---
license: mit
---
The dataset consists of synthetic OCR training content featuring a font distribution ratio of approximately 60% serif, 32% sans-serif, and 8% monospace typefaces. The text fragments are randomly selected, and the entire paragraph has no actual meaning to prevent overfitting during training. Additionally, the images have been incorporated with slight rotations, noise, and brightness variations. It is designed to train optical character recognition models across diverse typographical styles, balancing common serif/sans-serif combinations with a smaller proportion of fixed-width fonts for specialized applications.