HumanAct12

HumanAct12 是从极坐标图像和 3D 姿势数据集 PHSPD 中采用的新的 3D 人体运动数据集，具有适当的时间裁剪和动作注释。据统计，共有 1191 个 3D 运动剪辑（总共 90,099 个姿势），分为 12 个动作类和 34 个细粒度子类。动作类型包括走路、跑步、坐下、跳起来、热身等日常动作。细粒度的动作类型包含更具体的信息，如左鞠躬热身、左腿压腿热身等。

HumanAct12 is a novel 3D human motion dataset adapted from the polar image and 3D pose dataset PHSPD, with appropriate temporal cropping and action annotations. Statistically, it comprises a total of 1,191 3D motion clips (totaling 90,099 poses), which are categorized into 12 action classes and 34 fine-grained subcategories. The action types cover daily movements such as walking, running, sitting, jumping up, warming up, and so on. The fine-grained subcategories include more specific details, such as left-bowing warm-up, left-leg stretching warm-up, and the like.

搜集汇总

背景与挑战

背景概述

HumanAct12是一个3D人体运动数据集，包含1191个运动剪辑和90,099个姿势，分为12个动作类和34个细粒度子类，涵盖日常动作如走路、跑步等。该数据集源自PHSPD数据集，经过时间裁剪和动作注释处理，适用于人类动作生成研究，由耶鲁大学等机构于2020年发布。