几何图形测试数据集
收藏资源简介:
几何图形识别数据集由中国科学院自动化研究所构建。我们的数据集由10000个图表样本组成,由16个形状组成,涵盖5种位置关系、22种符号类型和6种文本类型,在原始级别标记了更细粒度的注释,包括原始类、位置和关系,其中3626个非重复图像是从Geometric3K数据集中选择的,其他6374个图像是从数学课程网站上6-12年级的三本流行教科书中通过从PDF书籍中截取屏幕截图收集的。结合上述注释和几何先验知识,可以自动、唯一地生成可理解的几何命题。
The geometric shape recognition dataset was constructed by the Institute of Automation, Chinese Academy of Sciences. This dataset contains 10,000 graphic samples, covering 16 shape categories, 5 types of positional relationships, 22 symbol types and 6 text types. Fine-grained annotations including original category, position and corresponding relationship are marked at the raw image level. Among them, 3,626 non-duplicate images are selected from the Geometric3K dataset, while the remaining 6,374 images are collected by capturing screen shots from PDF versions of three popular mathematics textbooks for grades 6 to 12 on math course websites. Combined with the aforementioned annotations and geometric prior knowledge, understandable geometric propositions can be generated automatically and uniquely.




