TimeLogic QA (TLQA)
收藏资源简介:
TimeLogic QA (TLQA) 数据集由中佛罗里达大学和图宾根大学的研究团队创建,旨在评估视频问答(VideoQA)模型对时间逻辑的理解能力。该数据集基于现有的视频数据集(如STAR、Breakfast、AGQA和CrossTask),通过自动生成问题-答案对,涵盖了16个时间逻辑类别,复杂度从1到5不等。TLQA包含128k个问题-答案对,分为布尔问题和多项选择题,适用于大规模评估。数据集的创建过程涉及定义时间逻辑类别、生成模板问题、构建实例状态以及自动生成正负样本。TLQA数据集的应用领域主要集中在视频分析,旨在解决复杂时间逻辑推理问题,提升模型对视频中事件顺序和因果关系的理解能力。
The TimeLogic QA (TLQA) dataset was created by a research team from the University of Central Florida and the University of Tübingen, aiming to evaluate the temporal logic understanding capability of video question answering (VideoQA) models. Built upon existing video datasets such as STAR, Breakfast, AGQA, and CrossTask, this dataset automatically generates question-answer pairs, covering 16 temporal logic categories with complexity levels ranging from 1 to 5. TLQA contains 128k question-answer pairs, which are categorized into boolean questions and multiple-choice questions, making it suitable for large-scale evaluation. The dataset creation process involves defining temporal logic categories, generating template questions, constructing instance states, and automatically generating positive and negative samples. The application scenarios of the TLQA dataset mainly focus on video analysis, aiming to solve complex temporal logic reasoning problems and improve the model's understanding of event sequences and causal relationships in videos.

- 1TimeLogic: A Temporal Logic Benchmark for Video QA中佛罗里达大学, 图宾根大学 · 2025年



