PatFigVQA, PatFigCLS
收藏资源简介:
PatFigVQA和PatFigCLS是由德国汉诺威莱布尼茨科学与技术信息中心创建的两个数据集,旨在通过视觉问答(VQA)和分类任务来提升专利图的分类效果。PatFigCLS数据集包含35,926个专利图,分为10种类型,数据来源于扩展的CLEF-IP 2011数据集和DeepPatent2数据集。数据集的创建过程包括对专利图的类型、投影、对象和USPC类别的分类,并通过规则匹配和聚类方法对数据进行标准化处理。该数据集主要用于专利检索系统中的多面搜索,帮助专利审查员更高效地找到相关专利。
PatFigVQA and PatFigCLS are two datasets developed by the Leibniz Information Centre for Science and Technology (TIB) Hannover, Germany. These datasets are designed to improve the classification performance of patent figures through visual question answering (VQA) and classification tasks. The PatFigCLS dataset consists of 35,926 patent figures categorized into 10 types, with data sourced from the extended CLEF-IP 2011 dataset and the DeepPatent2 dataset. The creation process for these datasets includes classifying patent figures by their type, projection, objects, and USPC categories, as well as standardizing the data via rule-based matching and clustering methods. This dataset is primarily used for faceted search in patent retrieval systems, helping patent examiners find relevant patents more efficiently.
数据集概述
数据集名称
Patent Figure Classification using Large Vision-language Models
数据集来源
该数据集是论文《Patent Figure Classification using Large Vision-language Models》的官方GitHub页面,作者为Sushil Awale, Eric Müller-Budack, Ralph Ewerth。
数据集相关论文
- 论文标题: Patent Figure Classification using Large Vision-language Models
- 会议: European Conference on Information Retrieval (ECIR)
- 会议地点: Lucca, Italy
- 会议时间: 2025年
数据集状态
- 当前状态: 该仓库即将更新,详细信息将在未来发布。

- 1Patent Figure Classification using Large Vision-language Models德国汉诺威莱布尼茨科学与技术信息中心, 德国汉诺威莱布尼茨大学L3S研究中心 · 2025年



