Out the Window

Out the Window (OTW) 数据集是一个众包活动数据集，包含来自 NIST 扩展视频活动 (ActEV) 挑战的 17 个活动的 5,668 个实例。这些视频是从 Amazon Mechanical Turk 上的工作人员使用一种新颖的场景代理策略众包而来的，该策略收集每个场景的多个自然活动实例。土耳其人被指示将他们的移动设备靠在可以俯瞰室外空间的上层窗户上，走到外面执行涉及人、车辆和物体的场景，最后将视频上传给我们进行注释。 VIRAT Ground 2.0 上活动分类的性能评估表明，OTW 数据集的平均分类准确度提高了 8.3%，在涉及车辆的最具挑战性的活动上提高了 12.5%。Out the Window (OTW) dataset is a crowdsourced activity dataset containing 5,668 instances of 17 activities from the NIST Extended Video Activity (ActEV) Challenge. These videos were crowdsourced from Amazon Mechanical Turk workers via a novel scenario agent strategy, which collects multiple natural activity instances for each scenario. Workers were instructed to prop their mobile devices against upper-story windows overlooking outdoor spaces, go outdoors to enact scenarios involving people, vehicles, and objects, and finally upload the recorded videos to us for annotation. Performance evaluations of activity classification on VIRAT Ground 2.0 demonstrate that the OTW dataset improves average classification accuracy by 8.3%, and enhances performance on the most challenging vehicle-related activities by 12.5%.

搜集汇总

背景与挑战

背景概述

Out the Window 数据集是一个众包活动视频数据集，包含来自NIST扩展视频活动挑战的17个活动的5,668个实例，通过移动设备在窗外场景收集。该数据集在VIRAT Ground 2.0上的评估表明，它能将活动分类平均准确度提高8.3%，在涉及车辆的活动上提升12.5%。