WueNLP/Synthdog-Multilingual-100

Synthdog Multilingual是一个多语言数据集，旨在用于训练大型视觉语言模型的OCR（光学字符识别）能力。该数据集包含了超过100万张训练样本图片，用于提高模型对多种语言的识别能力。这些样本中的文本来自于不同语言的维基百科，并且采用了GoNotoKurrent-Regular字体。数据集的大小在1M到10M之间，针对英语有500,000个样本，针对非拉丁脚本有10,000个样本，其他语言则各有5,000个样本。

The Synthdog Multilingual dataset is designed for training the OCR (Optical Character Recognition) capabilities of large vision-language models. The dataset contains over 1 million training sample images aimed at improving the models ability to recognize various languages. The text in these samples is sourced from Wikipedia in different languages and is presented using the GoNotoKurrent-Regular font. The dataset size ranges from 1M to 10M, with 500,000 samples for English, 10,000 for non-Latin scripts, and 5,000 for each of the other languages.