issai/BeyneleBench

Beynele-Bench 是一个包含750个示例的基于参考的基准测试，用于评估文本到图像生成中的哈萨克文化保真度。每个示例配对一个英语提示、一个哈萨克语提示、一张参考图像和一个三级文化分类体系。该基准测试旨在解决通用纯文本评判可能不足以指定文化特定元素（如人物、物体、建筑、食物、活动、自然场景或哈萨克文本渲染）的情况。

Beynele-Bench is a 750-example reference-based benchmark for evaluating Kazakh cultural fidelity in text-to-image generation. Each example pairs an English prompt, a Kazakh prompt, a reference image, and a three-level cultural taxonomy. The benchmark is designed for cases where generic text-only judging may under-specify culture-specific people, objects, buildings, food, activities, natural scenes, or Kazakh text rendering.