DeepPatent2
收藏资源简介:
DeepPatent2是一个大规模的技术图纸理解基准数据集,由洛斯阿拉莫斯国家实验室创建。该数据集包含超过270万个技术图纸,这些图纸是从2007年至2020年的美国设计专利文档中提取的,每个图纸都配有132,890个对象名称和22,394个视点信息。创建过程中,使用了自然语言处理模型来自动提取对象名称和视点,以及计算机视觉方法来分割复合图纸。DeepPatent2的应用领域包括3D图像重建和图像检索,旨在解决从2D草图中理解技术信息的问题。
DeepPatent2 is a large-scale technical drawing understanding benchmark dataset developed by Los Alamos National Laboratory. This dataset comprises over 2.7 million technical drawings extracted from U.S. design patent documents spanning from 2007 to 2020. Each drawing is paired with 132,890 object names and 22,394 pieces of viewpoint information. During its creation, natural language processing (NLP) models were used to automatically extract object names and viewpoint information, while computer vision methods were employed to segment composite drawings. The application fields of DeepPatent2 include 3D image reconstruction and image retrieval, aiming to address the problem of understanding technical information from 2D sketches.

- 1DeepPatent2: A Large-Scale Benchmarking Corpus for Technical Drawing Understanding洛斯阿拉莫斯国家实验室 · 2023年



