AlexandreSheva/rukopys-curated-mvp
收藏资源简介:
RUKOPYS Curated MVP是一个乌克兰手写识别数据集,是原始RUKOPYS数据集的任务就绪衍生版本,专门用于乌克兰手写文档AI处理。该数据集将原始RUKOPYS版本转换为可重复的训练工件,支持全页视觉语言微调、裁剪级手写文本转录和布局检测。数据集内容包含元数据文件、区域注释、视觉语言模型微调示例、YOLO格式布局检测数据等,适用于端到端手写文本识别(HTR)流程,包括模型训练、推理和评估。数据集涵盖多种来源,如档案、学校作业和大学考试,并包含不同类型区域(如手写文本、公式、表格等),适用于乌克兰语手写文档的解析、布局检测和OCR/HTR基准测试。
RUKOPYS Curated MVP is a Ukrainian handwriting recognition dataset, a task-ready curated derivative of the original RUKOPYS dataset for Ukrainian handwritten document AI. It transforms the raw RUKOPYS release into reproducible training artifacts for full-page vision-language fine-tuning, crop-level handwriting transcription, and layout detection. The dataset includes files such as metadata.jsonl, regions.jsonl, VLM and page SFT examples, and YOLO-format layout detection data, supporting an end-to-end HTR pipeline for model training, inference, and evaluation. It covers various sources like archives, school assignments, and university exams, with region types including handwritten text, formulas, and tables, and is recommended for Ukrainian handwritten page parsing, layout detection, and OCR/HTR benchmarking.




