MEVID

我们展示了具有身份的多视图扩展视频 (MEVID) 数据集，用于在野外进行大规模的视频人员重新识别 (ReID)。据我们所知，MEVID代表了最多样化的视频人ReID数据集，涵盖了广泛的室内和室外环境，跨越了73天窗口中的9个独特日期，各种摄像机视点和实体服装变化。具体来说，我们标记了158独特的人穿着598服装的身份，这些服装取自8,092个轨道，平均长度约为590帧，从超大规模MEVA人活动数据集的33个摄像机视图中可以看到

We present the Multiview Extended Video with Identity (MEVID) dataset for large-scale video-based person re-identification (ReID) in the wild. To the best of our knowledge, MEVID is the most diverse video person ReID dataset, covering a wide range of indoor and outdoor environments, spanning 9 distinct dates within a 73-day window, and encompassing diverse camera viewpoints and physical clothing variations. Specifically, we annotated identities for 158 unique persons wearing 598 outfits, derived from 8,092 tracks with an average duration of approximately 590 frames, captured from 33 camera views of the ultra-large-scale MEVA human activity dataset.

搜集汇总

背景与挑战

背景概述

MEVID是一个大规模视频人员重新识别数据集，包含158个独特人物的598套服装，覆盖多样化的室内外环境和摄像机视点。数据集由Kitware, Inc.于2022年发布，适用于视频人ReID研究。