[论文解读] Towards Cross-Platform Generalization: Domain Adaptive 3D Detection with Augmentation and Pseudo-Labeling

Xiyan Feng, Wenbo Zhang+4 more|arXiv (Cornell University)|Jan 13, 2026

Advanced Neural Network Applications被引用 0

一句话总结

这篇论文在基于 PVRCNN++ 的域自适应三维对象检测方法中，利用有针对性的增强和伪标签自训练来提升跨平台泛化，在 RoboSense2025 基准测试中获得领奖台表现。

ABSTRACT

This technical report represents the award-winning solution to the Cross-platform 3D Object Detection task in the RoboSense2025 Challenge. Our approach is built upon PVRCNN++, an efficient 3D object detection framework that effectively integrates point-based and voxel-based features. On top of this foundation, we improve cross-platform generalization by narrowing domain gaps through tailored data augmentation and a self-training strategy with pseudo-labels. These enhancements enabled our approach to secure the 3rd place in the challenge, achieving a 3D AP of 62.67% for the Car category on the phase-1 target domain, and 58.76% and 49.81% for Car and Pedestrian categories respectively on the phase-2 target domain.

研究动机与目标

• 在 RoboSense 基准测试中 motivating cross-platform generalization for 3D detection across different domain targets.
• Leverage a strong baseline (PVRCNN++) to fuse point-based and voxel-based features for efficiency.
• Narrow domain gaps via tailored data augmentation and a self-training strategy with pseudo-labels.
• Demonstrate competitive performance on phase-1 and phase-2 target domains in RoboSense2025.

提出的方法

• Base framework: PVRCNN++ integrating point-based and voxel-based features.
• Apply targeted domain gap narrowing through specialized data augmentation techniques.
• Employ a self-training strategy using pseudo-labels to further align target-domain distributions.
• Evaluate on RoboSense2025 phase-1 and phase-2 targets for Car and Pedestrian detection.

实验结果

研究问题

• RQ1 How can cross-platform (cross-domain) generalization be improved for 3D object detection?
• RQ2 What role do augmentation and pseudo-labeling play in reducing domain gaps in 3D detection?
• RQ3 How does the proposed approach perform on RoboSense2025 phase-1 and phase-2 target domains for Car and Pedestrian categories?
• RQ4 What is the impact of the proposed method on 3D AP metrics compared to baseline?

主要发现

• Achieved 3D AP of 62.67% for Car on the phase-1 target domain.
• Achieved 3D AP of 58.76% for Car and 49.81% for Pedestrian on the phase-2 target domain.