tberkane/WB-ChartExtract
收藏官方服务：
资源简介：
WB-ChartExtract 是一个用于图表到表格提取的基准：从图表图像中恢复基础数值表格。它基于世界银行开放数据的时间序列（52个指标，218个国家，65年）构建，设计得比现有基准更具挑战性：图表平均包含比ChartQA多7倍的数据点，不打印值标签（移除了OCR捷径），并涵盖4种图表类型 × 4种渲染库，具有显著风格变化。数据集包含1000个图表，包括图像、真实表格和元数据。
WB-ChartExtract is a benchmark for chart-to-table extraction: recovering the underlying numerical table from a chart image. It is built from World Bank Open Data time series (52 indicators, 218 countries, 65 years) and is designed to be more challenging than existing benchmarks: charts contain on average 7× more datapoints than ChartQA, print no value labels (removing the OCR shortcut), and span 4 chart types × 4 rendering libraries with heavy stylistic variation. The dataset contains 1,000 charts, including images, ground-truth tables, and metadata.
提供机构：
tberkane
