BABEL 动作分类数据集

BABEL 是一个大规模数据集，其中的语言标注描述了 mocap 序列中正在进行的动作。 BABEL 包括来自 AMASS 数据集的约 43 小时的 mocap 序列的动作标注。动作标注有两个抽象层次– 序列标注描述序列中的整体动作，而帧标注描述序列中每一帧的所有动作。每一帧标注都与 mocap 序列中相应动作的持续时间精确对齐，而且多个动作可以重叠。BABEL is a large-scale dataset where linguistic annotations describe the ongoing actions within motion capture (mocap) sequences. BABEL contains action annotations for approximately 43 hours of mocap sequences sourced from the AMASS dataset. Action annotations have two levels of abstraction: sequence-level annotations describe the overall action within a sequence, while frame-level annotations describe all actions present in each frame of the sequence. Each frame-level annotation is precisely aligned with the duration of the corresponding action in the mocap sequence, and multiple actions may overlap.

搜集汇总

背景与挑战

背景概述

BABEL是一个大规模动作分类数据集，包含约43小时的mocap序列，提供序列和帧两个抽象层次的标注，涉及250多个独特动作类别。该数据集适用于动作识别、时间性动作定位和运动合成等任务。