Semi-Supervised Diversity-Aware Domain Adaptation for 3D Object detection

TL;DR

This paper introduces a semi-supervised, diversity-aware domain adaptation method for 3D object detection in autonomous vehicles, leveraging neuron activation patterns and minimal annotations to improve cross-domain generalization.

Contribution

It proposes a novel lidar domain adaptation technique based on neuron activation patterns that requires minimal annotation and incorporates continual learning-inspired post-training methods.

Findings

Outperforms linear probing and existing domain adaptation techniques

Achieves high performance with small, diverse annotated target samples

Effectively prevents weight drift from the original model

Abstract

3D object detectors are fundamental components of perception systems in autonomous vehicles. While these detectors achieve remarkable performance on standard autonomous driving benchmarks, they often struggle to generalize across different domains - for instance, a model trained in the U.S. may perform poorly in regions like Asia or Europe. This paper presents a novel lidar domain adaptation method based on neuron activation patterns, demonstrating that state-of-the-art performance can be achieved by annotating only a small, representative, and diverse subset of samples from the target domain if they are correctly selected. The proposed approach requires very small annotation budget and, when combined with post-training techniques inspired by continual learning prevent weight drift from the original model. Empirical evaluation shows that the proposed domain adaptation approach outperforms both linear probing and state-of-the-art domain adaptation techniques.