GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection

2025-03-11

AI Summary

LiDAR-based 3D detectors suffer from limited cross-domain generalization, primarily because existing domain generalization methods over-rely on global geometric coordinates (e.g., Cartesian positions), causing models to learn domain-specific absolute positional biases. Method: We propose a novel paradigm that eliminates dependence on global coordinates and instead explicitly models local point cloud structure. To this end, we introduce Gaussian Blobs (GBlobs)—a zero-parameter, highly efficient encoding scheme that applies Gaussian kernels to local neighborhoods for learning-free feature encoding, thereby enhancing domain invariance. GBlobs is plug-and-play: it integrates seamlessly into mainstream 3D detectors without modifying backbones or introducing extra parameters. Contribution/Results: On single-source and multi-source domain generalization benchmarks, our method outperforms state-of-the-art approaches by +21/13/12 mAP and +17/12/5 mAP, respectively, while preserving in-domain performance on source data.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

LiDAR-based 3D detectors need large datasets for training, yet they struggle to generalize to novel domains. Domain Generalization (DG) aims to mitigate this by training detectors that are invariant to such domain shifts. Current DG approaches exclusively rely on global geometric features (point cloud Cartesian coordinates) as input features. Over-reliance on these global geometric features can, however, cause 3D detectors to prioritize object location and absolute position, resulting in poor cross-domain performance. To mitigate this, we propose to exploit explicit local point cloud structure for DG, in particular by encoding point cloud neighborhoods with Gaussian blobs, GBlobs. Our proposed formulation is highly efficient and requires no additional parameters. Without any bells and whistles, simply by integrating GBlobs in existing detectors, we beat the current state-of-the-art in challenging single-source DG benchmarks by over 21 mAP (Waymo->KITTI), 13 mAP (KITTI->Waymo), and 12 mAP (nuScenes->KITTI), without sacrificing in-domain performance. Additionally, GBlobs demonstrate exceptional performance in multi-source DG, surpassing the current state-of-the-art by 17, 12, and 5 mAP on Waymo, KITTI, and ONCE, respectively.

Problem

• Improves cross-domain LiDAR-based 3D object detection
• Encodes local point cloud structure using Gaussian blobs
• Enhances domain generalization without additional parameters

Innovation

• Uses Gaussian blobs for local structure encoding
• Enhances cross-domain LiDAR 3D object detection
• Improves domain generalization without extra parameters