PT³：面向可泛化毫米波人体点云表示的原型引导自监督预训练
PT$^3$: Prototype-Guided Self-Supervised Pretraining for Generalizable mmWave Human Point Cloud Representation
AI 导读
研究者提出 PT³，一个针对毫米波人体点云的原型引导自监督预训练框架，用共享原型的稀疏组合表示局部状态，并通过自回归预测建模时间演化，观测一致性可稳定雷达扰动下的表示。在多源未标注序列上预训练后，同一冻结编码器可支持动作分类、稠密点生成、关节姿态估计和未来运动预测。PT³ 在这些任务上优于对比预训练基线，并达到与任务专用模型相当或更好的性能。
正文 · 原文
Abstract:Multi-source millimeter-wave (mmWave) observations offer an opportunity to learn human representations reusable across sensing tasks. However, sparse detections and multipath effects make these observations unstable, challenging existing point cloud pretraining methods. We propose PT$^3$, a prototype-guided self-supervised pretraining framework for mmWave human point clouds. PT$^3$ represents local states in spatial regions as sparse compositions of shared prototypes and learns their temporal evolution through autoregressive prediction. Observation consistency further stabilizes representations under radar perturbations. After pretraining on unlabeled sequences from multiple sources, the same frozen encoder supports action classification, dense point generation, joint pose estimation, and future motion prediction on datasets not used for pretraining. Across these tasks, PT$^3$ outperforms the compared pretraining baselines and achieves competitive or superior performance to task-specific models.
| Comments: | 30 pages |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.38551 [cs.AI] |
| (or arXiv:2610.38551v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.38551 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ziyang Liu [view email]
[v1]
Wed, 28 Oct 2026 09:49:12 UTC (1,270 KB)
来源：arXiv cs.AI · arxiv.org