DPGLA: Bridging the Gap between Synthetic and Real Data for Unsupervised Domain Adaptation in 3D LiDAR Semantic Segmentation

2025-10-27

AI Summary

This work addresses unsupervised domain adaptation (UDA) for 3D LiDAR semantic segmentation from synthetic to real domains, where existing methods suffer from suboptimal utilization of unlabeled real data due to fixed-confidence thresholds for pseudo-label filtering. We propose a self-training framework featuring three key innovations: (1) a dynamic pseudo-label filtering mechanism that adaptively adjusts confidence thresholds based on per-sample uncertainty estimates; (2) a prior-guided data augmentation strategy integrating geometric and semantic priors to enhance the reliability and utility of low-confidence predictions; and (3) a data-mixing consistency loss that jointly promotes context-agnostic feature learning and cross-domain alignment. Our method achieves state-of-the-art performance on two standard synthetic-to-real UDA benchmarks. Ablation studies confirm the effectiveness of each component. The source code is publicly available.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: 3D Computer Vision
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Annotating real-world LiDAR point clouds for use in intelligent autonomous systems is costly. To overcome this limitation, self-training-based Unsupervised Domain Adaptation (UDA) has been widely used to improve point cloud semantic segmentation by leveraging synthetic point cloud data. However, we argue that existing methods do not effectively utilize unlabeled data, as they either rely on predefined or fixed confidence thresholds, resulting in suboptimal performance. In this paper, we propose a Dynamic Pseudo-Label Filtering (DPLF) scheme to enhance real data utilization in point cloud UDA semantic segmentation. Additionally, we design a simple and efficient Prior-Guided Data Augmentation Pipeline (PG-DAP) to mitigate domain shift between synthetic and real-world point clouds. Finally, we utilize data mixing consistency loss to push the model to learn context-free representations. We implement and thoroughly evaluate our approach through extensive comparisons with state-of-the-art methods. Experiments on two challenging synthetic-to-real point cloud semantic segmentation tasks demonstrate that our approach achieves superior performance. Ablation studies confirm the effectiveness of the DPLF and PG-DAP modules. We release the code of our method in this paper.

Problem

• Improves unsupervised domain adaptation for 3D LiDAR semantic segmentation
• Enhances real data utilization via dynamic pseudo-label filtering
• Mitigates domain shift between synthetic and real point clouds

Innovation

• Dynamic Pseudo-Label Filtering enhances real data utilization
• Prior-Guided Data Augmentation Pipeline mitigates domain shift
• Data mixing consistency loss learns context-free representations