AI-4-Everyone/Visual-TableQA

Visual-TableQA是一个开放域的视觉表格问答基准数据集，包含与LaTeX表格图片相关的合成问题-答案对。每个条目由唯一的表格ID、PNG格式的表格图片、自然语言问题和真实答案组成。数据集适用于表格推理任务的LLM或视觉-语言模型的微调、表格问答流程的基准测试以及对结构化视觉数据的模型鲁棒性评估。Visual-TableQA is an open-domain visual table question answering benchmark dataset that includes synthetic question-answer pairs associated with LaTeX table images. Each entry consists of a unique table ID, a PNG rendering of the LaTeX table, a natural language question, and a ground-truth answer string. The dataset is suitable for fine-tuning LLMs or vision-language models on table reasoning tasks, benchmarking pipelines for table question answering, and evaluating model robustness on structured visual data.