使用骨骼云彩色化进行自监督的 3D 动作表示学习（英文版）

英文标题：Self-Supervised 3D Action Representation Learning with Skeleton Cloud Colorization中文摘要：本文提出了一种自监督的方法，设计了一种新颖的骨骼云着色技术，并通过骨骼序列数据的无标签学习学习骨骼的空间和时态表示。该方法在不同配置下训练的线性分类器上评估，并在多个数据集上进行了广泛的实验。结果表明，该方法在无监督和半监督 3D 动作识别方法中取得了很大的提升，并且在监督 3D 动作识别方面也取得了竞争性能。英文摘要：3D Skeleton-based human action recognition has attracted increasing attentionin recent years. Most of the existing work focuses on supervised learning whichrequires a large number of labeled action sequences that are often expensiveand time-consu