---
title: "PT$^3$: Prototype-Guided Self-Supervised Pretraining for Generalizable mmWave Human Point Cloud Representation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.38551v1"
permalink: "https://feed.astralor.com/items/itm_9b667fb7e2dacb58fde6ae09"
published: "2026-10-28T09:49:12.000Z"
score: 53
channel: "tech"
tags: ["Prototype-Guided Self-Supervised Pretraining","Generalizable","Human Point Cloud","Representation Multi-source","However","Multi-source","Observation"]
---

# PT$^3$: Prototype-Guided Self-Supervised Pretraining for Generalizable mmWave Human Point Cloud Representation

## 速览

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, ch…

## 关键点

- 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.

## 全文

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.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_9b667fb7e2dacb58fde6ae09