DoMSEV (Dataset of Multimodal Semantic Egocentric Video)

我们提出了一个 80 小时的多模式语义以自我为中心的视频 (DoMSEV) 数据集，涵盖了广泛的活动。这些视频是使用 GoPro Hero 相机或由连接到英特尔 Realsense R200 RGB-D 相机的 3D 惯性运动单元 (IMU) 组成的内置设置录制的。不同的人在不同的照明和天气条件下录制了视频。记录器标记视频，告知拍摄某些片段的场景（例如，室内、城市、拥挤的环境或自然）、进行的活动（步行、跑步、站立、浏览、驾驶、骑自行车、吃饭、烹饪、吃饭、观察，在谈话、玩耍或购物中），如果有什么东西引起了他们的注意，以及他们与某个物体互动时。此外，我们为每个记录者创建了一个配置文件，代表他们对一组对象和视觉概念的偏好。

We present an 80-hour multimodal semantic egocentric video dataset (DoMSEV) that covers a wide range of activities. These videos were recorded using either a GoPro Hero camera or a built-in setup consisting of a 3D inertial measurement unit (IMU) connected to an Intel RealSense R200 RGB-D camera. The videos were captured by different individuals under varying lighting and weather conditions. Recorders annotated the videos to indicate the scene (e.g., indoor, urban, crowded environments, or natural settings) for the recorded segments, the activities being performed (walking, running, standing, browsing, driving, cycling, eating, cooking, eating, observing, conversing, playing, or shopping), whether anything attracted their attention, and when they interacted with an object. Additionally, we created a profile for each recorder that represents their preferences over a set of objects and visual concepts.

搜集汇总

背景与挑战

背景概述

DoMSEV是一个80小时的多模态以自我为中心视频数据集，涵盖多样活动，使用GoPro或配备IMU的RGB-D相机在不同条件下录制，并包含场景、活动标注及记录者偏好信息。该数据集由米纳斯吉拉斯联邦大学于2018年发布。