FooDI-ML (Food Drinks and groceries Images Multi Lingual)

食品饮料和杂货图像多语言 (FooDI-ML) 是一个数据集，其中包含从 Glovo 应用程序收集的超过 150 万个独特图像和超过 950 万个商店名称、产品名称描述和集合部分。提供的数据对应于来自欧洲、中东、非洲和拉丁美洲 37 个国家的食品、饮料和杂货产品。该数据集包含 33 种语言，包括来自东欧和西亚国家（如乌克兰语和哈萨克语）的 870K 语言样本，迄今为止，这些语言在公开可用的视觉语言数据集中的代表性不足。该数据集还包括广泛使用的语言，例如西班牙语和英语。描述来自：FooDI-ML：食品、饮料和杂货图像和描述的大型多语言数据集

FooDI-ML (Food, Beverage and Grocery Multilingual) is a dataset consisting of over 1.5 million unique images collected from the Glovo application, along with more than 9.5 million records of store names, product name descriptions and product collection sections. The data provided corresponds to food, beverage and grocery products from 37 countries spanning Europe, the Middle East, Africa and Latin America. This dataset encompasses 33 languages, including 870,000 language samples from Eastern European and West Asian nations such as Ukrainian and Kazakh — languages that have been notably underrepresented in publicly available vision-language datasets to date. It also features widely utilized languages including Spanish and English. This description is sourced from: FooDI-ML: A Large Multilingual Dataset of Food, Beverage and Grocery Images and Descriptions.

搜集汇总

背景与挑战

背景概述

FooDI-ML是一个多语言食品、饮料和杂货图像数据集，包含超过150万张图像和950万条文本描述，覆盖37个国家和33种语言，尤其纳入了乌克兰语、哈萨克语等代表性不足的语言。该数据集由Glovo在2021年发布，旨在支持视觉语言任务。