PoseTrack21

我们介绍了positrack21，这是一个大规模的数据集，用于在具有高度多样性的姿势的现实场景中进行人员搜索，多对象跟踪和多人姿势跟踪。该数据集提供了丰富的注释，例如人类姿势注释，包括关节闭塞的注释，甚至对于小人物的边界框注释以及视频序列内和跨视频序列的人id。数据集允许与人的重新识别一起评估多对象跟踪和多人姿势跟踪，或者利用人类姿势的结构知识来改善人的搜索和跟踪，特别是在严重闭塞的情况下。借助positrack21，我们希望鼓励研究人员研究在所有三个任务上都表现良好的联合方法。

We introduce positrack21, a large-scale dataset for person search, multi-object tracking and multi-person pose tracking in real-world scenes with highly diverse poses. This dataset provides rich annotations, including human pose annotations with occlusion-aware joint annotations, bounding box annotations even for small-sized persons, and person IDs within and across video sequences. The dataset enables evaluation of multi-object tracking and multi-person pose tracking together with person re-identification, or leveraging structural knowledge of human poses to improve person search and tracking, especially under severe occlusion conditions. With positrack21, we aim to encourage researchers to explore joint methods that perform well across all three tasks.

搜集汇总

背景与挑战

背景概述

PoseTrack21是一个2022年发布的大规模数据集，专注于人员搜索、多对象跟踪和多人姿势跟踪任务，适用于现实场景中的高度多样性姿势分析。它提供了丰富的注释，包括人类姿势、关节闭塞、边界框和人ID，旨在支持联合方法研究，以改善在严重遮挡情况下的跟踪和搜索性能。