Thousand Character Classification

Characters of the Thousand Character Classic (千字文) in 50 different fonts. Individual characters are saved as 28 x 28 NumPy arrays depicting the characters as white-on-black images, roughly matching the MNIST dataset, though binary (0 or 1) rather than grayscale (0-255). The characters are rendered in simplified text (简体字), which brings the number of unique characters down from 1000 to 994. Also, some fonts do not render all characters, resulting in fewer than 994 characters for some fonts.

Under "Characters", the rendered characters and labels are saved under each font. Characters are saved as N x 28 x 28 arrays, where N is the number of characters rendered in that font. Labels are saved as a csv listing all characters rendered in that font.

Under "Fonts", the fonts used to render the characters are saved as either .ttf or .otf files.

"Thousand_Character_Classic.txt" is the simplified character text of the Thousand Character Classic.

All fonts were downloaded from chinesefonts.org, with the exceptions of Liu Jian Mao Cao, Long Cang, Noto Sans SC, Noto Serif SC, ZCOOL KuaiLe, ZCOOL XiaoWei, ZCOOL QingKe HuangYou, and Zhi Mang Xing, which were downloaded from fonts.google.com.

Code to reproduce the character arrays can be viewed and run in Google Colab. The notebook can also be copied in order to save or edit it.