FaceXBench
收藏资源简介:
FaceXBench是由约翰霍普金斯大学的研究团队创建的一个综合性基准,旨在评估多模态大语言模型(MLLMs)在复杂人脸理解任务中的表现。该数据集包含5000个多模态选择题,涵盖了6大类14个任务,数据来源于25个公共数据集和新创建的FaceXAPI数据集。FaceXBench包含10441张独特的人脸图像,涵盖了不同年龄、性别、种族、分辨率和表情的多样性。数据集的创建过程包括从现有数据集中提取测试集、手动创建问题模板以及生成答案选项。FaceXBench的应用领域包括人脸识别、人脸认证、人脸分析等,旨在解决MLLMs在人脸理解任务中的不足,推动该领域的研究进展。
FaceXBench is a comprehensive benchmark created by a research team at Johns Hopkins University, designed to evaluate the performance of multimodal large language models (MLLMs) on complex facial understanding tasks. This dataset includes 5,000 multimodal multiple-choice questions covering 14 tasks across 6 categories, with data sourced from 25 public datasets and the newly created FaceXAPI dataset. FaceXBench contains 10,441 unique facial images, encompassing diversity across age, gender, ethnicity, resolution, and facial expressions. The dataset creation process includes extracting test subsets from existing datasets, manually crafting question templates, and generating answer options. Application scenarios of FaceXBench include facial recognition, face authentication, facial analysis, and others, aiming to address the limitations of MLLMs in facial understanding tasks and advance research progress in this field.

- 1FaceXBench: Evaluating Multimodal LLMs on Face Understanding约翰霍普金斯大学 · 2025年



