MOTCOM

如果浏览MOT文献，通常是遮挡，不稳定的运动和视觉相似性，它们要么直接被称为障碍物，要么通过跟踪器的设计间接被提及 (而不是对象的数量)。因此，我们提出了有史以来第一个新颖的MOT数据集复杂度度量，称为MOTCOM，它是受MOT关键问题启发的三个子度量的组合: 遮挡，不稳定运动和视觉相似性。MOTCOM的见解可以开启有关跟踪器性能的细微讨论，并可能导致对为鲜为人知的数据集或旨在解决子问题的数据集开发的新颖贡献的广泛认可。Upon reviewing the MOT literature, occlusion, unstable motion, and visual similarity are often either directly referred to as tracking obstacles or indirectly addressed through tracker design, rather than focusing solely on the quantity of objects. We hereby propose the first-ever novel MOT dataset complexity metric, termed MOTCOM, which combines three sub-metrics inspired by core MOT challenges: occlusion, unstable motion, and visual similarity. The insights derived from MOTCOM can facilitate nuanced discussions regarding tracker performance, and may foster broader recognition of novel contributions developed for understudied datasets or datasets intended to solve specific sub-problems.

搜集汇总

背景与挑战

背景概述

MOTCOM是一个创新的多目标跟踪数据集复杂度度量工具，由遮挡、不稳定运动和视觉相似性三个关键问题启发而来。该数据集由慕尼黑工业大学等机构于2022年发布，旨在通过量化跟踪障碍物来细化性能分析并推动相关研究。