Richly Annotated Pedestrian (RAP) dataset
收藏资源简介:
RAP数据集是由中国科学院自动化研究所创建的大型行人属性识别数据集,包含41,585个行人样本,每个样本标注了72个属性以及视角、遮挡和身体部位信息。该数据集来源于真实的监控场景,旨在解决监控系统中行人属性识别的问题,特别是在视角、遮挡和身体部位变化下的属性识别。数据集的创建过程涉及从监控视频中提取样本,并使用高斯混合模型进行行人检测和跟踪。RAP数据集的应用领域包括监控系统中的行人分析、属性识别和多标签学习,旨在提高大规模属性识别系统的性能。
The RAP dataset is a large-scale pedestrian attribute recognition dataset developed by the Institute of Automation, Chinese Academy of Sciences. It contains 41,585 pedestrian samples, where each sample is annotated with 72 attributes, along with information about viewpoint, occlusion, and body parts. Derived from real-world surveillance scenarios, this dataset is designed to tackle the challenges of pedestrian attribute recognition in surveillance systems, particularly those involving variations in viewpoint, occlusion, and body part conditions. The construction of the RAP dataset entails extracting samples from surveillance videos and employing Gaussian Mixture Models (GMM) for pedestrian detection and tracking. Application areas of the RAP dataset include pedestrian analysis, attribute recognition, and multi-label learning in surveillance systems, with the objective of enhancing the performance of large-scale attribute recognition systems.



