Teklia/ATR-benchmark
收藏Hugging Face2025-11-26 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/Teklia/ATR-benchmark
下载链接
链接失效反馈官方服务：
资源简介：
ATR基准数据集是一个多语言数据集，包含83个文档图像，涵盖页面或段落级别。该数据集旨在测试ATR模型，并整合了多个公开数据集的数据，包括BnL历史报纸、CASIA-HWDB2、DIY历史-社会正义、DAI-CRETDHI、FINLAM历史报纸、Horae-时辰书、IAM、NorHand v3、Marius PELLET、RASM、READ-2016、RIMES和ScribbleLens等。图像保持原始大小。数据集分为测试集，包含133个样本。数据集的字段包括图像、文本、数据集名称、arkindex_id、语言和级别。
The ATR benchmark dataset is a multilingual dataset containing 83 document images at the page or paragraph level. It is designed to test ATR models and combines data from several public datasets, including BnL Historical Newspapers, CASIA-HWDB2, DIY History - Social Justice, DAI-CRETDHI, FINLAM - Historical Newspapers, Horae - Books of hours, IAM, NorHand v3, Marius PELLET, RASM, READ-2016, RIMES, and ScribbleLens. Images are in their original size. The dataset is split into a test set with 133 samples. The fields of the dataset include image, text, dataset name, arkindex_id, language, and level.
提供机构：
Teklia
