standard-cognition/RetailAction

RetailAction数据集是一个大规模的数据集，旨在用于多视角时空定位零售环境中的客户与产品交互。该数据集包含了来自10个便利商店的21000个多视角注释样本，总共约41小时的视频。每个样本包含两个同步的顶部视角视频，记录了真实客户的行为。数据集以交互为中心的标签，而不是围绕人的边界框，而是使用点注释来标记物品被触摸的确切位置。数据集包括取、放、触摸等动作类别，并且每个视频都经过高效剪辑，以保证信息的最大化。所有数据都经过匿名处理，以保护隐私。

The RetailAction dataset is a large-scale dataset designed for multi-view spatio-temporal localization of customer-product interactions in real-world retail environments. The dataset consists of 21,000 multi-view annotated samples (approximately 41 hours of video) from 10 convenience stores. Each sample contains two synchronized top-view videos capturing real customer behaviors. The dataset uses interaction-centric labels, with point annotations marking the exact location where an item is touched. It includes action categories such as Take, Put, and Touch, and each video is剪辑ed efficiently to ensure maximum informativeness. All data are anonymized to protect privacy.