anshulsc/MirageTVQA

MirageTVQA是一个大规模的多语言视觉表格问答基准数据集，旨在评估视觉语言模型(VLMs)在真实表格理解任务上的表现。它包含45种语言、108,772个问答对和40,730个独特的表格图像。数据集包括干净和嘈杂的表格图像，以模拟真实世界的场景。问题涵盖了10种推理类型，例如数值聚合和多跳推理。数据集以JSONL和图像文件的形式组织，可以使用Python进行加载和评估。数据集还提供了针对各种模型的语言特定评估和基准结果。

MirageTVQA is the first large-scale multilingual visual table question-answering benchmark designed to evaluate Vision-Language Models (VLMs) on realistic table understanding tasks. It includes 45 languages, 108,772 question-answer pairs, and 40,730 unique table images, with both clean and noisy variants. The questions cover 10 reasoning types, such as numerical aggregation and multi-hop reasoning. The dataset is structured with JSONL and image files, and can be loaded and evaluated using Python. The dataset also provides language-specific evaluation and benchmark results for various models.