论文

DINOv3+PiSSA 首次用于雷达微多普勒人体活动识别,小模型参数少 4.5 倍

Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR

精选理由

把图像领域自监督模型 DINOv3 直接迁移到雷达信号识别,居然比全量微调更抗分布偏移,参数还少 4.5 倍。做毫米波雷达、可穿戴感知或边缘设备部署的工程师可以试试这个 PiSSA 配方,代码已开源。

一篇 arXiv 论文(arXiv:2611.25874)首次把自监督图像基础模型 DINOv3 引入雷达微多普勒人体活动识别(HAR),并搭配 PiSSA(主奇异方向)这一参数高效微调方法。作者在受试者不相交的分布内交叉验证,以及跨年龄、跨朝向、跨数据集的少样本迁移设定下,比较了全量微调、冻结探测和 PEFT 三类方案:全量微调出现跨数据集负迁移,冻结特征在跨朝向场景欠适配,只有 PEFT 两类失败都避开。

正文

DINOv3-PiSSA 在所有评估设定中排名第一。其 Small 版本用仅 3% 可训练参数、总参数量减少 4.5 倍,就追平了此前 Base 规模 PEFT 方法的分布内准确率,并在每个设定上超过同等参数量的该基线,迁移增益随分布偏移程度递增,分别为 2.1、6.6 和 8.9 个百分点。消融实验显示预训练与适配器配方均有贡献,在复合偏移下预训练贡献最大。代码和协议已在 GitHub 开源(udoppler-ssl)。

对做非接触式感知、隐私保护 HAR 或边缘部署的团队来说,这提供了一条低参数开销、抗分布偏移的可复现路线。

原文 · arXiv cs.LG

Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR

Micro-Doppler radar enables contactless, privacy-preserving human activity recognition (HAR), but deployment robustness remains underexplored. We compared full fine-tuning, frozen probing, and parameter-efficient fine-tuning (PEFT) using subject-disjoint in-distribution cross-validation and few-shot cross-age, cross-orientation, and cross-dataset transfer. Full fine-tuning suffered negative cross-dataset transfer, while frozen features under-adapted on cross-orientation. Only PEFT avoided both failures. We paired DINOv3 with all-layer principal-singular-direction adaptation (PiSSA), to our knowledge the first application of a self-supervised image foundation backbone to radar micro-Doppler HAR. DINOv3-PiSSA ranked first among PEFT methods in every evaluated setting. Under our protocol, its Small variant matched the in-distribution accuracy of the prior Base-size PEFT state of the art using 4.5x fewer total parameters, with only 3% trainable. It surpassed a size-matched variant of that baseline in every evaluated setting, with transfer gains increasing with shift severity: 2.1, 6.6, and 8.9 percentage points, respectively. Ablations attributed improvements to both pretraining and adapter recipe, with the pretraining contributing most under compound shift. Code and protocol are available at https://github.com/KeryanChelouche/udoppler-ssl.

产业坐标与信源核验

本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:DINOv3、PiSSA、人体活动识别、参数高效微调,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始线索采编自「arXiv cs.LG」(@Keryan Chelouche, Petr Dobias, Nistor Grozavu, Olivier Romain),经 AITOP 多维评分管线结构化提炼。