UDIVA

UDIVA数据集由来自22个国家 (68% 个来自西班牙) 的4至84岁 (平均值 = 31.29) 的147名自愿参与者 (55.1% 名男性) 之间的dyadic互动记录组成。大多数参与者是学生 (38.8%)，并将自己标识为白人 (84.4%)。参与者分为188个二元会议，平均参与2.5个会议/参与者 (最多5个会议)。最常见的互动群体是男性-男性/年轻-年轻/未知 (15%)，其中43% 互动发生在已知人群之间。西班牙语是大多数互动语言 (71.8%)，其次是加泰罗尼亚语 (19.7%) 和英语 (8.5%)。一半的会议包括以西班牙为原籍国的两位对话者。从技术上讲，数据是使用6个高清三脚架安装的摄像机 (1280 × 720px，25fps)，每个参与者1个翻领麦克风和桌子上的全向麦克风获取的。每个参与者还在脖子上佩戴以自我为中心的相机 (1920 × 1080px，30fps)，手腕上佩戴心率监测器。所有捕获设备都是时间同步的，并且安装在三脚架上的摄像机进行了校准。图1说明了UDIVA数据集的记录设置和不同视图。图2说明了UDIVA数据集的不同上下文 (即任务)。The UDIVA dataset consists of dyadic interaction recordings from 147 voluntary participants aged 4 to 84 years (mean = 31.29) across 22 countries, 68% of whom are from Spain, with 55.1% of the participants being male. Most participants were students (38.8%) and self-identified as White (84.4%). The participants were divided into 188 dyadic sessions, with an average of 2.5 sessions per participant (maximum of 5 sessions per participant). The most common interaction groups were male-male/young-young/unknown (15%), and 43% of all interactions occurred between known individuals. Spanish was the most frequently used interaction language (71.8%), followed by Catalan (19.7%) and English (8.5%). Half of the sessions involved two interlocutors originally from Spain. Technically, the data was captured using 6 high-definition tripod-mounted cameras (1280 × 720px, 25fps), one lapel microphone per participant, and an omnidirectional microphone placed on the table. Each participant also wore an egocentric camera around their neck (1920 × 1080px, 30fps) and a heart rate monitor on their wrist. All capture devices were time-synchronized, and the tripod-mounted cameras were calibrated. Figure 1 illustrates the recording setup and different views of the UDIVA dataset, while Figure 2 demonstrates its various contexts (i.e., tasks).

搜集汇总

背景与挑战

背景概述

UDIVA数据集记录了来自22个国家的147名志愿者（平均年龄31.29岁，55.1%为男性）之间的二元互动，共包含188次会议，主要使用西班牙语进行。数据通过同步的摄像机、麦克风和心率监测器等设备采集，由巴塞罗那大学等机构于2021年发布。