视频行人属性识别数据集

该视频行人属性数据库采用了 PETA 数据库建库过程中所采用的数据选取策略，选取了两个目前规模最大的视频行人重识别数据库 MARS 和 Duke-VID 作为建库所需的视频数据来源，在此基础之上以视频为单位进行行人属性手工标注，保证了数据库的扩展性。根据两个数据库行人和季节的不同特点，采用了不同的属性标注策略。This video-based pedestrian attribute database adopts the data selection strategy used in the database construction process of the PETA database. It selects the two currently largest-scale video-based person re-identification databases, MARS and Duke-VID, as the video data sources for its own construction. On this basis, manual annotation of pedestrian attributes is performed on a per-video basis, which ensures the scalability of this database. Different attribute annotation strategies are adopted based on the distinct characteristics of pedestrians and seasons across the two source databases.

搜集汇总

背景与挑战

背景概述

该数据集基于PETA数据库的数据选取策略，从MARS和Duke-VID两个视频行人重识别数据库中选取视频数据，并以视频为单位进行行人属性的手工标注。根据行人和季节特点，采用了不同的属性标注策略，确保了数据库的扩展性。