Towards 3D Object Detection with 2D Supervision

TL;DR

This paper presents a hybrid training framework for 3D object detection that leverages abundant 2D labels and temporal 2D supervision to reduce reliance on expensive 3D annotations, achieving near fully-supervised performance.

Contribution

It introduces a novel temporal 2D supervision method that transforms 3D predictions into 2D for training without 3D annotations, significantly reducing annotation costs.

Findings

Achieves nearly 90% of fully-supervised performance on nuScenes with only 25% 3D annotations.

Proposes a temporal 2D transformation using homography wrapping and 2D box deduction.

Demonstrates effective use of 2D supervision for 3D object detection.

Abstract

The great progress of 3D object detectors relies on large-scale data and 3D annotations. The annotation cost for 3D bounding boxes is extremely expensive while the 2D ones are easier and cheaper to collect. In this paper, we introduce a hybrid training framework, enabling us to learn a visual 3D object detector with massive 2D (pseudo) labels, even without 3D annotations. To break through the information bottleneck of 2D clues, we explore a new perspective: Temporal 2D Supervision. We propose a temporal 2D transformation to bridge the 3D predictions with temporal 2D labels. Two steps, including homography wraping and 2D box deduction, are taken to transform the 3D predictions into 2D ones for supervision. Experiments conducted on the nuScenes dataset show strong results (nearly 90% of its fully-supervised performance) with only 25% 3D annotations. We hope our findings can provide new insights for using a large number of 2D annotations for 3D perception.