MEMO-Bench
收藏资源简介:
MEMO-Bench是由上海交通大学、哈佛医学院和鹏城实验室联合创建的一个大型情感分析数据集,旨在评估文本到图像模型和多模态大语言模型在人类情感分析中的能力。该数据集包含7145张由12种文本到图像模型生成的肖像图像,每张图像代表六种基本情感之一。数据集的创建过程包括使用100个不同的提示词生成图像,并通过人工审查确保图像与情感表达的一致性。MEMO-Bench主要应用于情感分析领域,旨在解决AI在情感理解和表达方面的挑战。
MEMO-Bench is a large-scale sentiment analysis dataset jointly developed by Shanghai Jiao Tong University, Harvard Medical School, and Peng Cheng Laboratory. It aims to assess the capabilities of text-to-image models and multimodal large language models (LLMs) in human sentiment analysis. This dataset contains 7,145 portrait images generated by 12 distinct text-to-image models, with each image corresponding to one of the six basic emotions. The dataset construction process involves generating images using 100 different prompts, and ensuring the alignment between the images and their expressed emotions through manual review. MEMO-Bench is primarily applied in the field of sentiment analysis, with the objective of addressing the challenges encountered by AI in emotional understanding and expression.




