Hangul Fonts Dataset (HFD)

Hangul Fonts Dataset (HFD) 是一个专为研究深度学习表示中的层次性和组合性而设计的数据集。该数据集由劳伦斯伯克利国家实验室生物科学与工程部创建，包含了35种韩文字体中的11,172个字符块，总计391,020个标注图像。每个字符块由初始辅音、中间元音和最终辅音的组合构成，具有明确的层次结构和组合规则。HFD数据集不仅用于分析现有深度学习方法的局限性，还旨在推动开发能够从自然变异数据中提取层次和组合结构的新机器学习算法。

The Hangul Fonts Dataset (HFD) is a specialized dataset designed for researching hierarchy and compositionality in deep learning representations. It was created by the Bioscience and Engineering Department of Lawrence Berkeley National Laboratory, encompassing 11,172 character blocks across 35 distinct Korean fonts, totaling 391,020 annotated images. Each character block is composed of a combination of initial consonant, medial vowel, and final consonant, featuring a clear hierarchical structure and combinatorial rules. The HFD dataset not only serves to analyze the limitations of existing deep learning methods, but also aims to promote the development of novel machine learning algorithms that can extract hierarchical and compositional structures from naturally variant data.