% of their performance on the Waymo dataset with about 15x less annotation cost. The additional unlabeled training scenes could further boost the performance.