MARS (Motion Analysis and Re-identification Set)

MARS (Motion Analysis and Re-identification Set) 是一个基于视频的大规模行人再识别数据集，是 Market-1501 数据集的扩展。它是从六个近乎同步的相机中收集的。它由 1,261 名不同的行人组成，这些行人至少被 2 台摄像机捕获。行人的姿势、颜色和光照的变化，以及较差的图像质量，使得很难产生高匹配精度。此外，该数据集包含 3,248 个干扰项，以使其更加真实。可变形零件模型和 GMMCP 跟踪器用于自动生成轨迹（大部分为 25-50 帧长）。

MARS (Motion Analysis and Re-identification Set) is a large-scale video-based person re-identification dataset, serving as an extension of the Market-1501 dataset. It is collected from six nearly synchronized cameras. The dataset comprises 1,261 distinct pedestrians, each captured by at least two cameras. Variations in pedestrian pose, color and illumination, as well as poor image quality, make it challenging to attain high matching accuracy. Furthermore, the dataset contains 3,248 distractors to enhance its realism. Deformable part models and GMMCP trackers are employed to automatically generate trajectories, most of which are 25–50 frames in length.

搜集汇总

背景与挑战

背景概述

MARS是一个基于视频的大规模行人再识别数据集，扩展自Market-1501，包含1,261名行人和3,248个干扰项，以模拟真实场景中的挑战。它通过六个相机采集，具有姿势、光照和图像质量变化，并采用自动轨迹生成方法，适用于视频人员再识别研究。