BQA
收藏资源简介:
BQA数据集由奈良先端科学技术大学院大学创建,旨在评估视频大语言模型(VideoLLMs)对人类身体语言情感的理解能力。该数据集包含7632个5-10秒的短视频,每个视频带有26种情感标签和元数据(性别、年龄、种族)。数据集的创建过程包括提取候选答案、生成问题、过滤不适当的问题以及分配难度标签。BQA数据集主要用于评估模型在理解人类情感表达方面的能力,特别是在对话系统等应用中,旨在解决模型在情感理解和交互中的准确性问题。
The BQA dataset was developed by the Nara Institute of Science and Technology (NAIST) to evaluate the capacity of Video Large Language Models (VideoLLMs) to understand human emotional expressions conveyed through bodily language. This dataset contains 7,632 short videos with a duration of 5 to 10 seconds, each paired with 26 emotional labels and metadata including gender, age, and ethnicity. The dataset construction process involves extracting candidate answers, generating questions, filtering out inappropriate questions, and assigning difficulty labels. The BQA dataset is primarily used to assess a model's ability to comprehend human emotional expressions, particularly in applications such as dialogue systems, with the objective of addressing the accuracy issues faced by models in emotional understanding and interactive scenarios.



