基于自监督的骨架行为识别的跨流对比学习（英文版）

英文标题：Cross-Stream Contrastive Learning for Self-Supervised Skeleton-Based Action Recognition中文摘要：本文提出了一种基于骨架的自监督动作识别框架，采用跨流对比学习来设计硬样本对，结合积极变换策略，提高了对复杂动作模式的处理能力，并表现出优于三个基准数据集的效果。英文摘要：Self-supervised skeleton-based action recognition enjoys a rapid growth alongwith the development of contrastive learning. The existing methods rely onimposing invariance to augmentations of 3D skeleton within a single datastream, which merely leverages the easy positive pairs and limits the abilityto explore the co