mmWaveFlow：毫米波人体点云的统一增强与生成
mmWaveFlow: Unified Enhancement and Generation of mmWave Human Point Clouds
提出可逆的流匹配框架，在稠密与稀疏毫米波点云之间建立双向传输，并用跨模态潜变量对齐与原点感知流匹配缓解潜路径交叉，为稀疏点云增强与生成提供可复用范式。
原文依据 we propose a unified flow-matching framework mmWaveFlow that unifies enhancement and generation of mmWave point clouds
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论毫米波点云因其鲁棒性、隐私保护性和低成本而在人体感知中受到越来越多的关注。然而，其固有的数据稀疏性和缺乏大规模标注数据集阻碍了实际应用。我们重新审视生成建模，提出一个统一的流匹配框架 mmWaveFlow，通过学习稠密与稀疏点云之间的可逆传输来统一毫米波点云的增强与生成。我们利用配对数据和跨模态潜变量对齐模块来强制语义对齐并弥合模态差距。我们发现无条件的流匹配更容易受到潜路径交叉的影响，从而损害传输。因此，我们提出原点感知流匹配（OA-Flow），通过将传输条件建立在路径原点上，以减轻双向传输中的歧义。在多个数据集上的实验结果表明，mmWaveFlow 在毫米波人体点云生成和增强方面是有效的。我们还观察到下游任务中的一致增益，突显了我们的框架在人体感知中的前景。代码可在 https://github.com/suchang-99/mmWaveFlow 获取。
英文原文摘要 1222 字符
Millimeter-wave (mmWave) point clouds have attracted increasing interest in human sensing due to their robustness, privacy preservation, and low cost. However, their practical adoption is hindered by the inherent sparsity of data and the lack of large-scale annotated dataset. We revisit generative modeling and propose a unified flow-matching framework mmWaveFlow that unifies enhancement and generation of mmWave point clouds by learning an invertible transport between dense and sparse point clouds. We leverage paired data and a Cross-modal Latent Alignment module to enforce semantic alignment and bridge the modality gap. We find that condition-free flow matching is more vulnerable to latent path crossings, which impair transport. Therefore, we propose Origin-Aware Flow Matching (OA-Flow) by conditioning transport on the path origin to mitigate ambiguity in bidirectional transport. Results of experiments across multiple datasets demonstrate the effectiveness of mmWaveFlow for mmWave human point clouds generation and enhancement. We also observe consistent gains in downstream tasks, highlighting the promise of our framework for human sensing. Codes are available at https://github.com/suchang-99/mmWaveFlow.