An Empirical Study on Knowledge Transfer under Domain and Label Shifts in 3D LiDAR Point Clouds

2026-01-09

AI Summary

This work addresses the challenge of simultaneously handling sensor-induced domain shift and dynamic label evolution—such as class splitting, expansion, and insertion—in real-world 3D perception scenarios, a setting inadequately tackled by existing methods. We propose ROAD, the first benchmark for joint domain and label shift evaluation in 3D point clouds, built upon Waymo, NuScenes, and Argoverse2. The framework supports systematic experiments across zero-shot transfer, linear probing, and continual learning strategies. By explicitly modeling three forms of label evolution and establishing strong baselines, our study exposes the limitations of current approaches under realistic shifts and provides a reliable evaluation platform along with performance references for robust 3D perception research.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

For 3D perception systems to be practical in real-world applications -- from autonomous driving to embodied AI -- models must adapt to continuously evolving object definitions and sensor domains. Yet, research on continual and transfer learning in 3D point cloud perception remains underexplored compared to 2D vision -- particularly under simultaneous domain and label shifts. To address this gap, we propose the RObust Autonomous driving under Dataset shifts (ROAD) benchmark, a comprehensive evaluation suite for LiDAR-based object classification that explicitly accounts for domain shifts as well as three key forms of label evolution: class split, class expansion, and class insertion. Using large-scale datasets (Waymo, NuScenes, Argoverse2), we evaluate zero-shot transfer, linear probe, and CL, and analyze the impact of backbone architectures, training objectives, and CL methods. Our findings reveal limitations of existing approaches under realistic shifts and establish strong baselines for future research in robust 3D perception.

Problem

• domain shift
• label shift
• 3D point clouds
• knowledge transfer
• continual learning

Innovation

• domain shift
• label shift
• 3D point clouds
• continual learning
• knowledge transfer