SMV-EAR：将时空多视图表征学习引入高效事件相机动作识别
SMV-EAR: Bring Spatiotemporal Multi-View Representation Learning into Efficient Event-Based Action Recognition
把事件流沿 H/W 轴投影成平移不变的稠密多视图表征，并用双分支动态融合建模样本级互补性，这套 SMVRL 设计可直接迁移到任何需要从稀疏异步事件中提取时空特征的视觉任务（如事件相机上的检测、跟踪、姿态估计）。
原文依据 a principled spatiotemporal multi-view representation through translation-invariant dense conversion of sparse events
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论事件相机动作识别（EAR）具有隐私保护与效率方面的显著优势，其中时序运动动态至关重要。现有的面向事件物体识别（EOR）的时空多视图表征学习（SMVRL）方法通过将 H-W-T 事件单独投影到空间轴 H 和 W 上提供了有前景的解决方案，但受限于其平移可变的空间分箱表征以及朴素的早期拼接融合架构。本文重新审视了面向 EAR 的 SMVRL 关键设计阶段，并提出：(i) 通过对稀疏事件进行平移不变的稠密转换，得到一种有原则的时空多视图表征；(ii) 一种双分支动态融合架构，用于建模来自不同视图的运动特征之间的样本级互补性；(iii) 一种受生物启发的时序扭曲增强，用以模拟真实世界人体动作的速度变化。在 HARDVS、DailyDVS-200 和 THU-EACT-50-CHL 三个具有挑战性的 EAR 数据集上，相较于现有 SMVRL EOR 方法，我们分别取得了 +7.0%、+10.7% 和 +10.2% 的 Top-1 准确率提升，同时参数量意外减少 30.1%、计算量降低 35.7%，从而确立了我们的框架作为一种新颖而强大的 EAR 范式。代码见 https://github.com/Fineshawray/SMV-EAR。
英文原文摘要 1284 字符
Event cameras action recognition (EAR) offers compelling privacy-protecting and efficiency advantages, where temporal motion dynamics is of great importance. Existing spatiotemporal multi-view representation learning (SMVRL) methods for event-based object recognition (EOR) offer promising solutions by projecting H-W-T events alone spatial axis H and W, yet are limited by its translation-variant spatial binning representation and naive early concatenation fusion architecture. This paper reexamines the key SMVRL design stages for EAR and propose: (i) a principled spatiotemporal multi-view representation through translation-invariant dense conversion of sparse events, (ii) a dual-branch, dynamic fusion architecture that models sample-wise complementarity between motion features from different views, and (iii) a bio-inspired temporal warping augmentation that mimics speed variability of real-world human actions. On three challenging EAR datasets of HARDVS, DailyDVS-200 and THU-EACT-50-CHL, we show +7.0%, +10.7%, and +10.2% Top-1 accuracy gains over existing SMVRL EOR method with surprising 30.1% reduced parameters and 35.7% lower computations, establishing our framework as a novel and powerful EAR paradigm. Code are avaliable in https://github.com/Fineshawray/SMV-EAR.