论文标题
sotverse：单个对象跟踪的用户定义的任务空间
SOTVerse: A User-defined Task Space of Single Object Tracking
论文作者
论文摘要
单个对象跟踪（SOT）研究属于一个周期 - 跟踪器在大多数基准测试中表现良好，但在具有挑战性的情况下很快失败了，导致研究人员怀疑数据内容不足，并付出更多的努力来构建具有更具挑战性情况的较大数据集。但是，效率低下的数据利用和有限的评估方法更严重地阻碍了SOT研究。前者无法全面利用现有数据集，而后者忽略了评估过程中具有挑战性的因素。在本文中，我们将代表性基准系统化，并形成一个单个对象跟踪元视频（SOTVERSE） - 用户定义的SOT任务空间，以突破瓶颈。我们首先提出一个3E范式来描述三个组件（即环境，评估和执行者）的任务。然后，我们总结了任务特征，澄清组织标准，并以1256万帧的形式构建Sotverse。具体而言，Sotverse会自动标记每个框架挑战因素，从而使用户可以通过施工规则有效地生成用户定义的空间。此外，Sotverse提供了两种具有新指标的机制，并成功评估了各种子任务下的跟踪器。因此，Sotverse首先提供了改善计算机视觉领域资源利用率的策略，从而使研究更具标准化和科学作用。 Sotverse，工具包，评估服务器和结果可在http://metaverse.aitestunion.com上找到。
Single object tracking (SOT) research falls into a cycle -- trackers perform well on most benchmarks but quickly fail in challenging scenarios, causing researchers to doubt the insufficient data content and take more effort to construct larger datasets with more challenging situations. However, inefficient data utilization and limited evaluation methods more seriously hinder SOT research. The former causes existing datasets can not be exploited comprehensively, while the latter neglects challenging factors in the evaluation process. In this article, we systematize the representative benchmarks and form a Single Object Tracking metaverse (SOTVerse) -- a user-defined SOT task space to break through the bottleneck. We first propose a 3E Paradigm to describe tasks by three components (i.e., environment, evaluation, and executor). Then, we summarize task characteristics, clarify the organization standards, and construct SOTVerse with 12.56 million frames. Specifically, SOTVerse automatically labels challenging factors per frame, allowing users to generate user-defined spaces efficiently via construction rules. Besides, SOTVerse provides two mechanisms with new indicators and successfully evaluates trackers under various subtasks. Consequently, SOTVerse first provides a strategy to improve resource utilization in the computer vision area, making research more standardized and scientific. The SOTVerse, toolkit, evaluation server, and results are available at http://metaverse.aitestunion.com.
