从书籍构建的图像-文本对数据集

本数据集由NAVER云公司等机构创建，专门从书籍中提取图像与文本对，旨在通过机器学习模型自主学习知识。数据集包含9516对图像与文本，主要来源于1868至1945年间出版的日本摄影书籍，涉及大量风景和历史建筑描述。创建过程涉及使用OCR、对象检测器和布局分析器自动提取图像与文本对，并利用书籍的元数据作为监督标签。该数据集的应用领域包括图像-文本检索和洞察提取，特别是在历史和文化研究领域，有助于机器学习模型理解和分析人类历史记录。

Created by NAVER Cloud and other institutions, this dataset is specifically designed to extract image-text pairs from books, with the goal of enabling machine learning models to autonomously acquire knowledge. It comprises 9,516 image-text pairs, primarily sourced from Japanese photographic books published between 1868 and 1945, covering extensive descriptions of landscapes and historical buildings. During its development, image-text pairs were automatically extracted via OCR, object detectors and layout analyzers, with book metadata utilized as supervised labels. The dataset finds applications in image-text retrieval and insight extraction, especially in the fields of historical and cultural research, where it aids machine learning models in understanding and analyzing human historical records.