saluslab/Your_actions_talk

该数据集名为Your actions talk，是一个多模态数据集，包含12种两人活动。这些活动源自Paul Ekman等人的非语言行为研究，旨在从身体动作中提取语义。数据集通过Microsoft Azure Kinect v2收集，包含14,400个样本，每个样本都有四种模态：RGB、深度、红外（IR）和3D骨架关节。数据集的组织结构分为四层：模态、位置组合、交互标签和主题、时间戳以及图像或CSV文件。此外，文件还详细说明了每种数据格式的规格，并提供了样本帧的示例。

This repository introduces a multi-modal dataset, **Your actions talk**, which contains 12 two-person—or dyadic—activities. These activities are adopted from **The Repertoire of Nonverbal Behavior: Categories, Origins, Usage, and Coding** by Paul Ekman et al, which allows us to distill the semantics embedded in bodily movements. The dataset is collected via Microsoft Azure Kinect v2 and constitutes of **14,400** samples, all of which come with 4 modalities: **RGB**, **depth**, **infrared (IR)**, and **3D skeleton joints**. The folder structure used to categorize our data comprises four layers: (1) modality, (2) location combination, interaction label, and subject, (3) timestamps, and (4) image or csv files. Each sample’s 3D skeleton joints position relative to the Azure Kinect in 3D space are written to a *.csv* file, where each row represents a frame and each column captures the x, y, and z joint coordinates, measured as the distance (in millimeters) to the camera.