OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking Supplementary Material School of Software Technology, Zhejiang University

Motivation For what purpose was the dataset created? Was there a specific task in mind? Was there a specific gap that needed to be filled? In the current task of open-vocabulary multi-object tracking (OVMOT), there is only one benchmark available, which lacks high-quality, large-scale datasets. The existing dataset suffers from several limitations, including insufficient categories, limited video data, and a significant imbalance between base classes and novel classes. These deficiencies make it inadequate for supporting the evaluation of new OVMOT models. Our proposed dataset aims to provide a more comprehensive evaluation platform for the OVMOT task. Who created this dataset (e.g., which team, research group) and on behalf of which entity (e.g., company, institution, organization)? This dataset was constructed by collecting and extracting data from seven other datasets and applying unified annotations. This work was completed by Haiji Liang and Ruize Han. Who funded the creation of the dataset?