Watch-n-Patch 动作识别数据集

Watch-n-Patch 数据集的重点是对人类活动进行建模，涵盖在完全无监督的情况下的多种动作。数据由微软 Kinect One 传感器收集，总时长约为 230 分钟，包含 458 段视频。该数据集要求 7 名受试者在 8 个办公室和 5 个具有复杂背景的厨房中进行日常活动。此外，该数据集还提供了骨架数据作为 ground truth 标注。

The Watch-n-Patch dataset focuses on modeling human activities, covering a diverse range of actions under fully unsupervised settings. The data was collected using Microsoft Kinect One sensors, with a total duration of approximately 230 minutes and comprising 458 video segments. This dataset involves 7 subjects performing daily activities across 8 offices and 5 kitchens with complex backgrounds. Additionally, skeleton data is provided as ground truth annotations.

搜集汇总

背景与挑战

背景概述

Watch-n-Patch 动作识别数据集专注于人类活动建模，在无监督条件下涵盖多种动作。该数据集通过微软 Kinect One 传感器采集，总时长约230分钟，包含458段视频，记录了7名受试者在办公室和厨房环境中的日常活动，并提供骨架数据作为标注。