WorldCuisines

该数据集是一个大规模的多语言和多文化视觉基础语言理解基准，涵盖了视觉问答（VQA）数据集，其中包含跨越30种语言和方言的文本-图像对。该基准还包括识别菜肴名称及其起源的任务，并提供了一个带有注释的食物条目和图片的知识库。数据集规模超过100万数据点，评估数据集分别有12,000和60,000个实例，以及一个包含100万个实例的训练数据集。其主要任务是视觉问答（Vqa）。

This dataset is a large-scale multilingual and multicultural visual grounding language understanding benchmark. It encompasses visual question answering (VQA) datasets containing text-image pairs spanning 30 languages and dialects. Additionally, this benchmark includes tasks for identifying dish names and their origins, and provides a knowledge base of annotated food entries and images. The entire dataset contains over 1 million data points, with the evaluation datasets holding 12,000 and 60,000 instances respectively, and a training dataset consisting of 1 million instances. The primary task of this benchmark is visual question answering (VQA).