Multi-Dimensional Insights (MDI) Benchmark
收藏资源简介:
Multi-Dimensional Insights (MDI) Benchmark 是一个用于评估大型多模态模型(LMMs)在真实世界场景中个性化能力的数据集。该数据集包含超过500张真实世界的图像和1298个人工提出的问题,涵盖了六个主要的生活场景。数据集通过复杂性和年龄两个维度进行分类,旨在评估模型在不同年龄段和问题复杂度下的表现。数据集的创建过程包括图像收集、问题生成和多轮验证,确保数据的多样性和平衡性。该数据集主要用于解决LMMs在实际应用中对不同用户需求的个性化响应问题。
Multi-Dimensional Insights (MDI) Benchmark is a dataset developed to evaluate the personalized capabilities of large multimodal models (LMMs) in real-world scenarios. It comprises over 500 real-world images and 1,298 manually authored questions, spanning six core daily life scenarios. The dataset is categorized along two dimensions: question complexity and age cohort, with the objective of assessing model performance across different age groups and varying levels of question complexity. The construction of this benchmark involves three key phases: image collection, question generation, and multi-round validation, to guarantee the diversity and balanced distribution of the collected data. This benchmark is primarily intended to address the challenge of enabling LMMs to deliver personalized responses tailored to diverse user requirements in practical real-world applications.




