hujiujiu12138/reevaluate-image-text-pairs

这是一个为知识增强多模态检索系统构建的图像-文本对数据集，基于REEVLAUATE KG ArtKB知识库。数据集包含训练集、验证集和测试集，每个样本包含图像、唯一标识符、查询文本和目标文本。数据来源包括Wikidata和Pilot Museums。文本描述由元数据部分和内容部分组成，元数据部分通过不同模板生成，内容部分使用Salesforce/BLIP2-OPT-2.7B模型自动生成。数据集可用于训练和评估CLIP模型。

This is an image-text pair dataset constructed for the Knowledge-Enhanced Multimodal Retrieval System, built upon the REEVLAUATE KG ArtKB. The dataset is designed for training and evaluating the CLIP model for the retrieval system. It contains train, validation, and test splits, with each sample including an image, unique identifier, query text, and target text. Data sources include Wikidata and Pilot Museums. Text descriptions consist of metadata portions generated using templates and content portions generated automatically using the Salesforce/BLIP2-OPT-2.7B model.