Face-Human-Bench
收藏资源简介:
Face-Human-Bench是由北京邮电大学和北京师范大学的研究团队创建的一个综合性基准数据集,旨在评估多模态助手在面部和人类理解方面的能力。该数据集包含2700条问题,分为开发集(900条)和测试集(1800条),支持中英文评估。数据来源于公开的面部和人类理解数据集,通过半自动数据管道生成多模态问答问题。数据集的应用领域包括多模态助手的面部和人类理解能力评估,旨在提升多模态助手在社交互动、媒体分析和人机交互等领域的响应质量和应用范围。
Face-Human-Bench is a comprehensive benchmark dataset created by research teams from Beijing University of Posts and Telecommunications and Beijing Normal University, aiming to evaluate the facial and human understanding capabilities of multimodal assistants. This dataset contains 2700 questions, divided into a development set (900 questions) and a test set (1800 questions), and supports evaluations in both Chinese and English. The data is sourced from public facial and human understanding datasets, and multimodal question-answering pairs are generated through a semi-automatic data pipeline. Its application scenarios include the evaluation of facial and human understanding capabilities of multimodal assistants, with the goal of improving the response quality and application scope of multimodal assistants in fields such as social interaction, media analysis, and human-computer interaction.

- 1Face-Human-Bench: A Comprehensive Benchmark of Face and Human Understanding for Multi-modal Assistants北京邮电大学, 北京师范大学 · 2025年



