issai/OCRBench-Kazakh

OCRBench-Kazakh是一个手工收集的评估基准，旨在评估大型多模态模型（LMMs）在哈萨克语中的光学字符识别（OCR）和视觉文本理解能力。该数据集用于测试模型在哈萨克语提示下识别、定位和推理各种格式文本的能力，包括标准数字字体、复杂手写脚本和结构化图表。数据集包含三个手工整理的类别：常规文本识别（199个QA对和199张图像）、手写文本识别（100个QA对和100张图像）和图表VQA（142个QA对和71张图像）。它作为一个专门的基准，用于评估模型在真实哈萨克语境中的语言精确性和感知准确性，填补了本土多模态评估资源的关键空白。

OCRBench-Kazakh is a manually collected evaluation benchmark designed to assess the Optical Character Recognition (OCR) and visual-text understanding capabilities of Large Multimodal Models (LMMs) specifically for the Kazakh language. This dataset is used to test how well models can recognize, localize, and reason about text in various formats—ranging from standard digital fonts to complex handwritten scripts and structured charts—when prompted in Kazakh. The dataset consists of the following manually curated categories: Regular Text Recognition (199 QA Pairs and 199 Images), Handwritten Text Recognition (100 QA Pairs and 100 Images), and Charts VQA (142 QA Pairs and 71 Images). It serves as a specialized benchmark for evaluating a models linguistic precision and perceptual accuracy in a real-world Kazakh context, filling a critical gap in native multimodal evaluation resources.