swap-uniba/MMMEB-Benchmark

MMMEB是一个多语言和多模态嵌入模型的基准数据集，支持英语、法语、德语、意大利语和西班牙语五种语言。它分为四个任务元类别：图像到文本检索（I2T）、文本到图像检索（T2I）、视觉问答（VQA）、视觉定位（VG）和分类（C）。所有考虑的数据集要么是由人手写，要么是经过错误检查的。

MMMEB (Massive Multimodal and Multilingual Embedding Benchmark) is a benchmark for multilingual and multimodal embedding models, supporting five languages: English, French, German, Italian, and Spanish. It is structured into four task meta-categories: Image-to-Text Retrieval (I2T), Text-to-Image Retrieval (T2I), Visual Question Answering (VQA), Visual Grounding (VG), and Classification (C). All datasets considered in this benchmark have been either handwritten by humans or checked for errors.