OpenBox: Annotate Any Bounding Boxes in 3D

TL;DR

OpenBox is a two-stage automatic annotation pipeline that leverages 2D vision foundation models to generate high-quality 3D bounding boxes, reducing annotation costs and improving detection of unseen objects in autonomous driving datasets.

Contribution

We introduce OpenBox, a novel method that automates 3D bounding box annotation by integrating 2D vision models and categorizing object states, eliminating the need for self-training and enhancing annotation quality.

Findings

Improves annotation accuracy over baseline methods.

Reduces computational overhead in 3D annotation.

Effective across multiple autonomous driving datasets.

Abstract

Unsupervised and open-vocabulary 3D object detection has recently gained attention, particularly in autonomous driving, where reducing annotation costs and recognizing unseen objects are critical for both safety and scalability. However, most existing approaches uniformly annotate 3D bounding boxes, ignore objects' physical states, and require multiple self-training iterations for annotation refinement, resulting in suboptimal quality and substantial computational overhead. To address these challenges, we propose OpenBox, a two-stage automatic annotation pipeline that leverages a 2D vision foundation model. In the first stage, OpenBox associates instance-level cues from 2D images processed by a vision foundation model with the corresponding 3D point clouds via cross-modal instance alignment. In the second stage, it categorizes instances by rigidity and motion state, then generates adaptive bounding boxes with class-specific size statistics. As a result, OpenBox produces high-quality 3D bounding box annotations without requiring self-training. Experiments on the Waymo Open Dataset, the Lyft Level 5 Perception dataset, and the nuScenes dataset demonstrate improved accuracy and efficiency over baselines.