NIST SD 19

该数据集包含了26个大写英文字母的图像，这些图像由99种精选字体生成，旨在模仿《科学鸟》游戏中的结构。由于公开可用的适合评估竞赛性能的数据集匮乏，因此创建了这一数据集。它被用于对视觉变压器（ViT）模型进行微调。该数据集的规模为2574张生成图像，每个字符有90张训练图像和9张测试图像，总计130张图像用于评估。其任务是针对程序内容生成中的字符识别，对图像分类器进行微调。

This dataset consists of images of 26 uppercase English letters, generated with 99 carefully curated fonts, and is constructed to mimic the in-game structure of the *Science Bird* game. This dataset was developed to address the scarcity of publicly available datasets tailored for evaluating competitive model performance. It is employed for fine-tuning Vision Transformer (ViT) models. In total, the dataset contains 2574 generated images: each character has 90 training images and 9 test images, with 130 images in total set aside for evaluation. The core objective of this dataset is to fine-tune image classifiers for character recognition tasks within program content generation workflows.