teds指标测试集

数据集来源:百度TableGeneration产生20张有线，20张无线表格，来源WTW 15张表格，来源Pubnet 20张表格,自己标注产生 18张图片，覆盖有线无线，拍照，商品包装，excel，房产证件，景深，轻度偏转，模糊，手写等多维度场景，用于测试表格模型的TEDS指标，自己标注图片产生html来源自制小工具https://github.com/Joker1212/EasyTedsLabeler,欢迎尝试我们的开源表格识别方案https://github.com/RapidAI/TableStructureRec

Dataset Sources: 20 wired and 20 wireless tables generated by Baidu TableGeneration, 15 tables sourced from WTW, 20 tables sourced from Pubnet, and 18 manually annotated images. This dataset covers diverse scenarios including wired/wireless tables, photographed images, product packaging, Excel spreadsheets, real estate certificates, varying depth of field, slight deflection, blurriness, handwritten content, etc., and is designed for testing the TEDS metric of table models. The HTML files generated during the annotation process were created using our homemade tool available at https://github.com/Joker1212/EasyTedsLabeler. You are welcome to try our open-source table recognition solution at https://github.com/RapidAI/TableStructureRec

搜集汇总

背景与挑战

背景概述

该数据集用于评测表格TEDS指标，包含93张表格图片，涵盖多种场景和图像分辨率。数据格式为.jpg，可自由使用。