HueManity

该数据集名为HueManity，包含了83,850张图片，这些图片以石原测试风格的点阵图案中嵌入双字符字母数字字符串的形式呈现，旨在评估机器学习模型中的视觉感知能力。此外，该数据集还包括了专门针对不同任务的子集：数字识别集（1,000张图片）和文本识别集（1,000张图片），它们各自专注于不同类型的字符。总的来说，这个数据集共有83,850张图片，其任务是评估多语言机器学习模型（MLLMs）的视觉感知能力。

The dataset named HueManity contains 83,850 images, which are formatted as Ishihara-style dot matrix patterns embedded with two-character alphanumeric strings, and it is designed to evaluate the visual perception capabilities of machine learning models. Additionally, the dataset includes task-specific subsets: the digit recognition subset (1,000 images) and the text recognition subset (1,000 images), which focus on different types of characters respectively. Overall, this dataset has a total of 83,850 images, and its core task is to evaluate the visual perception capabilities of multilingual large language models (MLLMs).