GMOT-40
收藏资源简介:
GMOT-40是首个公开的密集通用多目标跟踪数据集,由天普大学和石溪大学联合创建。该数据集包含40个精心标注的视频序列,涵盖10个不同类别的对象,每个类别有4个序列。每个序列平均包含约22个同类别对象,所有序列均经过人工标注和验证。GMOT-40旨在解决现有MOT数据集对特定对象类别的依赖问题,适用于视频编辑、动物行为分析和基于视觉的对象计数等领域。数据集创建过程中,从YouTube等来源筛选并编辑视频,确保每个序列至少包含100帧,且80%的帧中至少有10个目标。
GMOT-40 is the first publicly available dense general multi-object tracking (MOT) dataset, jointly developed by Temple University and Stony Brook University. This dataset includes 40 meticulously annotated video sequences spanning 10 distinct object categories, with 4 sequences per category. Each sequence contains an average of approximately 22 objects of the same category, and all sequences have been manually annotated and verified. GMOT-40 aims to address the dependency issue of existing MOT datasets on specific object categories, and is applicable to fields such as video editing, animal behavior analysis, and vision-based object counting. During the dataset creation process, videos were screened and edited from sources including YouTube to ensure that each sequence contains at least 100 frames, and at least 10 targets appear in 80% of the frames.

- 1GMOT-40: A Benchmark for Generic Multiple Object Tracking天普大学 · 2021年



