EEmo-Bench
收藏资源简介:
EEmo-Bench是由上海交通大学研究团队创建的一个专门用于评估多模态大型语言模型在图像引发的情绪分析能力的基准。该数据集包含1960张涵盖广泛内容类别的图像,经人工标注,具有情绪排名、效价、唤醒度和支配度等情感属性。它旨在通过四项任务——感知、排名、描述和评估,对多模态大型语言模型进行全面的评估,以提升模型在图像引发的情绪感知和理解能力。
EEmo-Bench is a benchmark developed by the research team at Shanghai Jiao Tong University, specifically designed to evaluate the emotion analysis capabilities of multimodal large language models triggered by images. This dataset contains 1,960 images covering a wide range of content categories, which have been manually annotated with emotional attributes including emotion ranking, valence, arousal, and dominance. It aims to comprehensively evaluate multimodal large language models through four tasks: perception, ranking, description and assessment, so as to enhance the models' abilities in image-triggered emotion perception and understanding.




