Replication Data for DeepPatent2: A Large-Scale Benchmarking Corpus for Technical Drawing Understanding

DeepPatent2 consists of over 2.8 million design patents made public for the purpose of technical drawing understanding. It is made up of both original compound figures. and segmented figures obtained via application of deep-learning-based model on the original patents. It also include the metadata that about the patent id, type of object, patent figure release date, aspect view of object, etc. which are useful for NLP techniques.

DeepPatent2数据集包含超280万件面向技术图纸理解任务公开的外观设计专利。该数据集由原始复合图纸，以及基于深度学习模型处理原始专利得到的分割图纸两部分构成。此外，数据集还涵盖专利编号、目标对象类型、专利图纸发布日期、目标对象视角等元数据，此类元数据可用于自然语言处理（Natural Language Processing，NLP）相关技术的研究与应用。