Large Labelled Logo Dataset (L3D)
收藏资源简介:
在这项工作中,我们介绍了大型标签徽标数据集 (L3D),这是一个多用途的,手工标记的,持续增长的数据集。它由从欧盟知识产权局 (EUIPO) 开放注册表中提取的0.7M 256x256 RGB图像组成。它们中的每一个都与多个标签相关联,这些标签对图像中出现的图形和文本元素进行了分类,这些元素与它们相关联。EUIPO评估人员使用Vienna分类 (具象标记的分层分类) 对这些注释进行了分类。我们建议对该数据集进行两种直接应用,即徽标分类和徽标生成。
In this work, we introduce the Large Labeled Logo Dataset (L3D), a versatile, manually annotated, and continuously growing dataset. It consists of 0.7M 256x256 RGB images extracted from the open registry of the European Union Intellectual Property Office (EUIPO). Each image is associated with multiple labels that classify the graphical and textual elements present in the corresponding image. EUIPO examiners used the Vienna Classification, a hierarchical classification system for figurative marks, to categorize these annotations. We propose two direct applications for this dataset: logo classification and logo generation.




