BDIQA

BDIQA是一个专为视频问答设计的新数据集，旨在通过心智理论探索认知推理。该数据集由上海多维信息处理重点实验室创建，包含3,527个视频和19,932个问答对，用于评估机器在简单和复杂场景中对信念、欲望和意图（BDI）的推理能力。BDIQA的创建过程受到儿童心智理论发展的启发，通过动画方式生成视频，并经过人工评估和筛选。该数据集的应用领域主要集中在提升视频问答模型在复杂场景中的理解能力，解决现有视频问答数据集在心智理论推理任务上的不足。

BDIQA is a novel dataset specifically designed for video question answering (Video QA) tasks, aiming to explore cognitive reasoning through Theory of Mind (ToM). Developed by the Shanghai Key Laboratory of Multidimensional Information Processing, this dataset comprises 3,527 videos and 19,932 question-answer pairs, which are used to evaluate machines' reasoning abilities regarding beliefs, desires, and intentions (BDI) across both simple and complex scenarios. The creation of BDIQA is inspired by the developmental trajectory of Theory of Mind in children; all videos are generated through animation and subjected to manual evaluation and screening. The primary application of this dataset focuses on enhancing the comprehension capabilities of video QA models in complex scenarios, addressing the gaps in existing Video QA datasets for Theory of Mind-based reasoning tasks.