DIALab/MMCricBench
收藏Hugging Face2025-08-27 更新2025-11-01 收录
下载链接：
https://hf-mirror.com/datasets/DIALab/MMCricBench
下载链接
链接失效反馈官方服务：
资源简介：
MMCricBench是一个用于评估大型视觉语言模型在数值推理、跨语言理解和多图像推理方面的基准数据集。该数据集包含英语和印地语板球计分卡图像，所有问题/答案均为英语。数据集包括1463个合成的计分卡图像（PNG格式），分为822个单图像计分卡和641个多图像计分卡。共有1500个问题-答案对，分为三种推理类别：C1-直接检索与简单推理，C2-基本算术与条件逻辑，C3-多步骤定量推理（通常跨图像）。
MMCricBench is a benchmark dataset for evaluating Large Vision-Language Models (LVLMs) on numerical reasoning, cross-lingual understanding, and multi-image reasoning over semi-structured cricket scorecard images. It includes English and Hindi scorecards; all questions/answers are in English. The dataset consists of 1,463 synthetic scorecards (PNG) divided into 822 single-image scorecards and 641 multi-image scorecards. There are 1,500 QA pairs categorized into three reasoning categories: C1 - Direct retrieval & simple inference, C2 - Basic arithmetic & conditional logic, C3 - Multi-step quantitative reasoning (often across images).
提供机构：
DIALab
