EGO-CH(EGOcentric-Cultural Heritage)

我们提出了EGO-CH，这是一个以自我为中心的视频数据集，用于了解访问者的行为。该数据集已在两个不同的文化场所中收集，其中包括70名受试者 (包括志愿者和60名真实访问者) 获得的27个小时以上的视频。整个数据集包括26个环境和超过200个兴趣点 (poi) 的标签。具体来说，EGO-CH的每个视频都带有1) 时间标签，指定访问者的当前位置和观察到的POI，2) 围绕POI的边界框注释。由60个视频组成的数据集的很大一部分也与访问者在每次访问结束时填写的调查相关联。为了鼓励对该主题的研究，我们提出了4项具有挑战性的任务，这些任务有助于了解访问者的行为并报告数据集上的基线结果

We introduce EGO-CH, an egocentric video dataset developed for analyzing visitor behavior. This dataset was collected across two distinct cultural venues, containing over 27 hours of footage from 70 subjects, including volunteers and 60 real-world visitors. The complete dataset encompasses 26 environments and labels for more than 200 points of interest (POI). Specifically, every video in EGO-CH is paired with two annotation types: 1) temporal labels that indicate the visitor's current location and observed POI, and 2) bounding box annotations surrounding the target POI. A large subset of the dataset, comprising 60 videos, is also associated with surveys completed by visitors immediately after each visit. To promote research in this domain, we present four challenging tasks aimed at advancing the understanding of visitor behavior, and report baseline performance results evaluated on the EGO-CH dataset.

搜集汇总

背景与挑战

背景概述

EGO-CH是一个以自我为中心的视频数据集，用于研究文化场所中访客的行为，包含70名受试者采集的27小时以上视频，涵盖26个环境和200多个兴趣点，并提供了时间标签和边界框注释。该数据集还关联了调查问卷，并提出了4项挑战性任务以支持行为理解研究。